This file lists all changes made to the game driver in all gory detail. See the file HISTORY for a user-oriented summary of all the changes. 02-Feb-2002 (Lars Duening) (3.2.9-dev.534 == 3.2.9) - (many files) Some cosmetic cleanup. 26-Jan-2002 (Lars Duening) (3.2.9-dev.533) - (prolang.y) Corrected a small bug in the code generation for operators &&= and ||=. 26-Jan-2002 (Lars Duening) (3.2.9-dev.532) - (prolang.y) One more detail to the compilation problem (reported by Freaky). 26-Jan-2002 (Lars Duening) (3.2.9-dev.531) - (prolang.y) Removed a compilation problem (reported by Freaky). 26-Jan-2002 (Lars Duening) (3.2.9-dev.530) - (prolang.y) Removed a compilation problem. 25-Jan-2002 (Lars Duening) (3.2.9-dev.529) - (object.c) Corrected a compilation problem (reported by Joseph). - (prolang.y, interpret.c, func_spec) Implemented short-circuit evaluation for the operators '&&=' and '||='. 24-Jan-2002 (Lars Duening) (3.2.9-dev.528) - (lex) type #pragmas set in an auto-include were reset at the end of it (reported by Gnomi). 24-Jan-2002 (Lars Duening) (3.2.9-dev.527) - (object.c) save/restore didn't look out for shared closures. 22-Jan-2002 (Lars Duening) (3.2.9-dev.526) - (etc/lpc.xml) LPC syntax highlighting for KDE's kate (thanks, Menaures!) 19-Jan-2002 (Lars Duening) (3.2.9-dev.525) - (interpret.c) Changed the error thrown by replace_program() to a mere debug message. 19-Jan-2002 (Lars Duening) (3.2.9-dev.524) - (interpret.c) Restricted replace_program() on the first inherited program if that program has virtually inherited variables. See b-030119 for an explanation. 16-Jan-2002 (Lars Duening) (3.2.9-dev.523) - (prolang.y) When redefining a function, the redefinition can now have more arguments than the original. - (lex.c) Escaping newlines in #if conditions on Windows didn't work (reported by Zonk) 14-Jan-2002 (Lars Duening) (3.2.9-dev.522) - (several files) Removed no longer used diagnostic code. 07-Jan-2002 (Lars Duening) (3.2.9-dev.521) - (several files) Removed the 'no-blueprint' configuration option: there code is stable. 05-Jan-2002 (Lars Duening) (3.2.9-dev.520) - (make_func.y) When creating type/hookmaps in prolang.y, the linecounting didn't count the map's continuation lines. - (interpret.c) string lvalue indices can no longer index past the end of the string. - (interpret.c) Removed a couple sanity checks which are no longer needed. 20-Dec-2002 (Lars Duening) (3.2.9-dev.519) - (lex.c) Added a workaround for an optimizer bug in Cygwin 3.2 (reported by Zonk). 20-Dec-2002 (Lars Duening) (3.2.9-dev.518) - (actions.c) Throw an error if H_NOTIFY_FAIL hook is missing. This also prevents the driver from crashing if the H_SEND_NOTIFY hook is defined (reported by Zonk). 18-Dec-2002 (Lars Duening) (3.2.9-dev.517) - (object.c) When calling privilege_violation() for restoring efuns, the current object is passed as 'who' argument (suggested by Gnomi). - (object.c) When saving lfuns, I accidentally deleted the lookup for inherited lfuns (reported by Gnomi). 17-Dec-2002 (Lars Duening) (3.2.9-dev.516) - (object.c) restore_svalue() did not check if a restored efun was overriden by a nomask simul-efun (reported by Gnomi). - (comm.c) With pthreads, the driver could crash in remove_interactive() when the write_cond was destroyed while the thread was still waiting on it (reported by Lee). 16-Dec-2002 (Lars Duening) (3.2.9-dev.515) - (object.c) restore_map_size() stumbled over closure entries (reported by Jason). 16-Dec-2002 (Lars Duening) (3.2.9-dev.514) - (comm.c) If the login object destructed itself in the logon() call, the driver crashed in the print_prompt() routine (reported by Jason). 15-Dec-2002 (Lars Duening) (3.2.9-dev.513) - (object.c) Corrected a compilation error (thanks, Dafire and Jason!). 15-Dec-2002 (Lars Duening) (3.2.9-dev.512) - (prolang.y) It is no longer possible to defined a variable to be of type 'void' (reported by Gnomi). 14-Dec-2002 (Lars Duening) (3.2.9-dev.511) - (object.c) Small corrections. 14-Dec-2002 (Lars Duening) (3.2.9-dev.510) - (object.c) Efuns save_() and restore_() can now handle non-lambda closures, symbols and quoted arrays. This required a new savefile format, though. 11-Dec-2002 (Lars Duening) (3.2.9-dev.509) - (doc/concepts/negotiation, src/wk/) Added some telnet protocol related documentation. - (func_spec, lex.c, prolang.y, interpret.c) New operators '&&=' and '||=' (suggested by Croft). 09-Dec-2002 (Lars Duening) (3.2.9-dev.508) - (mudlib/telnetneg.c) Updated (thanks, Fiona!) - (mudlib/test_master.c) Added testcode for using charmode (thanks, Fiona!) - (comm) Fiona provided more patches: SAVE_NOECHO, which makes sure that the client can switch to CHARMODE before also setting NOECHO; and SIMULATE_CHARMODE which allows the mudlib to switch to charmode even if the client can do only linemode. - (simulate.c) Efuns copy_file() and rename() no longer throw errors on failures, but just return a non-0 value (reported by Menaures). 05-Dec-2002 (Lars Duening) (3.2.9-dev.507) - (config.h, comm, gcollect.c, main.c) Replaced USE_PTHREAD by USE_PTHREADS. - (comm.c) Before printing the prompt, current_object is set to NULL so that the player's catch_tell() is sure to be able to see the prompt (reported by mudder@Furtual.ORG). 04-Dec-2002 (Lars Duening) (3.2.9-dev.506) - (make_func.y, comm.c, main.c, prolang.y) Removed warnings (reported by Coogan). - (comm.c) When logging on a new player, the prompt wasn't printed - in particular the prompt set by input_to(INPUT_PROMPT) (reported by Nostradamus). - (main.c) If the reserved areas can be reallocated, the out_of_memory flag is reset. - (simulate.c) catch() reports an 'out_of_memory' condition only if it was raised during the execution of the guarded code. 03-Dec-2002 (Lars Duening) (3.2.9-dev.505) - (patch-o-matic) Updated to new executable name. - (prolang.y, array.c) The filenames returned by include_list() now always start with a '/' to make their reuse in include statements simpler (suggested by Coogan). - (mudlib/telnet) Fiona donated the telnet negotiations module from Wunderland, containing a complete implementation of the telnet protocol (thanks!). - (sprintf.c) The %O/%Q formatter stumbled over really large arrays and mappings: the top-level function ignored if the conversion routine reallocated the string buffer (reported by Bardioc). 03-Dec-2002 (Lars Duening) (3.2.9-dev.504) - (Makefile.in) Corrected two typos. 03-Dec-2002 (Lars Duening) (3.2.9-dev.503) - (main.c) Forgot to add the configurable efuns to the '--options' output. 02-Dec-2002 (Lars Duening) (3.2.9-dev.502) - (func_spec, interpret.c) Efun member_array() is now deactivated with !USE_DEPRECATED (suggested by Fiona). - (configure, config.h, efuns, simulate, object, interpret.c): The efuns set_light(), set_is_wizard() and process_string() can be configured from the configure script. - (Makefile) Driver executable renamed from 'driver' to 'ldmud'. - (configure, Makefile) The configure script now tries to figure out on its own which warning options the compiler supports. 30-Nov-2002 (Lars Duening) (3.2.9-dev.501) - (comm.c) Accidentally deleted two lines in the noecho handling. 30-Nov-2002 (Lars Duening) (3.2.9-dev.500) - (comm.c) Fiona provided a number of bugfixes for CHARMODE - thanks! 30-Nov-2002 (Lars Duening) (3.2.9-dev.499) - (prolang.y) Brainfart: the last patch created a problem, instead of solving one. The bug was that "*"::fun() looked at _all_ inherited programs, not just the directly inherited ones. 30-Nov-2002 (Lars Duening) (3.2.9-dev.498) - (prolang.y) When making wildcarded supercalls ("*"::fun()), the compiler takes care that each function is called only once (reported by Gnomi). 29-Nov-2002 (Lars Duening) (3.2.9-dev.497) - (sysmalloc.c) This file is no longer needed (actually it has been useless since 3.2.8-dev.188). - (lex.c) pragma_combine_strings was never initialised with a default. - (lex.c) Changed the FLOAT_MIN/FLOAT_MAX predefined to reflect the C-float limits, not the C-double limits, as the gamedriver's floats are somewhere inbetween the two. 27-Nov-2002 (Lars Duening) (3.2.9-dev.496) - (patchlevel.h, hosts/be/Makefile, hosts/be/set-version) The version flags of the BeOS executable are now read from patchlevel.h and not set directly in the BeOS Makefile. 27-Nov-2002 (Lars Duening) (3.2.9-dev.495) - (efuns.c) Efun transfer() was still incorrect. Funny though that nobody noticed it before :-) (reported by Nostradamus). 27-Nov-2002 (Lars Duening) (3.2.9-dev.494) - (comm.c) The correction of the threads racing condition unfortunately introduced a deadlock - Gawain found and fixed it (thanks!). 27-Nov-2002 (Lars Duening) (3.2.9-dev.493) - (efuns.c) Corrected the deprecated efun transfer(): the check for dest->can_put_and_get() was implemented wrong (reported by Nostradamus). 27-Nov-2002 (Lars Duening) (3.2.9-dev.492) - (comm.c) The thread-cleanup function now locks the structure before performing the cleanup (suggested by Gawain). 26-Nov-2002 (Lars Duening) - (object.c) restore_object("#0:0") caught the driver in an endless loop (thanks, Ctx!). 22-Nov-2002 (Lars Duening) - (prolang.y, interpret.c) The __INIT code is no longer restricted to the first 64K of generated bytecode. 21-Nov-2002 (Lars Duening) - (etc/patch-o-matic) Contributed by Guandalug - thanks! 20-Nov-2002 (Lars Duening) (3.2.9-dev.491) - (closure.c, interpret.c) Corrected refcount errors in the handling of soon-to-be dangling alien-lfun closures. - (smalloc.c) Made sure that before each call to fatal() the thread-check is reset. 19-Nov-2002 (Lars Duening) (3.2.9-dev.490) - (closure.c, interpret.c) In order to convert a dangling lfun closure during a replace_program() into a dangling alien-lfun closure, the closure needs to be created as alien-lfun in the first place (reported by Pulami). 17-Nov-2002 (Lars Duening) (3.2.9-dev.489) - (many files) Pure source cosmetics: removed trailing spaces at lineends. 16-Nov-2002 (Lars Duening) (3.2.9-dev.488) - (smalloc.c) If the allocator ran out of memory, the exit handling triggered the 'Multiple threads' safety check. Not really a problem, as the driver would exit anyway, but it obscures what is happening. 15-Nov-2002 (Lars Duening) (3.2.9-dev.487) - (comm.c) The call to remove_flush_entry() could clobber the list of players to flush if it was called for a player not in the flush list. 13-Nov-2002 (Lars Duening) (3.2.9-dev.486) - (interpret.c) get_type_info(,2) on an unbound-lambda crashed the driver because an uninitialized variable was used (thanks, Ctx!) - (simulate.c) If the H_NOECHO hook is set, the telnet negotiations are redirected only after the hook setting is verified (thanks, Fiona!) - (comm.c) Corrected a potential problem when an interactive object with pending data is forcefully made non-interactive: from my reading of the code it seems that the object remained in the list of players-to-flush, even though it no longer had a valid interactive_t attribute. 12-Nov-2002 (Lars Duening) (3.2.9-dev.485) - (port.h) The 'malloc' attribute is now used only with gcc 3 or newer. - (lex.c) Corrected the line counting with auto-includes (reported by Coogan). 11-Nov-2002 (Lars Duening) (3.2.9-dev.484) - (comm.c) When calling H_NOECHO, the code did not check if the argument objects were still alive (reported by Fiona). 08-Nov-2002 (Lars Duening) - (port.h, comm.c, xalloc.h, smalloc.h) Activated GCC attributes for OS-X compilations, added GCC attribute 'malloc' to the allocator functions, removed some warnings. 07-Nov-2002 (Lars Duening) (3.2.9-dev.483) - (configure.in) The tests regarding which libraries are needed suffered from a disabling typo (reported by Nostradamus). - (port.h, hosts/unix.h) Adapted to the latest Cygwin release. 06-Nov-2002 (Lars Duening) (3.2.9-dev.482) - (closure.c) If a lfun-closure is left danging by a replace_program(), it is now converted into an alien-lfun closure pointing to master::dangling_lfun_closure(). This way the error handler has a chance to determine the originating object (suggested by Gnomi). - (closure.c) The handling of dangling alien-lfun closures actually destroyed the alien.ob entry. - (interpret.c) allocate() used with an array or mapping initialisation value now creates shallow copies of the initialisation value for each initialised element (suggested by Oskar). 04-Nov-2002 (Lars Duening) (3.2.9-dev.481) - (smalloc.c) The thread-check in xfree() also triggered when the memory reserves were released (reported by Bardioc). 03-Nov-2002 (Lars Duening) - (configure.in) Made the check for -lpthread more robust (reported by Markus Peter). 03-Nov-2002 (Lars Duening) - (configure.in) Made the check for -lpthread more robust (reported by Markus Peter). 01-Nov-2002 (Lars Duening) (3.2.9-dev.480) - (comm.c) *blush* Deleted TELOPT_EOR handling by act of stupidity.. 31-Oct-2002 (Lars Duening) (3.2.9-dev.478, -dev.479) - (comm.c) Added TELOPT_BINARY to the telnet negotiations forwarded to H_TELNET_NEG (suggested by Fiona). 30-Oct-2002 (Lars Duening) (3.2.9-dev.477) - (lex.c) When starting a new compile, the order of initialisations was wrong (in particular: auto-include was handled before the lexer was set up) (reported by Freaky). - (configure, (x)erq/Makefile.in) Only link in libraries actually needed by the erq (suggested by Freaky). - (configure) Modified the configure so that only required libraries are added to the compile (suggested by Freaky). 29-Oct-2002 (Lars Duening) (3.2.9-dev.476) - (comm, func_spec) find_input_to() and remove_input_to() can now match on both object and string name. - (interpret.c) unquote(symbol) added a quote level instead of removing it (reported by Gnomi). - (comm.c) Improved the 'already bound' error messages to list the port number. - (util/xerq) If the xerq couldn't send all data from a ERQ_SEND request immediately and thus returned a ERQ_E_INCOMPLETE, it will follow it up with a ERQ_OK message to the same handle later once all the pending data has been send (provided by Fiona, thanks!) - (util/xerq/lpc) Replaced the socketd with the Wunderland daemon (thanks!). - (lex.c, prolang, simulate.c) auto-includes are now handled like normal includes (instead of just placing the text into the textbuffer and fudging the current_line variable), correcting a source of false linenumber informations. 27-Oct-2002 (Lars Duening) (3.2.9-dev.474, .475) - (smalloc.c) Added a very primitive safeguard against interleaving calls to the allocator - I suspect that this is the root cause for the problems with pthreads. - (object.c) When resetting an object, the previous_object() was some random object (reported by Gnomi). - (comm, func_spec) New efuns find_input_to() and input_to_info() (suggested by Fiona). 16-Oct-2002 (Lars Duening) (3.2.9-dev.473) - (configure) Some systems need libresolv in order to provide herror() (reported by Coogan). - (comm.c) Removed a compile warning (thanks, Fiona!). - (array.c) regexplode: Removed a memory leak on stack overflow (thanks, Fiona!). 15-Oct-2002 (Lars Duening) (3.2.9-dev.472) - (exec.h, lex.c, prolang, object, gcollect.c, typedefs.h) Added complete include file information to compiled programs (backported from 3.3). - (object.c) Implemented efun include_list() properly (backported from 3.3). - (prolang.y) Adapted to bison 1.5 . 14-Oct-2002 (Lars Duening) - (configure) If a requested option (like use-mysql) can't be used because the system lacks the necessary support, the configure script issues a warning and exits with return code 1 (suggested by Freaky). 13-Oct-2002 (Lars Duening) (3.2.9-dev.471) - (prolang.y) Removed a compiler warning. - (configure.in) The test for pthread support was broken, and the test for 'long long' was run twice. And if pthreads are not found, USE_PTHREADS is now deactivated (reported by Freaky). 10-Oct-2002 (Lars Duening) (3.2.9-dev.470) - (efuns.c, svalue.h, interpret.c) Evaluation cost of deep_copy() is now proportional to size and nesting depth of copied arrays and mappings, if DYNAMIC_COST is defined (suggested by Gnomi). 09-Oct-2002 (Lars Duening) (3.2.9-dev.469) - (port.h, swap.c) No longer includes <values.h> which has been obsoleted by <limits.h> and <float.h>. - (config.h, driver.h) Removed the compilation warning when YYDEBUG was not defined. - (mud/lp-245) Corrected the transfer() simul-efun (thanks, Ralph!) 09-Oct-2002 (Lars Duening) (3.2.9-dev.467, .468) - (lex.c, prolang) Implemented #'inh::fun lfun closures (b-010801-0). 07-Oct-2002 (Lars Duening) (3.2.9-dev.466) - (comm.c) Removed compilation problems with BeOS. - (prolang.y) Oh My Ghod. The range index [..<expr0>] created faulty bytecode, maybe even clobbered random memory, if the expression was anything complicated, like 'm++' or 'm = 4'. This bug was caused by uninitialized the LPC parser not initializing the internal return value from parsing <expr0>, so the bug didn't manifest itself every time (reported by Zonk). 04-Oct-2002 (Lars Duening) (3.2.9-dev.465) - (autoconf/configure.in) Corrected the problem with newer autoconfs. 01-Oct-2002 (Lars Duening) (3.2.9-dev.464) - (comm.c) Forgot a parameter in an error message (thanks, Freaky!) - (main.c, comm) The hostname was initialized to late, so the debug.log file was opened with the wrong name (reported by Dafire). 30-Sep-2002 (Lars Duening) (3.2.9-dev.463) - (comm.c) Changed the fprintf() in the thread cleanup routine into a malloc-friendlier dprintf1(). - (main.c) Added a check into debug_message() if the name of the debug log file has already been set (reported by Dafire). - (simulate.c) The driver now checks if an object to be loaded appears out of nowhere while it is being compiled (this can happen if the mudlib warning handler is careless with rename_object()) and throws an error (reported by Gnomi). - (main.c, comm) New commandline options --hostname and --hostaddr to override the values returned by the system. In particular, --hostaddr will be used as the address to open to the driver ports. 28-Sep-2002 (Lars Duening) - (config.h.in) USE_PTHREAD was permanently commented out (reported by Gawain). 27-Sep-2002 (Lars Duening) (3.2.9-dev.461+.462) - (comm, main.c, configure) Based on a patch by Gawain, PTHREADS and PCRE (requires Fiona's patch) can now be configured like other options. New commandline option --max-thread-pending to set the limit for pending data to write. 25-Sep-2002 (Lars Duening) - (comm.c) If a connection is marked for removal (.do_close != 0), telnet commands can still be sent. This is to allow the input_to() handling to properly reset the connection to linemode even if the input_to() handler has called remove_interactive(). - (backend, interpret.c, smalloc.c) Made the GC request log output nicer. 25-Sep-2002 (Lars Duening) (3.2.9-dev.460) - (comm, gcollect.c) Removed compiler warnings, and a crash in pthreads thread cleanup (reported by Dafire) 25-Sep-2002 (Lars Duening) (3.2.9-dev.459) - (interpret.c) The LPC tracing code crashed when used with lambda closures or with call_resolved() (reported by Nostradamus). 24-Sep-2002 (Lars Duening) (3.2.9-dev.458) - (comm, backend.c, gcollect.c) smalloc is not threadsafe, which caused the implementation of USE_PTHREAD code to trash the memory structures. To solve the problem, all memory allocations and deallocations are done by the main thread. Patches from Guandalug: - (interpret.c) Converting a sefun-closure to a string looked up the name in the efun table instead of the sefun table. 20-Sep-2002 (Lars Duening) (3.2.9-dev.457) - (prolang.y) The test for duplicate toplevel inherits got confused when virtual inherits occured (reported by Zonk via Dafire). 19-Sep-2002 (Lars Duening) - (etc/lpc-mode.el) LPC mode for EMACS, by Vivek Dasmohapatra. 17-Sep-2002 (Lars Duening) - (comm.c) ECONNREFUSED is recognized as error (suggested by Coogan). 16-Sep-2002 (Lars Duening) - (comm.c) With PThreads: forgot to detach new threads after creation (thanks, Gawain!) - (configure) The object statistic seems to be correct now, so the OSTAT diagnostics are disabled by default. 15-Sep-2002 (Lars Duening) (3.2.9-dev.456) - (simulate) Both 'warn' and 'warning' are already symbols on some systems, so I now used 'warnf' as name of the warning function. - (regexp, rxcache, array.c, backend.c, ed.c) Removed name clashes with system libraries. 15-Sep-2002 (Lars Duening) (3.2.9-dev.455) - (prolang.y, lex, interpret.c, simulate, func_spec) New pragma 'warn_deprecated' causes the compiler to print warnings when a deprecated efun is used (suggested by Dafire). 03-Sep-2002 (Lars Duening) (3.2.9-dev.454) - (swap.c) The object statistics get messed up when the GC swaps objects. Let's see if this change corrects all of the discrepancies. In general the reason for the discrepancies was that variables were counted more often as 'swapped in' than they had been swapped out. 30-Aug-2002 (Lars Duening) (3.2.9-dev.453) - (efuns.c, lex.c, make_func.y) Some chars were casted to 'unsigned int' instead of 'unsigned char', breaking support for international character sets (thanks, Oskar!). 26-Aug-2002 (Lars Duening) - (swap.c) The OSTAT diagnostic for swap-in subtracted instead of added to compute the expected new tot_alloc_object_size value (reported by Coogan). 21-Aug-2002 (Lars Duening) - (object.c) Oops, omitted one argument in one of the OSTAT outputs. 21-Aug-2002 (Lars Duening) (3.2.9-dev.452) - (comm.c, object.c, simulate.c, swap.c) Corrected some warnings (reported by Dafire). - (interpret, and some more files) Master applies caused by external events (like ERQ messages) are now distinguished from internal calls (like valid_write()) so that the error handling for external calls receive the full eval cost benefits (suggested by Fiona). - (Makefile.in s): Added missing @EXEEXT@ tags. 20-Aug-2002 (Lars Duening) (3.2.9-dev.451) - (main.c) Needed to #include <netdb.h> on Solaris to get the MAXHOSTNAMELEN macro (thanks, Coogan). - (util/erq/erq.c, util/xerq/erq.c) Forgot the timestamp in the initial startup message. - (port, simulate.c, comm.c, hosts/unix.h) Cygwin corrections. - (config.h, configure, driver.h, erqs) New configuration option with-erq-debug sets the new config.h macro ERQ_DEBUG which defines the ERQ debug level. No more '-DDEBUG' to enable the debugging code in the ERQs (suggested by Bill). 19-Aug-2002 (Lars Duening) (3.2.9-dev.450) - (closure.c) Efun symbol_efun() freed symbol-svalues as strings (reported by Dafire). 19-Aug-2002 (Lars Duening) (3.2.9-dev.449) - (lex.c) An error in a permanent define given on the command line would crash the driver because the lexer called error() before the VM had been set up. - (Makefile.in) Some makes don't like ':=' variable assignments (reported by Coogan). - (main.c) If the driver uses the default debug.log file, its (mudlib) name is available in the predefined macro __DEBUG_LOG__ (suggested by Mandragon). - (driver.h, config.h) Made the config.h macro DEBUG invisible to the ERQs as their debugging mode is controlled by their Makefiles. - (util/erq/erq.c, util/xerq/erq.c) The ERQs now print their name and compilation date to stderr on startup. 18-Aug-2002 (Lars Duening) (3.2.9-dev.448) - (lex.c) Added predefined macros __RESET_TIME__ and __CLEANUP_TIME__ (suggested by Elric). 14-Aug-2002 (Lars Duening) - (configure) Optimization option -funroll-all-loops seems to create buggy code, so I removed it from the configure. 14-Aug-2002 (Lars Duening) (3.2.9-dev.447) - (comm.c) Compilation broke under BeOS. - (object.c, swap.c, simulate.c) Removed warnings. - (Makefile.in) Removed a GNUMake'ism (reported by Coogan). 13-Aug-2002 (Lars Duening) (3.2.9-dev.446) - (configure) The VERBOSE_OPCPROF define in config.h got clobbered. - (interpret.c) I removed the copy-on-write for mappings again - it broke more than it fixed. 12-Aug-2002 (Lars Duening) (3.2.9-dev.445) - (lex.c) New predefined macros __CATCH_EVAL_COST__ and __MASTER_EVAL_COST__ to carry the corresponding configuration options over to LPC (suggested by Menaures). - (Makefiles, configure, config.h) Moved the debugging options into config.h to ensure clean recompilations, and made them configurable through the normal configure mechanism. 11-Aug-2002 (Lars Duening) - (driver.h, object.c, simulate.c, swap.c) Defining debug macro CHECK_OBJECT_STAT allows tracing of all changes to the tot_alloc_object(_size) statistics. Aim is to find the discrepancy in the 'status' command. To activate the tracing, the commandline option '--check-object-stat' must be given. - (interpret.c) The operations <map> += <map> and <map> -= <map> now implement the copy-on-write semantic to mirror the behaviour of the related array operations (suggested by Dafire). - (interpret.c) get_type_info(<closure>,2) did not return the bound object for efun, simul-efun and operator closures (reported by Gnomi). - (interpret.c) Attempting to replace a program in an object which already has closures bound no longer raises an error, but issues a debug_message() (suggested by Pulami). - (make_func.y) Removed two warnings by bison (reported by Gawain). 10-Aug-2002 (Lars Duening) (3.2.9-dev.444) - (closure.c) In efun symbol_function(), symbols were freed as strings (reported by Freaky). - (main, backend.c) Signal USR2 forces the driver to reopen the host.debug.log file (thanks, Gawain!). - (comm, gcollect.c) Experimental: Defining USE_PTHREAD (if your system has pthread support) activates code which uses threads to write the socket data to the network in the background. Intention is to avoid EWOULDBLOCK errors on slow connections (thanks, Gawain!) 08-Aug-2002 (Lars Duening) - (comm.c) When using IPv6, get_host_ip_number() used a too small buffer (thanks, Fiona!) - (backend.c) Improved the 'doing a GC' messages somewhat. 05-Aug-2002 (Lars Duening) (3.2.9-dev.443) - (mapping.c) Embarassing: the bugfix broke map() and filter() for all non-multi-value mappings (reported by Casey). 04-Aug-2002 (Lars Duening) (3.2.9-dev.442) - (prolang.y) Inheriting the same program twice at the top level now raises a warning, resp. an error in pedantic mode. - (simulate.c) If a closure is called from limited() and self-destructs, the driver would try to free the now-invalid closure argument and crash (this happened a lot in EotL) (thanks, Casey!). - (interpret.c) In DEBUG mode, the functions free_object_svalue() and free_string_svalue() check that they really get the proper data type. - (efuns.c) Efuns filter() and map(): when used with multi-value mappings, the driver checks if the called function kept a reference to the passed array with the values. If yes, a new array is created for the next iteration to avoid overwriting the old one (reported by Fiona). - (interpret.c) When collecting the traceback in a catch(), the file and line information was used, but not initialized, occasionally crashing the driver. In addition, in a couple of cases when the file and line information can't be determined, the file was left undefined (thanks, Gnomi!) 15-Jul-2002 (Lars Duening) (3.2.9-dev.441) - (mud/lp-245) The wizlist() simul-efun didn't print proper failure notices (reported by Vandergnome). - (gcollect.c) In a GC the list of destructed objects was handled last, by which time the list might already have been clobbered. Moving the handling first should plug a memory leak (reported by Dafire). - (prolang.y) The crasher correction from dev.439/dev.440 still had a small bug which caused variables and functions to be assigned wrongly (reported by Fiona and Gnomi, fixed by Menaures). - (prolang.y) If a variable was inherited multiple times, the driver generated a senseless warning (reported by Zora). - (efuns.c, interpret.c, main.c) Added some support for locales; in particular the driver now uses the locale defined in the LANG environment variable for character conversions (thanks, Oskar!). - (simulate.c) Efun callbacks misparsed the arguments when only the function name and extra arguments, but no object was provided. 14-Jul-2002 (Lars Duening) - (etc/memusage) Fiona updated the script to make it more robust and work with the new OBJ_DUMP format. Furthermore the constants for $DOMAINS and so on do work now. 05-Jun-2002 (Lars Duening) (3.2.9-dev.440) - (prolang.y) Oops, yesterday's bugfix was not quite corrected. 04-Jun-2002 (Lars Duening) (3.2.9-dev.439) - (prolang.y) If a method in a virtual inherit is rendered invisible, but also made available again through a parallel inherit of the same program, the compiler resolved the visible function to the wrong inheritance, causing the driver to crash (b-020530). 13-May-2002 (Lars Duening) (3.2.9-dev.438) - (config.h.in, Makefile.in, autoconf/configure.in) TRACE_CODE is now configurable through the configure script, and thus is located in config.h and no longer in Makefile. - (gcollect, interpret.c, func_spec) Optional filename argument to efun garbage_collection() can specify a new GC log file for the next GC. The filename is validated through master::valid_write() (f-020312). 07-May-2002 (Lars Duening) (3.2.9-dev.437) - (mapping.c) Improved the checks for empty mappings in add_mapping() and add_to_mapping(). - (mapping.c) 'mapping += mapping' now throws an error when the mappings are of differing width (reported by Alfe and Coogan). - (prolang.y) New notation '([:<expr>])' as shortcut for m_allocate(0, <expr>) (suggested by Alfe). 06-May-2002 (Lars Duening) (3.2.9-dev.436) - (smalloc.c) With CHECK_OBJECT_GC_REF, small chunks could lead to misleading diagnostics during the GC. 03-May-2002 (Lars Duening) (3.2.9-dev.435) - (prolang.y) Reworked the program block reallocations to make sure that their bounds won't be overwritten. This could have been the reasons for the crashers on corrupted memory (reported by Coogan). 01-May-2002 (Lars Duening) - (settings/dt2) Updated the DeeperTrouble settings file; and also removed the mud/dt2 directory since it's no longer necessary (thanks, Lasse!). 26-Apr-2002 (Lars Duening) (3.2.9-dev.434) - (prolang.y) Hiding an inherited private nosave variable with an inherited non-private variable didn't work (reported by Gnomi). - (prolang.y, lex.c, simulate.c) Compiler error and warning messages all have the format '<file> line <line>: <text>' now. - (prolang.y, swap.c) Oops, the linenumber structure was allocated too small on creation, and read in incorrectly from swap. - (driver.h) Deactivated the debug defines CHECK_MAPPING_TOTAL and CHECK_OBJECT_REF since these bugs seem to be fixed. 25-Apr-2002 (Lars Duening) (3.2.9-dev.433) - (comm.c) Using unbound lambdas for prompts crashed the driver (reported by Gnomi). - (driver.h, gcollect, smalloc) When macro CHECK_OBJECT_GC_REF is activated, every supposedly non-object memory reference in a GC is checked if it points to an object. If yes, a diagnostic is printed to gcollect out. Requires MALLOC_TRACE to work. 21-Apr-2002 (Lars Duening) (3.2.9-dev.432) - (exec.h, prolang.y, interpret.c, object.c, swap.c) Linenumbers are now kept in a separate memory block, which makes swapping them out and back in simpler. 20-Apr-2002 (Lars Duening) (3.2.9-dev.431) - (prolang.y, lex.c) Calls to the master during a compilation now check the available evaluation costs. If the limit is exceeded, a compilation error is generated. - (swap.c) Removing a program block from the swap no longer crashes if the program doesn't have line numbers (reported by Pixie). - (prolang.y) Calls to undefined functions now generate a warning when detected in functions without typechecking if #pragma pedantic is in effect (before the compiler silently inserted a prototype). 19-Apr-2002 (Lars Duening) - (prolang.y) Removed a few bison warnings (reported by Menaures). 18-Apr-2002 (Lars Duening) (3.2.9-dev.430) - (prolang.y) When passing a wrong argument to a varargs function, the error message gave the wrong argument number (reported by Gnomi). - (func_spec, interpret.c) New efun m_add(). 09-Apr-2002 (Lars Duening) (3.2.9-dev.429) - (prolang.y) The improved type checking on <op>= accidentally broke the type checking on the plain assignment (reported by Dafire). 08-Apr-2002 (Lars Duening) (3.2.9-dev.428) - (prolang.y) Improved the type checking on the <op>= operators to match what the VM actually implements (reported by Menaures). 04-Apr-2002 (Lars Duening) (3.2.9-dev.427) - (mapping.c) During a GC, the .hash member in a mapping does not point to a mapping-hash structure - the statistics code ignored that (b-011015-0). - (gcollect, swap.c) Made the GC status variable less reliant on magic numbers. - (swap, gcollect.c, object.c, simulate.c) When removing the swap entry for a modified program, the code now also makes sure that the linenumbers are loaded back into memory. 03-Apr-2002 (Lars Duening) (3.2.9-dev.426) - (prolang.y) '<string> *= <int>' didn't compile (reported by Dafire and Menaures). - (swap, gcollect.c, object.c, simulate.c) A hunch: when a program is swapped for the first time, the driver writes the struct as it is to the disk. On later swap-ins, this data is read directly into memory, but never written again. If in the meantime the prog->blueprint pointer changed, this change will be lost after the next swap in. It would require a rare sequence of activities to make this a problem, but it's possible. The change is therefore to remove the swap block whenever prog->blueprint changes; let's see if this helps (b-020228-0). 31-Mar-2002 (Lars Duening) (3.2.9-dev.425) - (simulate.c) Load errors during the initial master load crashed the driver in the error handling. 30-Mar-2002 (Lars Duening) (3.2.9-dev.424) - (driver.h) The blueprint code helped some people (but not all), so I re-activated it as a counter-check. - (bugs/) Updated the files. 27-Mar-2002 (Lars Duening) (3.2.9-dev.423) - (mapping.c) *sigh* The mapping check itself contained a fatal error (reported by Freaky). - (wiz_list.c) Don't call master::get_wiz_name() when there is no master. Otherwise a simple warning during the initial master load would abort the driver (reported by Frank Syse). - (driver.h, lots of files) If debug macro NO_BLUEPRINT is defined, the driver no longer tracks the blueprint for a program structure. This somewhat disables the blueprint() efun, but if it solves the object refcount problem, I know at least where to look. - (func_spec, actions, interpret.c) Efun query_verb() now takes an optional flag to return the verb as specified in the add_action() instead of the verb given by the user (suggested by Dafire). 26-Mar-2002 (Lars Duening) (3.2.9-dev.422) - (array.c, mapping.c) Made sure that the callback structures are properly released on errors. 25-Mar-2002 (Lars Duening) - (lex.c) *sigh* 'ifstate_t' is also a system type (reported by Alwin). - (comm.c) Removing the last pending input_to() did not reset the 'input_to pending' flag in the interactive structure (reported by Gnomi). - (mapping) The inaccuracy in the mappings statistics is caused by impossibly high numbers of mapping elements. The checks done by CHECK_MAPPING_TOTAL have been tightened around possible places where the too big number could be created. 21-Mar-2002 (Lars Duening) (3.2.9-dev.421) - (lex.c) The lexer's 'struct ifstate' clashes with an 'enum ifstate' defined on some systems for the networking stack (reported by Alwin). - (interpret.c) Removed the negative indexing on control_stack and value stack by introducing dummy entries at the bottom of each. 19-Mar-2002 (Lars Duening) (3.2.9-dev.420) - (pkg-mysql.c) db_connect() didn't check if it got too many arguments, crashing in that case (reported by Menaures). - (interpret.c) CHECK_OBJECT_REF: Checking the destructed objects after every VM instruction used too much time - it is now limited to F_RETURN, vefuns and tefuns. 16-Mar-2002 (Lars Duening) (3.2.9-dev.419) - (lex.c) Fixed a memory leak, caused again by a piece of code which relied on free_svalue() not changing the svalue .type (reported by Alwin). - (simulate.c) Setting a driver hook from an unshared string leaked this string (reported by Alwin). - (array.c) And this should be the last piece of code which made assumptions about free_svalue(). - (func_spec, efuns) New efun blueprint(). 13-Mar-2002 (Lars Duening) (3.2.9-dev.418) - (mapping.c) Since free_svalue() no longer leaves the .type untouched, the CHECK_MAPPINGS sanity stumbled at one point (reported by Bardioc). 12-Mar-2002 (Lars Duening) (3.2.9-dev.417) - (lex.c) Efun expand_define() did not set the proper type for the return value - so far undiscovered because free_svalue() didn't use to change it. - (Makefiles) Changed default optimization level from High (-O4) to Medium (-O2), since compilers get notoriously unreliable at high optimization. 11-Mar-2002 (Lars Duening) (3.2.9-dev.416) - (simulate.c) The CHECK_OBJECT_REF check was still overzealous; unfortunately not with the FF mudlib I tested it with. Hmpf. - (xalloc.c) The heap/stack gap check didn't expect the stack to be located under the heap, as Cygwin does it. 10-Mar-2002 (Lars Duening) (3.2.9-dev.415) - (actions.c, simulate.c) The CHECK_OBJECT_REF check was overzealous and didn't track properly the changes in object sentences. Sorry. - (interpret.c) Efun all_environment() did not accept destructed objects as argument as documented (reported by Pulami and Gnomi). 09-Mar-2002 (Lars Duening) (3.2.9-dev.414) - (xalloc) The implementation of the heap/stack check didn't work on architectures with really big address spaces - corrected that. - (mapping.c, interpret.c, object.c, efuns.c) When adding new values to mapping, the code didn't check if the memory could be allocated (reported by Menaures). - (doc/efun) More german translations from Pulami. 07-Mar-2002 (Lars Duening) (3.2.9-dev.413) - (xalloc, smalloc.c interpret.c, efuns.c, object.c) Added check to prevent the stack from growing into the heap. - (smalloc.c) Some expected/found magic word logs were reversed. - (interpret.c) atan2() did not accept an integer as first argument (reported by Pulami). - (doc/efun) Pulami provided a whole bunch of german translations. - (interpret.c) to_int() on closures didn't work as advertised (reported by Pulami) 06-Mar-2002 (Lars Duening) (3.2.9-dev.412) - (simulate, interpret.c, gcollect.c) With CHECK_OBJECT_REF, the driver continually checks the consistency of the list of destructed objects. - (doc/efun/get_dir) Updated the manpage and added the german version (both provided by Pulami). 05-Mar-2002 (Lars Duening) (3.2.9-dev.411) - (efuns.c) deep_copy() didn't handle quoted arrays properly (reported by Gnomi). - (interpret.c) New efun unquote() (suggested by Gnomi). - (object, gcollect.c, driver.h) Debug macro CHECK_OBJECT_REF activates a trace of blueprint allocations/deallocations, to be used to debug the strange refcount bugs recently encountered (b-020228-0). 04-Mar-2002 (Lars Duening) (3.2.9-dev.410) - (prolang.y) Publically inheriting functions which had been inherited protected before now overrides the earlier inherit (f-020303-1). - (ptrtable.c) When allocating a new ptable, the .table[] member is cleared. - (simulate.c) New get_dir() flag GETDIR_PATH adds the full absolute pathname to the returned filenames (suggested by Pulami). - (interpret.c) If a value of the wrong type is passed to a switch, the driver no longer throws a runtime error but instead jumps to the 'default' case (suggested by Pulami). 03-Mar-2002 (Lars Duening) (3.2.9-dev.409) - (simulate.c) When the driver runs on a fatal error, it calls the master lfun notify_shutdown() with "Fatal Error" as argument before it dumps core. - (configure) Added a check for 'bison -y' when looking for YACC. 28-Feb-2002 (Lars Duening) (3.2.9-dev.408) - (simulate, efuns.c, gcollect.c, interpret.c) In addition to the last uncaught-error trace, debug_info() can now also return the last error trace in general. 27-Feb-2002 (Lars Duening) - (smalloc.c) Corrected the statistics adjustment during the freelist consolidation. - (backend.c) The check_alarm() method did not recognize the '--no-heart' option, eventually turning the alarm() back on (reported by Dafire). 26-Feb-2002 (Lars Duening) (3.2.9-dev.407) - (efuns.c) Adjusted the terminal_colour() sanity check to acknowledge a certain boundary condition; added a test to make sure that the indentation is smaller than the wrapping width (reported by Menaures). - (actions.c) An error in add_action() could cause the driver to free memory twice (reported by Menaures). - (mapping.c) Several allocations weren't checked for success (reported by Menaures). 22-Feb-2002 (Lars Duening) (3.2.9-dev.406) - (mapping.c, smalloc.c) Improved the diagnostic messages: all are now printed to the GC log where they belong. - (smalloc.c) Added more data to the statistics diagnostics, by default deactivated since I can debug it over time in OSB. - (simulate.c) Made the load_object_error() method resistant against object names with printf format specifications (reported by Gnomi). - (mudlib/uni-crasher) Added Menaures' test object to the archive. 21-Feb-2002 (Lars Duening) - (smalloc.c) Oops, the statistics diagnostic lacked the 'DEBUG:' tag. 20-Feb-2002 (Lars Duening) (3.2.9-dev.405) - (prolang.y) Variable definitions are now normal statements and as such no longer restricted to the beginning of a block (f-000430-1). 17-Feb-2002 (Lars Duening) (3.2.9-dev.404) - (prolang.y) In a for() loop, variables may now be declared without initialisation (e.g. 'for (int i; ...)') (suggested by Pulami). - (random.c) Tentatively modified the scaling of the raw random number to the desired range so as to avoid overflow problems (investigated by Collin Baillie for an Alpha machine). 15-Feb-2002 (Lars Duening) - (prolang.y) The byte move routine to implement call_other sefuns by using call_other moved bytes in the wrong direction, crashing the driver if this piece of code was executed (reported by Coogan). - (simul_efun.c) simul-efuns shadowing real efuns were not properly discarded, making it impossible to reactive them (which in a way was lucky as it made it possible to find the call_other bug). - (interpret.c) The error handling for a secure call did not properly roll back the control stack, leaking any closure referenced from it (b-990505-0). 13-Feb-2002 (Lars Duening) (3.2.9-dev.403) - (efuns.c) Efun copy_bits(): I accidentally had deleted the lines handling negative deststart values. The efun was also prone to numerical overflow effects (reported by Menaures). - (gcollect.c) Hmm, clear_object_ref() unconditionally cleared the blueprint reference - this should happen only for destructed blueprints. Dunno if this was responsible for the recent crashes in FinalFrontier. - (sprintf.c) sprintf("%-=10s", "a\nb\n") printed two extra newlines at the end instead of just one (reported by Gnomi). 11-Feb-2002 (Lars Duening) (3.2.9-dev.402) - (comm.c) If on shutdown players were already destructed in notify_shutdown(), the following remove_player() was passed destructed objects (reported by Dafire). - (efuns.c) Corrected a crasher in the new efun copy_bits(). - (efuns.c) copy_bits() malfunctioned if src range to copy ended in unset bits (reported by Menaures). - (efuns.c) In efun terminal_colour(), the wrapping analysis had to be separated from the initial length gathering since the analysis requires all lengths to be available. Without the separation, the driver ran into a fatal sanity error on some input values (reported by Dafire). - (actions.c) Argument errors in add_action() would leak the action sentence. 09-Feb-2002 (Lars Duening) (3.2.9-dev.401) - (efuns, func_spec) New efun copy_bits() to copy ranges of bitstrings (based on code provided by Menaures). - (interpret.c) And nobody noticed that set_bit() was broken :-) - (efuns.c) Efun terminal_colour(): In wrapped formatting, when a word is too long to fit on a line by itself and there are at least two characters space left on the current line, the long word is begun immediately. Previously, such long words were always printed on a line of their own (suggested by Fiona). 07-Feb-2002 (Lars Duening) (3.2.9-dev.400) - (ed, comm, efuns) Modified the print_prompt() routine so that the editor prompt is take directly from the editor buffer. This allows to get rid of the swapping of editor and normal command prompts, which may also have been responsible for the memory leaks reported in b-020205-0 (though I can't prove that). As an aside, the f_set_prompt() function moved from efuns.c to comm.c . - (actions.c) If notify_fail() was not used from an undestructed command giver, the given failure notice value leaked (b-020205-1). 07-Feb-2002 (Lars Duening) (3.2.9-dev.399) - (bugs/) Cleaned up the bug reports. 06-Feb-2002 (Lars Duening) (3.2.9-dev.398) - (sprintf.c) In justified column formatting, when a word is too long to fit on a line by itself and there are at least two characters space left on the current line, the long word is begun immediately. Previously, such long words were always printed on a line of their own (suggested by Fiona). 06-Feb-2002 (Lars Duening) (3.2.9-dev.397) - (bugs) Added more information about memory leaks. 05-Feb-2002 (Lars Duening) (3.2.9-dev.396) - (mapping.c) The diagnostic in count_mapping_size() forgot to check for a NULL uid (thanks, Ben!) - (prolang.y) Inherited public but undefined function were hidden by (simul-)efuns of the same name, making a resolution by crossdefinition through another inherited module impossible (reported by Fiona). 03-Feb-2002 (Lars Duening) (3.2.9-dev.395) - (simulate, prolang.y, lex.c) Added a limit safety check to limit_error_format(), and increased the buffer sizes for the error messages (reported by Gnomi). 01-Feb-2002 (Lars Duening) - (smalloc.c) Added a check routine for the 'Total small memory' statistics - I think there is a miscalculation in the freelist consolidation. 31-Jan-2002 (Lars Duening) (3.2.9-dev.394) - (mapping.c) Added diagnostic statements to count_mapping_size() in order to find the place where the mappings statistics is screwed up. - (interpret.c) If input_to() was used without command giver, a NULL pointer string was pushed onto the interpreter stack due to a missing check in the privilege_violation4() function (reported by Menaures). 30-Jan-2002 (Lars Duening) (3.2.9-dev.393) - (gcollect.c) The GC forgot to clear the refcounts in the known destructed objects, thus sometimes causing a 'First reference to program, but refcount != 0' fatal error (reported by Coogan, reproduced by Menaures). - (interpret.c) replace_program() without argument can now skip the extra inherits added by the compiler to accomodate virtual variables (reported by Nathan). 26-Jan-2002 (Lars Duening) (3.2.9-dev.392) - (prolang.y) The wildcarded supercall "*":: no longer throws an error if none of the inherited objects defines the called function. It fits into the semantics of the '*' wildcard ("any or none"), and it is practical for supercalls like "*"::reset() (suggested by Zonk@Finalfrontier via Dafire). 25-Jan-2002 (Lars Duening) (3.2.9-dev.391) - (object.c) Made an error message more informative. 23-Jan-2002 (Lars Duening) (3.2.9-dev.390) - (prolang.y) The '::' operator did not recognize the separate namespaces of local variables and functions (reported by Imp via Gnomi). - (func_spec, prolang.y, interpret.c) Added new bytecode instructions nconst1 and nnumber to handle negative constants better (especially __INT_MIN__) (reported by Menaures). - (efuns.c, interpret.c, mapping.c) check_map_for_destr() wasn't called before copying or subtracting mappings (reported by Coogan). - (strfuns) Added some 'const' qualifiers. - (simulate.c) Improved the loadtime error messages: they now print the inheritance chain (suggested by Gnomi). - (etc/memusage) Provided by Fiona - thanks! 17-Jan-2002 (Lars Duening) (3.2.9-dev.389) - (interpret.c) The division <int>/<float> used a random value for the divident (reported by Pulami). - (simul_efun, simulate.c) Destructing the simul-efun object immediately invalidates all simul-efuns; this should reduce the number of 'undefined function' errors when doing manual reloads of the simul-efun object. - (make_func.y) Removed a small memory leak. - (smalloc.c, prolang.y, closure.c) Modified some memory accesses in order to keep Memory Check tools like ZeroFault happy. - (regexp.c) The .from_ed member in a newly compiled regexp wasn't set. - (object.c) One variable wasn't set. 16-Jan-2002 (Lars Duening) - (simul_efun) When reloading the simul-efun objects, all previous sefuns are discarded and removed from the symbol table. Otherwise if the simul-efun object used cross-defined sefuns itself, the recompiled object would not recognize the cross-defined calls and instead pick up the older inactive entry from the symbol table. This problem may still occur when reloading a simul-efun object by hand (reported by Coogan and some other people). - (lex) Inferior identifier entries now have their own string reference. - (lex.h, prolang.y) The magic values for ident_t.u.global now have symbolic names. 15-Jan-2002 (Lars Duening) - (configure) Fiona did more analysis on the rxcache hash and found out that using a prime table size doesn't help. I set the size back to 8192; let's see how bad the collisions are at this small table size. 14-Jan-2002 (Lars Duening) - (interpret.c) The overflow error messages for float divisions named the wrong values (thanks, Guandalug!). 12-Jan-2002 (Lars Duening) (3.2.9-dev.388) - (configure.in) Fiona pointed out that just halving the rxcache size doesn't seem to increase the fill ratio. As an experiment I now set the default size to prime 8191. - (interpret.c) The float range checks stupidly compared against DBL_MIN for the low limit instead of -DBL_MAX (reported by Ben). - (stralloc.c, debug_info) Added a few more string table statistics. 10-Jan-2002 (Lars Duening) (3.2.9-dev.387) - (lex.c) Added new predefined macros __FLOAT_MIN__ and __FLOAT_MAX__ for the float value limits. - (interpret.c) Added checks for numeric overflows to float operations. Btw, <int> / <float> returned a wrong result. - (configure.in) Reduced the default size of the RXCACHE table to 8192 instead of 16384 entries. 08-Jan-2002 (Lars Duening) (3.2.9-dev.386) - (efuns.c) Efun terminal_colour() removed a crasher which happened when too many control sequences appeared in a line (reported by Peng). - (autoconf/configure.in) Added icc compiler option '-ipo' to highest optimisation tests. - (make_func.y, comm.c, lex.c, interpret.c) Removed a couple (but not all) warnings generated by the Intel Compiler (reported by Dafire). 03-Jan-2002 (Lars Duening) (3.2.9-dev.385) - (gcollect.c) Forgot to clear the refcounts in the list of destructed objects. Oops. - (object.c) free_prog() now also frees its blueprint reference, which should remove at least one major refcount leak. - (lp-245/obj/master.c, lp-245/obj/player.c) Small corrections. 02-Jan-2002 (Lars Duening) (3.2.9-dev.384) - (simulate.c) Corrected the misformatting in 'status'. - (simulate, comm.c) Corrected a potential crasher when bad arguments are passed to efun input_to(): the embedded callback structure wasn't initialized properly, so that the free_input_to() got garbage data instead of pointers (reported by Menaures). - (util/xerq/erq.c) Reduced the amount of crap printed in DEBUG configuration. - (gcollect.c) There seems to be a problem with leaking object refcounts, so that destructed objects are tagged as 'referenced' even if they aren't. The GC will eventually free the memory itself, but inavoidably mess up the object statistics. As a first measure the GC after the mark phase checks the list of destructed objects if any of the objects on it haven't been referenced yet. If so, the objects concerned are logged and then explicitely included into the list of objects to free. This way at least the object statistics will be kept in sync for the time being. 01-Jan-2002 (Lars Duening) (3.2.9-dev.383) - (gcollect.c, simulate, comm) The GC now removes all stale objects referenced from the interactive structures during the cleanup phase. As a side effect, callbacks may no longer reference destructed objects during the clearing/counting phase of a GC. - (actions, main.c, backend.c, efuns.c, gcollect.c, interpret.c, simulate, dumpstat, mudlib/sys/debug_info) Destructed objects are now kept in a separate list in order to aid mudadmins in analysing the memory usage of their muds. debug_info(DINFO_STATUS) returns the statistics in the new fields DID_ST_OBJECTS_NEWLY_DEST and _DEST; debug_info(DINFO_DUMP, "destructed") dumps the lists to a file; and the builtin commands 'status tables' and 'dumpallobj' have been updated accordingly. 31-Dec-2001 (Lars Duening) (3.2.9-dev.382) - (interpret.c) to_string() can now handle operator closures (suggested by Gnomi). 30-Dec-2001 (Lars Duening) (3.2.9-dev.381) - (interpret.c) abs(INT_MIN) now generates an overflow error (reported by Menaures). - (driver.h, mapping, smalloc) If debug macro CHECK_MAPPING_TOTAL is defined, ever change in the registered mapping size statistics is accompanied with a sanity check. If the check fails, a DEBUG message is generated on stderr. - (sprintf.c) When adding columns, a one-line column ending with a newline inserted the newline on-the spot instead of delaying it until the end of the line as it was done with multiline columns (reported by Gnomi). 29-Dec-2001 (Lars Duening) - (driver/dumpallobj, efun/debug_info) Explained the format of the OBJDUMP file. - (settings/tubmud) Updated (thanks, Coogan!) 28-Dec-2001 (Lars Duening) (3.2.9-dev.380) - (interpret.c) When calling a simul_efun in a vanished object, the local stackpointer of eval_instruction() was not synchronized with inter_sp after the call (reported by Gnomi). - (simul_efun.c) When reloading a simul_efun object, the check for old sefuns to reactivate ignored changes in the TYPE_MOD_XVARARGS flag. Since the exact format of the F_SIMUL_EFUN instruction depends on this flag, this could trip the bytecode interpreter (reported by Gnomi). - (mudlib/astar.c) An implementation of the A* pathfinding algorithm, donated by Acius (thanks!). 21-Dec-2001 (Lars Duening) (3.2.9-dev.379) - (simulate.c) When checking the refcounts, the extra_refs in objects named in callbacks weren't counted corrected (reported by Tomba). - (gcollect, call_out.c, simulate.c) Removed knowledge about the GC internals from call_outs and callbacks. 19-Dec-2001 (Lars Duening) (3.2.9-dev.378) - (interpret.c) Corrected a wrong variable initialisation in efun say() (reported by Nostradamus). 14-Dec-2001 (Lars Duening) - (interpret.c) Efun set_this_object() queried the master and simul-efun object without checking if they actually exist. 12-Dec-2001 (Lars Duening) (3.2.9-dev.377) - (call_out.c) Efun call_out_info() did not return the target object if the call_out is a closure (reported by Menaures). - (interpret.c) The increment and decrement operators were missing the overflow checks (reported by Freaky). - (lex.c) New LPC defines __INT_MAX__ and __INT_MIN__ for the numeric integer limits. - (interpret.c) Another overflow check, this time for __INT_MIN__ / -1 (reported by Menaures). - (interpret.c) Corrected the bugfix to the say() refcount problem (thanks, Tomba!) 11-Dec-2001 (Lars Duening) - (comm.c) TELOPT_TM can now be negotiated from the mudlib as well (reported by Peng). 11-Dec-2001 (Lars Duening) (3.2.9-dev.376) - (interpret.c) Additions, subtractions and multiplications now check for numeric overflows. 11-Dec-2001 (Lars Duening) (3.2.9-dev.375) - (interpret.c, mapping.c) Oops, the new checks also introduced divisions by 0 (reported by Menaures). - (sprintf.c) Runtime fieldsize specification by '%*' didn't check if the size was given as MIN_INT: it would try to negate it to get the positive number, but fail and cause the formatting routines to crash under circumstances (reported by Menaures). 10-Dec-2001 (Lars Duening) (3.2.9-dev.374) - (interpret.c) gcc 2.91 on Linux/x86 generates buggy code for MIN_INT % -1, so this case is now explicitely check for (reported by Menaures). - (interpret.c) Multiplication of a string with a number was not checked for numeric overflows in the computation of the result length (reported by Menaures). - (smalloc.c) Added a check against numeric overflows in the computation of the final memory block size. - (mapping.c) When allocating/resizing mappings, a check against numeric overflows in the size computations has been added (reported Menaures). - 3.2.9-dev.373 had an extraneous debug statement. 05-Dec-2001 (Lars Duening) (3.2.9-dev.372) - (backend.c) Before calling reset() or clean_up() from the backend, the previous_object is cleared (reported by Gnomi). - (main.c) New commandline option: --random-seed <num> Seed value for the random number generator. If not given, the driver chooses a seed value on its own. This option is for debugging. - (interpret.c) The mysterious code in calling alien lfun closures to create a second control frame, deactivated in dev.318, revealed its purpose: when for such a closure the bound object l->ob differs from the code object l->function.alien.ob, this fact needs to be recorded in order to prevent security holes. So the code fragment has been activated again with a small change: the second control frame will be created only if the two objects differ (unknowingly suggested by Menaures). 04-Dec-2001 (Lars Duening) (3.2.9-dev.371) - (interpret.c) When doing a call_other on an array, inter_sp was incorrect while getting the object. An error there then could clobber memory. - (backend) Added in internal flag 'mud_is_up' for debugging. - (main.c) The initial random seed is printed for debugging purposes. 28-Nov-2001 (Lars Duening) (3.2.9-dev.370) - (interpret.c) In-place execution of invert_bits() returned an invalid svalue. - (prolang.y) Another correction to the bugfix of Gimli's problem (reported by Freaky). - (lex.c) Line number counting was thrown off by one by the auto-include string (reported by Menaures and Nathan). 27-Nov-2001 (Lars Duening) (3.2.9-dev.369) - (prolang.y) The bugfix to Gimlis problem unfortunately broke a couple of muds. This version works better (reported by many people). - (array.c) Removed a memory leak in inherit_list()s error handling. 25-Nov-2001 (Lars Duening) (3.2.9-dev.368) - (mudlib/sys/functionlist.h) Changed the <> include of lpctypes.h into a "" include (suggested by Heiko). - (prolang.y) Inheriting two private functions of the same caused the compiler to crossdefine the second one to the first. Not directly visible to the user, this messed up any use of these functions by INITIALIZATION_BY___INIT (reported by Gimli). - (interpret.c) Efun replace_program() can omit the argument if the object has only one inherit (suggested by Pulami). 23-Nov-2001 (Lars Duening) (3.2.9-dev.367) - (object.c) Efun say() could leak object references (reported by Tomba). - (mudlib/sys/functionlist.h) Changed the absolute include of lpctypes.h into a <> include (suggested by Heiko). 21-Nov-2001 (Lars Duening) - (pkg-mysql.c) Using the mysql->thread_id as handle proved to be unreliable as the thread_id may change. Instead an explicit handle is now used (thanks, Fuchur!). - (ed.c) Added command 'T' to perform Tab <-> Whitespace conversion (thanks, Menaures!). 20-Nov-2001 (Lars Duening) (3.2.9-dev.366) - (sprintf.c) Printing a table modified the internal precision information without making a backup. If now the table was applied to an array of data ("%#99999999999@s" format string), formatting the second line would cause a crash (reported by Menaures). - (doc/efun/snoop) Corrected the manpage: snooping an already snooped player terminates the previous snoop (reported by Zordiac). 19-Nov-2001 (Lars Duening) (3.2.9-dev.365) - (sprintf.c) It was possible to instruct sprintf to format a string as both columns and as a table. sprintf() then formatted it as columns, but any error (like buffer overflow) would crash the driver. The error recovery is now more robust, and such an invalid format string is rejected (reported by Menaures). - (sprintf.c) Max size of the result is now 128 KByte. - (lex.c) A H_AUTO_INCLUDE closure also receives a flag telling if it was a <>-type include or not (suggested by Freaky). - (interpret.c) Efuns and_bits(), or_bits(), xor_bits() mishandled the bits 5, 11, 17, ... (thanks, Fuchur!) - (prolang.y) The bugfix yesterday was not quite correct: a pair of missing parentheses completely changed the meaning of the check for private inherited functions, causing the driver to crash (reported by Freaky). 18-Nov-2001 (Lars Duening) (3.2.9-dev.364) - (array.c) Efun inherit_list: the tagging of virtual inherits returned a random value for the main program (reported by Freaky). - (prolang.y) Inherited private functions could hide simul-efuns or efuns (reported by Gnomi). - (lex.c) If H_AUTO_INCLUDE is set to a closure, it will be called for all files opened by the lexer. A second argument passed to the closure determines whether the call is for the object file itself or for an included file (suggested by Freaky). - (func_spec, array, exec.h, prolang.y) New efun include_list() to return the filenames read when compiling a program. - (efuns.c) Added the number of included files to the result of object_info(OINFO_MEMORY). 17-Nov-2001 (Lars Duening) (3.2.9-dev.363) - (comm.c) Oops, two bugs found. 17-Nov-2001 (Lars Duening) (3.2.9-dev.362) - (comm) Efun input_to() now accepts a prompt parameter, specifying the prompt to print. - (array.c, exec.h, mudlib/sys/inherit_list.h) New flag for efun inherit_list() allows to tag virtual inherits in the result (rest of f-991020-0). - (doc/concepts/hooks) Corrected the docs for the MODIFY_COMMAND hooks. 17-Nov-2001 (Lars Duening) (3.2.9-dev.361) - (func_spec, prolang.y, make_func, lex, simulate) Replaced the efun set_auto_include_string() with the driver hook H_AUTO_INCLUDE. Additionally, a closure can be used to generate the auto include data (f-000505, f-011026-2, f-990316-1). - (simulate.c) Removed a small memory leak in set_driver_hook(): replacing hooks leaked the old hook data. 16-Nov-2001 (Lars Duening) (3.2.9-dev.360) - (sys/debug_info.h, smalloc.c) Efun debug_info(): macro DID_SBRK_SIZE was typoed as DID_SBKR_SIZE. - (sys/debug_info.h, smalloc.c) Efun debug_info(): added the derived statistics DID_MEM_ALLOCATED, _USED and _TOTAL_UNUSED to the DINFO_DATA:DID_MEMORY data. - (interpret.c) The stack overflow error handling forgot to update the stack pointer after cleaning up the stack, causing some svalues to be freed twice (reported by Coogan). - (efuns.c) Efun terminal_colour(): Temporary values are no longer held directly on the stack, but instead in an array on the stack, thus greatly reducing the stack usage for strings with lots of colour codes. 15-Nov-2001 (Lars Duening) (3.2.9-dev.359) - (lex.c) If H_INCLUDE_DIRS is set to an array of directory names, the names can now have leading '/' (reported by Nathan). - (mapping.c) check_map_for_destr() didn't check if the mapping actually had a condensed part - this can happen when called from compact_mappings() (reported by Freaky). - (interpret.c) Improved the stack overflow checking a bit, and for some function calls the local variables were allocated twice. - (interpret.c) member() and member_array() used with closures didn't always find them (reported by Gnomi). 14-Nov-2001 (Lars Duening) (3.2.9-dev.358) - (gcollect.c) Tweaks to the information dumped on fatals. - (driver.h, gcollect) New debug macro DUMP_GC_REFS (in driver.h) instructs the GC to list all found references on the gcollect_outfd. Off by default, this macro should be defined only by people getting fatal()s like 'First reference to destructed object, but ref count != 0'. - (main) New function dprintf4(). - (mapping.c) compact_mappings() did not remove destructed objects and closures from the mapping (reported by Menaures). 13-Nov-2001 (Lars Duening) (3.2.9-dev.357) - (gcollect.c, smalloc.c) Removed the compilation problem when MALLOC_LPC_TRACE is not defined (reported by Menaures). - (smalloc.c) The computation of a 'magic' word in the freelist consolidation was wrong, causing bus errors in Tubmud. - (mud/lp-245) Modified to no longer use deprecated functions (thanks, Coogan!) - (simulate.c) Added missing newlines to some error messages (thanks, Freaky!) - (efuns.c) Efun transfer() miscalculated the weights when the target wasn't a living (reported by Zordiac). - (driver.h) Since FinalFrontier is now running two weeks without a problem, I deactivated the mapping check code. - And Zordiac informed me that Deeper Trouble is now running natively on LDMud. 12-Nov-2001 (Lars Duening) (3.2.9-dev.356) - (make_func.y) Lexer now understands lines terminated with CRLFs instead of just LFs (reported by Holger). - (gcollect.c) When a 'First ref to' with unmatching refcount is encountered, the memory allocation information is dumped, too. 24-Oct-2001 (Lars Duening) (3.2.9-dev.355) - (object.c) Efun inherit_list() can now return the inheritance structure as a tree (part of f-991020-0). - (backend.c) If a closure is used with regreplace(), it will be passed the position of the found substring, too (suggested by Alfe). - (comm.c) New efun remove_input_to() (part of f-000530). 23-Oct-2001 (Lars Duening) - (doc/efun/terminal_colour) Improved the explanation of the key recognition strategy, and added a MudOS simulation function courtesy of Fiona. 21-Oct-2001 (Lars Duening) (3.2.9-dev.354) - (actions.c) Removed a crasher when add_action() was performed between two objects with no environment (thanks, Fuchur!). 18-Oct-2001 (Lars Duening) (3.2.9-dev.353) - (mapping.c) The search loops for condensed misc keys terminated one iteration too late before declaring a key as not found, which occasionally led to 'finding' a key which wasn't really there. Especially in remove_mapping() this led to clobbered memory. This probably fixes b-000218, b-000822 and b-000207; maybe even some of the other strange crashers. - (sprintf.c) Removed a crasher in justified formatting when the string to format contained only one word (reported by Slava and Coogan). - (doc/LPC/arrays, doc/efun/sprintf) Improved examples by Coogan (thanks!). 17-Oct-2001 (Lars Duening) (3.2.9-dev.352) - (interpret.c) I had to deactivate the check against negative range assignments - this feature is in fact in use (thanks, Coogan and Alfe - and sorry!) 14-Oct-2001 (Lars Duening) (3.2.9-dev.351) - (mapping, xalloc.c) Didn't compile with sysmalloc as allocator. 12-Oct-2001 (Lars Duening) (3.2.9-dev.350) - (sprintf.c) Improved the flushing algorithm: spaces are more evenly distributed, newlines in columns didn't work properly (reported by Slava). - (lots of files) Renamed closure_hook[] to the more sensical driver_hook[]. - (erq, xerq) Erq now accepts an '--execdir' argument to override the default ERQ_DIR setting (suggested by Freaky). - (main, comm.c) It is now possible to pass simple arguments to the erq with the --erq commandline option (suggested by Freaky). 11-Oct-2001 (Lars Duening) (3.2.9-dev.349) - (sprintf.c) New formatting modifier '$' allows flush formatting of strings and columns of strings (f-010717). 09-Oct-2001 (Lars Duening) - (gcollect.c) Increased the refcount of an destructed referenced object before recursing into the substructures - I have the feeling that there was a possible race condition, but I can't prove it. 08-Oct-2001 (Lars Duening) (3.2.9-dev.348) - (prolang.y) Function definition errors and warnings now all print the name of the function involved (suggested by Freaky). - (simulate.c) Shadowing a nomask function now prints the names of objects involved as well (suggested by Freaky). 07-Oct-2001 (Lars Duening) (3.2.9-dev.347) - (mapping.c) Darn, I introduced a bug in the diagnostic code of the last patch. Sorry! (reported by Freaky) 06-Oct-2001 (Lars Duening) (3.2.9-dev.346) - (mapping) Added some more diagnostic regarding the Finalfrontier bug: the culprit really seems to be in remove_mapping(). Now, when the bug is detected, a trace of the activities in remove_mapping() is printed as well. 06-Oct-2001 (Lars Duening) (3.2.9-dev.345) - (prolang.y) Warnings about redefinition of visibility apply to prototype/functions only, but no longer to redefinitions of inherited functions. 05-Oct-2001 (Lars Duening) - (lex.c) Termination of a macro definition in an include file with End-of-file confused the lexer (reported by Tomi). - (comm) If a message had to be discarded due to contention (EWOULDBLOCK), the user will receive a information message at the next possible opportunity. 04-Oct-2001 (Lars Duening) - (efuns) Renamed efun md5_encrypt() to md5(). 03-Oct-2001 (Lars Duening) (3.2.9-dev.344) - (prolang.y) Prototypes no longer modify the definition of inherited functions in strange ways (reported by Fiona). 01-Oct-2001 (Lars Duening) (3.2.9-dev.343) - (prolang.y) The check for redefinition of 'nomask' functions no longer worked (reported by Gimli). 30-Sep-2001 (Lars Duening) (3.2.9-dev.342) - (lex.c) Added a warning about empty #pragmas (reported by Slava). - (interpret.c) Tracing no longer crashes when the command_giver turns non-interactive (reported by Dafire). 28-Sep-2001 (Lars Duening) (3.2.9-dev.341) - (lex.c) Spaces before the lineend caused the lexer to complain about an unrecognized pragma, even though the pragma had been recognized (reported by Dafire and Slava). - (efuns.c) Efun terminal_colour(): The escape sequence '%%^^' is now recognized and replaced by '%^' as is '%^%^' already. This is for compatibility with MudOS drivers (suggested by Fiona). - (telnet.h, comm.c) Added support for mudlib negotiation of MCP, MSP and MXP through the normal H_TELNET_NEG hook (suggested by Markus Peter). - (interpret.c, lex.c, smalloc.c) Made it work on Linux s/390 (a machine where char's are by default unsigned) (provided by Freaky). 27-Sep-2001 (Lars Duening) (3.2.9-dev.340) - (lex.c) Pragma recognition no longer stumbles over CRs at the line end (reported by Freaky and Slava). - (interpret.c) The idiom of assigning to a range [i..i-1] in order to insert elements is legal again - a lot of mudlibs already use it (reported by Andre). 26-Sep-2001 (Lars Duening) (3.2.9-dev.339) - (mapping.c) Finalfrontier was caught by the diagnostic added in dev.337, this time in remove_mapping(). I added more diagnostics there. - (interpret.c) Assignments to negative length ranges (eg. [3..2]) are caught and throw an error (reported by Gimli and Pulami). - (gcollect.c) Printing memory information during a GC is more robust against string NULL pointers (reported by Freaky). - (lex.c) It is possible to specify multiple pragmas in one #pragma statement, separated by commas (suggested by Largo). 18-Sep-2001 (Lars Duening) (3.2.9-dev.338) - (mapping.c) Somehow a syntax error managed to get into the file after I tested it :-( (reported by Dafire and Freaky). 17-Sep-2001 (Lars Duening) (3.2.9-dev.337) - (mapping.c) Some more diagnostic code for the hash->condensed_deleted suspect. Also, last_instructions() is no longer called when not using TRACE_CODE (reported by Fiona). 08-Sep-2001 (Lars Duening) (3.2.9-dev.336) - (interpret.c) The test if a function call should be trace or not was done against the command_giver, not against the current_interactive as documented (reported by Gnomi). - (simulate.c) When entering a fatal(), the given format string is printed as is to stderr (just a precaution). - (smalloc.c) The 'magic match' failed messages now give the expected block address and sizes, too. - (lex.c) Added warning when an unknown pragma is encountered (suggested by Robert). 06-Sep-2001 (Lars Duening) (3.2.9-dev.335) - (mapping.c) Some more diagnostic code for the mapping bug: at the moment I suspect that hash->condensed_deleted gets out of sync. 05-Sep-2001 (Lars Duening) (3.2.9-dev.334) - (prolang.y) The driver couldn't handle (multiple?) second-level inherits of a 'private nomask' function, which inline closure now are (reported by Bardioc). 03-Sep-2001 (Lars Duening) (3.2.9-dev.333) - (interpret.c, mapping) Some more debugging code: if an illegal mapping size is detected, the MALLOC_TRACE information is printed. I also added checks to certain suspicious VM instructions - I hope that I won't have to check all mappings after each instruction. - (xalloc, smalloc) New helper function dump_malloc_trace() to, well, dump the malloc trace if there is one. 31-Aug-2001 (Lars Duening) (3.2.9-dev.332) - (efuns.c) Well, the recent change didn't fix terminal_colour(), so I changed it back again. - (lex.c) The lfuns implementing inline closures are now 'private nomask', which is the real solution unlike the one implemented in 3.2.9-dev.260 (suggested by Freaky). - (efuns.c) If terminal_colour() was used with a closure returning a computed string, the ever call (but the first) within the same formatting clobbered the result from the previous one (reported by Ben and Elric). - (swap.c) Oops, when adjusting the allocation statistics after swapping out variables, the driver occasionally tried to dereference a non-existing pointer (reported by Ben and Nostradamus). - (mudlib/sys/lpcytpes.h) The values to be returned by functionlist() were out of sync with what functionlist() really returned (reported by Freaky). 30-Aug-2001 (Lars Duening) (3.2.9-dev.331) - (mapping.c) Since the Finalfrontier mapping bug still occurs, I added some more instrumentation to the module. - (efuns.c) Tentative change to correct an obscure bug in the terminal_colour() code. Let's see if it's going to work. 29-Aug-2001 (Lars Duening) (3.2.9-dev.330) - (smalloc.c) Corrected a small statistics problem in the freelist consolidation: some unusable space was counted twice (b-010713-0, b-000830). - (swap.c) When swapping variables, the object allocation size statistics wasn't updated. - (lex.c) Multi-dimensional array indices in macro calls are no longer confused with multiple macro arguments (reported by Freaky). 28-Aug-2001 (Lars Duening) (3.2.9-dev.329) - (interpret.c) More corrections to the refcount checking code. Furthermore, an failure no longer crashes the driver (but generates lots of debugging output) (reported by Tomba). - (pkg-mysql.c) Removed a memory leak caused by the result set not begin freed after all the data has been fetched. - (backend.c) Added safeguard so that slow_shut_down() is not called repeated while a previous one is still pending (suggested by Dafire). 28-Aug-2001 (Lars Duening) (3.2.9-dev.328) - (interpret.c) The reference loop check was too strict and also flagged legal assignments (thanks, Dan). 27-Aug-2001 (Lars Duening) (3.2.9-dev.327) - (interpret.c) Improved the check for reference loop creation on assignment (suggested by Menaures). - (interpret.c) When reloading the master object, the driver freed through the driver hooks some object references it shouldn't free (reported by Tomba). 26-Aug-2001 (Lars Duening) (3.2.9-dev.326) - (simulate.c) Running the driver with '-d -d' crashed it when loading objects. - (interpret.c) The refcount check for got to count the blueprint reference in programs (reported by Tomba). - (smalloc.c) The check against MAX_MALLOCED was incomplete and could lead to use of foreign memory (reported by Menaures). 24-Aug-2001 (Lars Duening) - (func_spec) Even though the efun get_type_info() was supposed to take an arbitrary type as second argument, it wasn't declared that way (thanks, Fiona!). 23-Aug-2001 (Lars Duening) (3.2.9-dev.325) - (array.c) ERQ callbacks for a destructed object don't have a current_object at the time the error occurs. The error handling code allocates an array, but the array code was still assuming that current_object is always valid (b-990217-4) (reported by Fiona). - (prolang.y) Redefining an inherited private function with a non-private one, and then inheriting the redefinition didn't work (reported by Menaures). 22-Aug-2001 (Lars Duening) (3.2.9-dev.324) - (interpret.c) The lambda protection had a bug: when a non-lambda closure was called from within an executed lambda, the lambda closure was accidentally freed (reported by Fiona). - (sprintf.c) *blush* I completely forgot to add the terminator 0 after creating the quoted representation of a string (thanks, Menaures!). 21-Aug-2001 (Lars Duening) (3.2.9-dev.323) - (prolang.y) The 'private' correction caused the driver to crash under obscure circumstances. I implemented the correction now in a different way, and it seems to work now (reported by Bardioc and many others). - (prolang.y) If an efun was called with too many arguments, the extraneous args were popped twice from the argument stack, so that subsequent function calls clobbered memory (reported by Dafire). - (sprintf.c) For the new '%Q' format, string length is now computed with strlen() instead of svalue_strlen() - the latter is too sensitive towards constructed strings (reported by Menaures). - (simulate.c) Added an error when the time left for a catch() is less than the reserved cost. 19-Aug-2001 (Lars Duening) (3.2.9-dev.322) - (prolang.y) Defining a function which is also a private function in an inherited object no longer redefines the inherited one (reported by Bardioc and Menaures). - (sprintf.c) Closures on inherited lfuns are printed with the defining program for the %O format. - (sprintf.c) New format specifier 'Q' acts like 'O', except that special characters in strings are quoted (suggested by Menaures). 15-Aug-2001 (Lars Duening) (3.2.9-dev.321) - (interpret.c) Holding a reference on a lambda works much better if the code actually increment the correct refcount. 14-Aug-2001 (Lars Duening) (3.2.9-dev.320) - (interpret) The VM now holds a reference on a lambda while it is executed, preventing premature destructions (reported by Dan and Fiona). - (simulate, actions.c, backend.c, comm.c, gcollect.c) Since the VM now protects lambdas during their execution, the free_closure_hooks() hack is no longer needed. For now, the code has been just deactivated. - (interpret.c) Corrected late-night errors in efuns ceil() and floor() (reported by Niels). 13-Aug-2001 (Lars Duening) - (mapping.c) Adding a mapping to itself via 'm += m' freed svalues first, then used them as data for assignment (reported by Patryn). 12-Aug-2001 (Lars Duening) - (func_spec) Somehow the change for sizeof() didn't make it into the distribution (reported by Menaures). 07-Aug-2001 (Lars Duening) (3.2.9-dev.319) - (lex.c) The recognition of numeric character constants was slightly wrong (thanks, Fiona). 06-Aug-2001 (Lars Duening) - (interpret.c) Efun program_name() now accepts 0 as argument and pass it through (suggested by Niels). - (func_spec, interpret.c) Efuns floor() and ceil() also accept integer argument values (suggested by Niels). 05-Aug-2001 (Lars Duening) (3.2.9-dev.318) - (simulate.c) Destructing an object with a single shadow crashed the driver (reported by Craig). - (interpret.c) Efuns load_name() and object_name() now accept 0 as argument and pass it through (suggested by Niels). - (interpret.c) Calling an alien-lfun created two control frames instead of just one, with no obvious purpose. I removed one of the frames but commenting out the code, so if it is needed nonetheless, it will be easy to reactivate (reported by Niels). - (interpret.c) For the first 255 simul-efuns it is now permissible to receive references in a varargs parameter, e.g. to implement a simulated call_other(). Handle with care! (reported by Coogan) - (configure) Improved the search for mySQL libraries - it is amazing in how many different ways this package is installed. 02-Aug-2001 (Lars Duening) - (interpret.c) Corrected a wrong function argument (reported by Freaky). 01-Aug-2001 (Lars Duening) (3.2.9-dev.317) - (prolang.y) The error message for the case that a wildcarded super call tried to pass arguments was confusing (reported by Gimli). - (interpret.c) Efun sizeof() can be used on strings (suggested by Niels). 31-Jul-2001 (Lars Duening) (3.2.9-dev.316) - (pkg-mysql.c) Fixed a syntax error. It sucks that I can't compile this file at home, doesn't it? - (lex.c) A preprocessor statement following a macro redefinition wasn't recognized because the macro parsing code clobbered the newline character (thanks, Fuchur!). 26-Jul-2001 (Lars Duening) (3.2.9-dev.315) - (lex.c) The redefinition of a macro messed up the list of defines, eventually crashing the driver (reported by Freaky). - (efuns.c) Using terminal_colour() with a closure as lookup method an a non-tabled keyword crashed the driver (reported by Elric). - (simulate.c) Clones no longer inherit the current O_WILL_CLEAN_UP flag from their blueprint - it didn't make sense (thanks, Niels!). 25-Jul-2001 (Lars Duening) (3.2.9-dev.314) - (sprintf.c) Printing numbers with leading zeroes and a sign character printed the sign in the wrong place (reported by Dafire). - (backend.c) The check_alarm() function did not quite work right (thanks, Coogan!). - (backend.c) Efun regreplace(): Pattern which could match on the empty string matched once too often at string end. E.g. regreplace("blah fasel", ".*", "ERSATZ", 1) returned "ERSATZERSATZ" instead of just "ERSATZ" (thanks, Fiona!). - (typedefs.h) Upped the StrRefCount type to 'unsigned long' (b-010320). - (lex.c) Redefinition of a macro yields a warning even if the definition doesn't change (suggested by Gnomi). - (backend.c, simulate.c) The initial setting of the O_WILL_CLEAN_UP flag checked for the hardcoded "clean_up" lfun instead for what the H_CLEAN_UP hook said. Also, the documentation was slightly out of sync with the reality (thanks, Niels!). - (mysql.c) db_error() used a non-existing argument (thanks, Dafire!). 17-Jul-2001 (Lars Duening) (3.2.9-dev.313) - (array, gcollect.c, mapping) The GC now recounts the number of arrays and mappings in the game (reported by Niels). - (prolang.y) If the compiler detects a function call with too many arguments, the subsequent argument type checks were wrong (reported by Fiona). - (sprintf.c) Instead of using the new modifier '$', the efun now uses the existence of an explicite pad string specification as flag whether to remove pad space before new lines or not. This is actually more logical to use than a modifier (suggested by Largo). - (actions.c) The special command 'malloc' now follows the 'is_wizard' handling of the other special commands (thanks, Niels). 16-Jul-2001 (Lars Duening) (3.2.9-dev.312) - (gcollect.c) Oops. Freeing a circular array reference with MALLOC_LPC_TRACE enabled crashed the driver in the dump of the array contents. 15-Jul-2001 (Lars Duening) (3.2.9-dev.311) - (actions, simulate) Removing or destructing a shadow now properly removes all associated actions (reported by Nathan). - (actions.c) New add_action() flag AA_IMM_ARGS implements the sane NOSPACE argument handling. 15-Jul-2001 (Lars Duening) (3.2.9-dev.310) - (configure) When testing for mySQL, the driver automatically checks in /usr/local/include as well (some systems don't have it the default include path). - (configure) Added a check for 'minimal optimization' options. - (configure) When compiling on Windows, link in a couple of more libraries. This might not be fully correct yet. - (make_func.y) Removed a couple of warnings under gcc 3.0 (thanks, Largo!). - (comm.c) On a regular shutdown, stop the erq demon explicitely. - (main.c) When compiling for MSDOS_FS, the erq pathname wasn't built correctly (reported by Coogan). - (sprintf.c) Simul-efun closures printed with %O are marked as such. - (closure.c) Comparison of operator, efun, and simul-efun closures was instable (reported by Nathan). 13-Jul-2001 (Lars Duening) (3.2.9-dev.309) - (interpret.c, comm.c, parse.c, parse_old.c) Removed a couple of undefined behaviours, detected by gcc 3.0 (thanks, Largo!) - (sprintf.c) Format modifier '$' instructs (s)printf() to not remove padding white space before a lineend. - (efuns) New efuns gmtime() and localtime() to return the time broken up into its constituents (suggested by Markus Peter). - (lex.c) Escaped character constants are now only parsed up to a maximum length. If no valid digits are found, the whole item is simply treated as escaped character (thanks, Fiona!). 05-Jul-2001 (Lars Duening) (3.2.9-dev.308) - (prolang.y) It is now possible again to redefine a non-static inherited variable with a static one (or vice versa), but it also generates a warning. - (doc/concepts/intermud*) Intermud-2 documentation added (thanks, Fiona!) 03-Jul-2001 (Lars Duening) (3.2.9-dev.307) - (prolang.y) How embarrassing! The code for !INITIALIZATION_BY_INIT had a bad syntax error - and I didn't notice it (thanks, Bardioc!) 01-Jul-2001 (Lars Duening) (3.2.9-dev.306) - (configure) 'with_wizlist_file=no' didn't work from settings files (reported by Freaky). - (configure) A superfluous 'fi' broke the configure. - (lex.c) '//' comments didn't like to be terminated by the end of file (reported by Matthew). 30-Jun-2001 (Lars Duening) (3.2.9-dev.304 and dev.305) - (hosts/os2) Updated the OS/2 port (provided by Bart). 30-Jun-2001 (Lars Duening) (3.2.9-dev.303) - (prolang.y) Relaxed type checking somewhat so that a non-mixed function argument can be redefined to mixed, and vice versa (suggested by Gnomi) - (lex.c) Escaped character constants in inline closures weren't recognized (reported by Freaky and Gnomi) - (doc/LPC/comments) New. - (configure.in) When using mySQL, also check for libnsl. - (configure.in) The mySQL path is now used for both include and library searches. 29-Jun-2001 (Lars Duening) (3.2.9-dev.302) - (prolang.y) Accidentally it was no longer possible to redefine privately inherited variables (reported by Gnomi). - (actions.c, simulate.c) New hook H_SEND_NOTIFY_FAIL to deliver the notify fail message. It is meant for those muds who want control over the command flow without having to implement everything themselves via the H_COMMAND hook. This hook is still experimental and may change in details (suggested by Gnomi). - (prolang.y) It is now possible to define a default visibility for functions and variables (see doc/LPC/modifiers) (suggested by Freaky). 29-Jun-2001 (Lars Duening) (3.2.9-dev.301) - (main.c) Commandline arguments can now be read from a textfile, using new commandline option '--args'. These argument files can be nested. (suggested by Elric) 28-Jun-2001 (Lars Duening) (3.2.9-dev.300) - (configure) The evaluation of the '--with-wizlist-file' configuration option was less than flawless. - (main.c) Instead of scanning the arguments twice, all encountered -f options are stored in a temporary list and evaluated when it's time (f-010603-1). - (main.c) Laid the groundwork for nested sources for the command arguments. 27-Jun-2001 (Lars Duening) (3.2.9-dev.299) - (simulate.c) When parsing callback arguments, the pattern 'function, object, extra' was parsed such that 'object' became part of the 'extra' arguments (reported by Marthin). - (lex.c) It is now possible to use symbols in inline closures, ie. "(: 'a :)" (reported by Menaures). - (main.c, lex.c, wiz_list, configure) The name of the WIZLIST file can now be configured at compile time and with commandline option '--wizlist-file'. The LPC macro __WIZLIST__ expands to the selected name. The WIZLIST file can also be disabled completely ('--no-wizlist-file'), in which case __WIZLIST__ is not defined at all (based on a patch from Elric). - (call_out.c) Corrected a fatal refcount bug in the garbage collection of call_outs with destructed command givers (reported by Gnomi). 22-Jun-2001 (Lars Duening) - (sprintf.c) Column-formatting a string with one newline printed one newline too few if the formatstring didn't end immediately (reported by Gnomi). 21-Jun-2001 (Lars Duening) - (make_func.y, lex.c) Identifier names can use the extended alpha characters from the ISO 8859-1 alphabet (suggested by Elric). 17-Jun-2001 (Lars Duening) (3.2.9-dev.298) - (backend.c) cleanup_stuff() also executes a clear_state() to avoid dangling object references (reported by Gnomi). - (pkg-mysql.c) Added efun db_error() to return the most recent error string (thanks to Casey and Dafire). - (efuns.c) If the driver is compiled with DYNAMIC_COSTS, the cost of efun clones() is proportional to the number of objects in the game. 13-Jun-2001 (Lars Duening) (3.2.9-dev.297) - (prolang.y) Global variables can no longer be redefined (reported by Coogan). - (doc/efun/operators) Clarified the behaviour of (<type>) casts. - (simulate.c, interpret.c) Normal runtime errors (and their traceback) are no longer reported on stdout at all (reported by Tomi). - (etc/startmud.pl) Added a simple start/keep-alive script - feel free to improve on it (thanks, Krystal-Anne!) 11-Jun-2001 (Lars Duening) (3.2.9-dev.296) - (interpret) Nifty little crasher: if the object holding the closure currently execute is destructed, the interpreter happily tried to free the closure after execution without noticing that it already has been freed (as part of the object destruction process) (reported by Freaky). 09-Jun-2001 (Lars Duening) (3.2.9-dev.295) - (main.c) Improved the formatting of the language options in the '--options' output. 07-Jun-2001 (Lars Duening) - (comm.c) The driver forgot that it is in noecho/charmode mode after the second consecutive input_to call, sending out unnecessary telnet negotiations. - (interpret.c) When an object is destructed, all references to it are removed from the stack - including references through closures. Before, only direct T_OBJECT references were removed (reported by Gnomi). 04-Jun-2001 (Lars Duening) (3.2.9-dev.294) - (prolang.y) The type on function return is too strict (or the type tracking in general too lacking) unfortunately, so I took it out again. 03-Jun-2001 (Lars Duening) (3.2.9-dev.293) - (prolang.y) A cast of a type to itself is now a no-op as it should be (reported by Gnomi). - (interpret.c, prolang.y) Integers and float can now be compared against each other (suggested by Gnomi). - (gcollect.c) The GC forgot to trace the program->blueprint pointer which was introduced dev.278 (reported by Coogan). 02-Jun-2001 (Lars Duening) - (backend.c) The check_alarm() function also restores the alarm signal handler. - (prolang.y) It is now possible to cast to '(mixed)'. - (prolang.y) Improved the type tracking of '?:': it now accepts operands of different types and evaluates to 'mixed' in that case (reported by Dennis VanHoey) - (prolang.y) Improved the type checking at function return under strict_types. 25-May-2001 (Lars Duening) - (configure) Comments may now appear at the end of the lines in a settings file (thanks, Bertram!). 24-May-2001 (Lars Duening) (3.2.9-dev.292) - (efuns) Efun terminal_colour() can now use closures to translate colour keys (suggested by Elric). - (backend.c) Removing destructed objects in the process-objects loop clobbered the list of programs to replace (reported by Gnomi). 22-May-2001 (Lars Duening) - (backend.c) Efun regreplace() can now take a closure to produce a replacement string (suggested by Fiona) (f-010516-2). 16-May-2001 (Lars Duening) (3.2.9-dev.291) - (backend.c) The call to clean_up() erroreously counted the internal blueprint-program reference to the reference count passed to the clean_up() apply. Additionally, before the call all pending destructed objects are removed (suggested by Elric). - (rxcache.c) The regexp cache forgot to compare the excompat flag, occasionally selecting the wrong regexp (thanks, Fiona). 16-May-2001 (Lars Duening) (3.2.9-dev.290) - (closure.c) Of course I forgot to remove one debugging related conditional, which effectively disabled yesterdays bugfix. 15-May-2001 (Lars Duening) (3.2.9-dev.289) - (sscanf) Clarified the documentation regarding the matching of '%s' (reported by Pulami and Gnomi) - (closure.c) Compiling a simul_efun with extended varargs in a lambda closure created faulty bytecode (reported by Gnomi). 13-May-2001 (Lars Duening) (3.2.9-dev.288) - (prolang.y) Undid the casting change, as it broke the idiom 's = (string)foo->bar();' where bar() may signal a failure by returning integer 0. It's the old 0/nil dichotomy again. - (efuns.c) Efun sscanf(): the format '%s%d' failed to parse 'ab1cd' correctly (reported by Pulami). 09-May-2001 (Lars Duening) (3.2.9-dev.287) - (prolang.y) Casting an unknown or mixed type now always inserts a proper conversion (reported by Tomi). - (heartbeat.c) The result for DID_ST_HBEAT_AVG_PROC was computed as int instead of float (reported by Freaky). - (mud/morgengrauen) Added basic support for the MorgenGrauen 3.2.6 mudlib. 08-May-2001 (Lars Duening) - (prolang.y) The expression 'a & b' failed to track the types if one of the operands was of TYPE_ANY (reported by Gnomi). 06-May-2001 (Lars Duening) - (interpret.c) Added a protection against simple reference loops as created by 'b = 0 || &a; b = 0 || &a;' (reported by Gnomi). 28-Apr-2001 (Lars Duening) (3.2.9-dev.286) - (several files) Added modifications for the Amiga/gcc port (provided by Niels). 27-Apr-2001 (Lars Duening) (3.2.9-dev.285) - (sprintf.c) Duh. A typo caused '%^' to eat the next character from the format string (reported by Robert). - (sprintf.c) A leading '0' used to pad the whole field with 0s. Now, only leading 0s are printed and this setting and the explicite pad string are independent of each other (suggested by Fiona). - (call_out.c) Removing call_outs, or call_outs going stale, messed up the 'number of callouts' statistics (provided by Gnomi). 25-Apr-2001 (Lars Duening) (3.2.9-dev.284) - (prolang.y) Pre-increment or -decrement failed to terminate the parser reduction, causing an internal fatal error when the result was used somewhere (reported by Dafire). 24-Apr-2001 (Lars Duening) (3.2.9-dev.283) - (efuns.c) Efun sgn() used on negative integers still returned +1 - a simple typo (thanks, Fiona!). 23-Apr-2001 (Lars Duening) - (func_spec) Efun copy_mapping() is deprecated. 22-Apr-2001 (Lars Duening) (3.2.9-dev.282) - (interpret.c) When using a multi-dimensional allocate(), specifying 0-sized dimensions could crash the driver (reported by Menaures). - (Makefile.in) When doing 'make install', the driver is now installed as 'driver', not 'parse' (reported by Eivind). - (docs) Some corrections, provided by Fiona. - (closure.c) The generation of sparse lookup ranges failed to properly generate default-jumps for the values between two consecutive ranges (e.g. 'case -2..-1:, case 1..2:' was smudged into 'case -2..1: case 0..2:') (reported by Menaures). 19-Apr-2001 (Lars Duening) (3.2.9-dev.281) - (configure, main.c, simulate, call_out.c) Runtime for the total number of callouts, settable with commandline option '--max-callouts'. 17-Apr-2001 (Lars Duening) (3.2.9-dev.280) - (prolang.y) Redefining an inherited private heart_beat() didn't work: the new heart_beat() wasn't called (reported by Gnomi). - (rxcache.c) The status functions were far too slow, because they counted the number of entries on each call (reported by Sissi). - (prolang.y) Function calls can distinguish between variables and functions of the same name (suggested by Gnomi). 16-Apr-2001 (Lars Duening) (3.2.9-dev.279) - (interpret.c) The call to privilege_violation() for efun call_out_info() passed some random data instead of constant 0 as argument (reported by Freaky). 13-Apr-2001 (Lars Duening) - (simulate.c) Made the 'illegal object to load' error message more informative (suggested by Freaky). 12-Apr-2001 (Lars Duening) - (sprintf.c) The formatter "%^" will print "%^" for compatibility with terminal_colour(). - (closure.c, array.c) Introduced a total order on closure svalues which allows their use in alists and array operations like intersection. The same could be done for their use in mappings, but that change is too big to be squeezed into 3.2.9 (reported by Gnomi). 11-Apr-2001 (Lars Duening) (3.2.9-dev.278) - (prolang.y) Additions of floats to constant 0 yielded TYPE_ANY instead of TYPE_FLOAT. - (interpret.c) Arguments passed to runtime_error() were wrong if the error happened in a master apply (reported by Gnomi). 10-Apr-2001 (Lars Duening) - (exec.h, prolang.y, simulate.c) The program structure received a pointer pointing to the blueprint object. Main purpose is to allow the cheap deep update of the master object, but other uses will probably show up as well. - (simulate.c) Added function deep_destruct() to destruct an object and the blueprints of all inherited programs. - (backend.c, interpret.c) The signal SIGUSR1 now destructs the master object and all its inherits, to allow a clean reload. 07-Apr-2001 (Lars Duening) (3.2.9-dev.277) - (smalloc.c) The multiple-of allocation was not quite correct. 06-Apr-2001 (Lars Duening) (3.2.9-dev.276) - (md5.c) Removed a compiler warning. - (smalloc.c) On Darwin, we get the memory from the system only in multiples of 64 Bytes, which caused the creation of too-small large blocks. Fixed that. 05-Apr-2001 (Lars Duening) (3.2.9-dev.275) - (actions.c) If an action doesn't have a verb defined, the error will list the object which created the action (suggested by Gnomi). 04-Apr-2001 (Lars Duening) (3.2.9-dev.274) - (Makefiles) Corrected. - (efuns, md5) New efun md5_encrypt() (thanks, Gawain!) - (interpret.c) Assigning a volatile strings sometimes crashed due to a faulty length computation (reported by Gnomi and Parsec!) - (lex.c) An attempt to '(: ;} fun() {... :)' is now detected and rejected (reported by Daniel Fischer). 03-Apr-2001 (Lars Duening) (3.2.9-dev.272, -dev.273) - (lex) Corrected a bug in yesterday's change. - (doc/efun) Removed references to deprecated efuns. - (erq, xerq) Fixed (hopefully) a type mismatch under Darwin. - (make_func, lpc parser) Wrapped the parser into wrappers which selectively define 'lint' to suppress spurious warnings, and under Darwin enable compilation in the first place. 02-Apr-2001 (Lars Duening) (3.2.9-dev.271) - (lex) When generating names for inline closures, the compiler makes sure that the name hasn't been used yet (reported by Dan). 30-Mar-2001 (Lars Duening) - (lex.c) The include path and filenames returned by the master the and H_INCLUDE_DIRS closure can now start with leading slashes (suggested by DaFire). - (xerq) The incoming data buffer is cleared before each new command, in order to detect errors more easily (suggested by Gawain). - (comm.c) When using charmode against non-charmode clients (TinyFugue), the line end wasn't handled properly. The solution is to receive the data in charmode, but to pass it on to the user as if in linemode. 27-Mar-2001 (Lars Duening) - (interpret.c) All <op>= assignment operators can now operate on a char lvalue (indexed string) on the left hand side. 20-Mar-2001 (Lars Duening) (3.2.9-dev.270) - (erq, xerq/execute.c) Quoted arguments are recognized and not split into words (provided by Christian W., thanks!) - (backend.c) The no-var-swapping time before a reset() is limited to 5 minutes max, otherwise a lot of muds will never swap variables. - (array.c) A correction to the correction to functionlist() (thanks, Fiona!). - (func_spec, interpret.c) Efun allocate() now accepts an initialisation value (f-981229-09a). - (configure.in) When using mySQL, the script also checks for and links in the libz (reported by Bardioc). 16-Mar-2001 (Lars Duening) (3.2.9-dev.269) - (pkg-mysql.c) Dafire provided to new mySQL efuns: db_insert_id() and db_coldefs() - thanks!. - (array.c) The list control for functionlist() was documented incorrectly, and also functioned incorrectly (thanks, Fiona!). 15-Mar-2001 (Lars Duening) (3.2.9-dev.268) - (settings/unitopia) Updated (thanks, Freaky!). - (comm.c) The pathlength safety check in the start-erq-demon routine was too cautious (thanks, Freaky). - (mudlib/dhrystone.c) Added a missing assignment. - (xerq/Makefile.in) Accidentally left the XDEBUG option enabled. - (comm.c) A prompt closure bound to a destructed object is detected and restored to the default prompt. An error is still thrown, to give the mudlib hackers something to do. - (pkg-mysql.c) The version check did not really work (thanks, Dafire!). 14-Mar-2001 (Lars Duening) (3.2.9-dev.267) - (main.c, comm.c) The command option '--erq' can now be used to specify the full pathname of the erq executable (suggested by Freaky). - (comm.c) Before printing a prompt via a closure, the current object is set to the command giver. This avoids a crash under (admittedly rare) circumstances of closure destruction. 13-Mar-2001 (Lars Duening) (3.2.9-dev.266) - (dumpstat.c, gcollect.c) Added uses of StrRefCount where I forgot them earlier (thanks, Freaky!) - (backend.c) When calling the master apply slow_shut_down(), it is made sure that previous_object(), this_player(), and this_interactive() are all 0 (reported by Freaky). - (comm.c, ed.c, interpret.c, simulate.c) Removed spurious checks for out-of-memory condition. While the intention is noble, it hasn't been implemented consequently enough to be of any use. - (comm.c) If a CRLF as expansion of a NL character was to be inserted, but only one character was left in the buffer, the code errorneously detected a 'message shorter than minimum, needs not be send' condition, thus swallowing all following characters after the NL. - (interpret.c) Argl. Letting the ==/!= comparison for floats fall through to the all new and improved closure comparison was not a good idea (reported by Gnomi and Freaky). 12-Mar-2001 (Lars Duening) (3.2.9-dev.265) - (pkg-mysql.c) db_connect() optionally takes a username and a password (thanks, Andre!) 09-Mar-2001 (Lars Duening) (3.2.9-dev.264) - (mud/lp-245) Small corrections. - (configure) Another define for 'void' signal return types - this one can be tested directly, as a comparison of the values of RETSIGTYPE is not portable. 05-Mar-2001 (Lars Duening) (3.2.9-dev.263) - (smalloc.c) Corrected a statistics error (reported by Freaky). - (simulate, interpret, efuns.c) Added the last error trace as possible return from debug_info(DINFO_TRACE) (suggested by Gnomi). 04-Mar-2001 (Lars Duening) (3.2.9-dev.262) - (interpret.c) Removed a VM stack leak in ++/-- on floats (reported by Dafire). - (object.c) The previous object for object resets is this_object() (reported by Gnomi). - (Makefiles) Added 'lint' to the compiler defines to ensure a smooth compilation of prolang.y. 28-Feb-2001 (Lars Duening) (3.2.9-dev.261) - (backend.c) Simplified the error message output in write_file(). - (simulate.c) The master apply runtime_error() receives an extra argument describing if the error is a normal error, or occured during a heartbeat (suggested by Freaky). - (func_spec) Efuns add_verb() and add_xverb() are now part of the deprecated bunch. - (gcollect.c) Added a dump function for closure literals. 27-Feb-2001 (Lars Duening) - (lex.c, gcollect.c) Cosmetics. - (smalloc.c, main.c) xfree() now accepts NULL pointers. Allocation of MIN_(SMALL_)MALLOCED blocks actually check if the memory is allocated (reported by Freaky). - (main.c, backend.c, interpret.c) Swapping is no longer an optional feature, though it still can be disabled by specifying swapping times of 0. - (configure.in, main.c, xalloc) New commandline options '--min-malloced' and '--min-small-malloced' to complement the configuration options (suggested by Freaky). - (hosts/win32) Updated the various files with those provided by Coogan (thanks!) 26-Feb-2001 (Lars Duening) - (smalloc) Renamed the constant MASK to M_MASK, as MASK seems to clash with one of Darwin's symbols. - (smalloc.h) svalue_strlen() did not use StrRefCount in all the places it should be used (reported by Freaky). 22-Feb-2001 (Lars Duening) (3.2.9-dev.260) - (func_spec, efuns) The efuns debug_info() and object_info() can now return single values instead of whole arrays. - (backend, comm.c) Added a check method which restarts the alarm after some time of activity. This is required at least for Cygwin on Windows, where the alarm often gets lost (thanks, Holger!). - (interpret.c) The master apply valid_trace() now receives as second argument the argument passed to the efun (suggested by Coogan). - (typedefs.h, interpret.c, gcollect.c, stralloc, smalloc.h) Introduced the type StrRefCount for the shared string refcounts. This will make it easier to switch from 16-Bit to 32-Bit refcounts (after 3.2.9 has been released, since it should undergo some more testing) (p-010208). - (lex.c, prolang) Inline closure 'inherit' the visibility modifiers of their defining function (suggested by Coogan). - (smalloc.c) Corrected an accounting error in consolidate_freelists(). 21-Feb-2001 (Lars Duening) (3.2.9-dev.259) - (autoconf/configure.in) Some more small fixes. - (prolang.y) Recognize an inherited heart_beat() only if it is visible (adopted from Morgengrauen). - (efuns.c) debug_info(DINFO_DATA, DID_STATUS) now returns the number of executed heartbeats as well as the total number of heartbeats. - (settings/morgengrauen) New configuration file. 20-Feb-2001 (Lars Duening) (3.2.9-dev.258) - (autoconf/configure.in) Corrected some replacement directives. - (func_spec) Removed two duplicate efun definitions which caused Cygwin compiles to hang. - (make_func.y, lex.c) Added diagnostics to detect duplicate efun and string definitions. - (autoconf/configure.in) The option --enable-use-mysql now also accepts a directory path under which the libmysqlclient can be found (f-000918). 16-Feb-2001 (Lars Duening) (3.2.9-dev.256 and -dev.257) - (main.c) Added code to unblock SIGALRM on program start - on systems it is blocked randomly for no obvious reason (thanks, Fini!) (maybe fixes b-000308) - (efuns, func_spec) New efun sgn() to retrieve the sign of a numeric argument (suggested by Croft). 13-Feb-2001 (Lars Duening) - (lots of files) New configuration option 'use-deprecated' allows to disable obsolete and deprecated efuns. If enabled, the macro __DEPRECATED__ is defined (suggested by Coogan). - (lots of files) The UDP port is now a standard feature. The option has been renamed from 'catch_udp_port' to just 'udp_port'. - (lots of files) The UDP efuns have been renamed from send_imp() and query_imp_port() to send_udp() and query_udp_port(); as were the associated privilege violation string and the master apply. The old names are still available with USE_DEPRECATED. 12-Feb-2001 (Lars Duening) - (interpret.c) Assigning volatile strings was somehow no longer supported under all circumstances (reported by Gnomi). 07-Feb-2001 (Lars Duening) (3.2.9-dev.255) - (smalloc.c) Removed a compilation error (reported by Coogan). 06-Feb-2001 (Lars Duening) - (strfuns.c) When (re)allocating the strbuf buffer, the length was accidentally casted down to unsigned short, causing the program to write outside the allocated memory (reported by Sebastian). 05-Feb-2001 (Lars Duening) - (smalloc.c) Using the time_stamp() from within the allocation routines was really stupid. - (smalloc.c) The large block allocator had an off-by-one error which left free large blocks of 'small' size behind. These blocks could never be allocated, just merged again. - (smalloc.c) The routine to split largish free small blocks was faulty, causing coredumps. - (parse_old.c) Expanded the list of hard coded prepositions. - (lex.c) Added new standard predefine __LDMUD__, expanding to an empty string. - (settings/heaven7, mud/heaven7/) Added basic support for Heaven7: you can boot the mud and log in. There are probably still errors in the areas. 31-Jan-2001 (Lars Duening) - (interpret.c) Attempts to set a single string character to 0 are now caught with a runtime error (reported by Nico). - (func_spec) The result of set_prompt() can be a string or a closure, yet it was declared to return only string (reported by Gnomi). - (prolang.y) The type tracking for multiplication returned an 'int' result even for 'mixed' operands (reported by Gnomi). - (interpret.c) The relations '==' and '!=' can now compare lfun- and variable closures (suggested by Gnomi). - (func_spec, efuns) New efuns min() and max(). 30-Jan-2001 (Lars Duening) - (configuration, smalloc.c) Added a new config parameter MIN_SMALL_MALLOCED which defines the size of the first small chunk allocated by the system. - (xerq) With ERQ_FORK started subprocesses are now as well handled with a child_t structure. While not exactly necessary, it allows for useful diagnostics in the sig_child() handler (reported by Freaky). - (dumpstat, efuns.c) The internal data measuring routines now return both the total data size, as well as the data size scaled down according to the extend of data sharing. The results are also returned by object_info(OI_MEMORY) (suggested by Freaky). 26-Jan-2001 (Lars Duening) (3.2.9-dev.254) - (smalloc.c) Introduced a free list for 'oversized' small blocks which is used much like the unused space in a small chunk. This list allows the consolidation function to keep large free blocks together even if they exceed the legal maximum size for small blocks. 25-Jan-2001 (Lars Duening) - (smalloc, gcollect.c) New function consolidate_freelists() to rebuild the small block free lists, merging adjacent free small blocks where possible. Additionally, when a small chunk is exhausted the small block allocator first tries to split one of the larger free blocks before getting a new small chunk. 25-Jan-2001 (Lars Duening) (3.2.9-dev.253) - (xerq/execute.c) Corrected some small bugs (thanks, Freaky!) 24-Jan-2001 (Lars Duening) - (port.c) The time_stamp routines recompute the string result only if the time has changed since the last call. - (efuns.c, smalloc.c) smalloc now keeps a statistic about the number and size of permanent and clib allocations; and debug_info() returns this statistics as part of (DINFO_DATA, DID_MEMORY). - (smalloc.c) Defining the macro DEBUG_SMALLOC_ALLOCS causes smalloc to log all new allocations which can't be satisfied directly from the free lists. The output is written to the same stream as the GC log (default is stderr). 19-Jan-2001 (Lars Duening) (3.2.9-dev.252) - (main.c) Change to the default MUD_LIB directory early, so that the debug.log file is opened in the right place (reported by Elric). - (xerq) The parent now synchronizes with the child by sending a specific byte of data. - (pkg-mysql.c) Markus Peter provided a patch for backwards compatibility with mySQL before 3.22.24. - (main, comm.c) The name of the erq executable can now be given on the commandline with the option '--erq <filename>' (suggested by Freaky). 18-Jan-2001 (Lars Duening) - (comm.c) In efun exec(), if this_interactive() points to one of the the involved objects, it is switched over as well, like this_player() already is (reported by Elric). - (xerq) Freaky found and corrected a couple race conditions in the clean-up loops of socket and children. - (backend.c) Efun debug_message() now allows to preprend the message with a timestamp for the current date and time, controlled by a new bitflag (suggested by Freaky). 17-Jan-2001 (Lars Duening) - (xerq) Added additional diagnostic output, activated with compiler option XDEBUG. Also, removed two uninitialized variables (the stdout/ stderr sockets in the child_t structure) in the handling of child processes, and added measures against buffer overflows. 16-Jan-2001 (Lars Duening) - (erq.c, xerq/socket.c) When closing the connection to a child process, the erq makes sure to read all still remaining data from the connection (reported by Freaky). 21-Dec-2000 (Lars Duening) (3.2.9-dev.251) - (comm.c) If the driver received only a NL in charmode, the first following character was ignored. 21-Dec-2000 (Gnomi) - (comm.c) Found and corrected a problem concerning the use of CHARMODE with Windows' telnet: the charmode negotiation were considere a failure too early, and moreover the CHARMODE_REQ flag was reset without adjusting the input buffer variables. Unfortunately this re-opens b-000818. (b-001214) - (comm.c) Clients sending CRLF in Charmode instead of just CR are now handled properly: the LF is ignored like the NUL in CR NUL already is. This also re-fixes b-000818. - (comm.c) When receiving DONT ECHO or WONT SGA, the check for if the option has already been granted was wrong. - (comm.c) After a charmode input_to() has been received, the driver did not properly reset the text pointers, causing !-escaped commands to go unrecognized. 20-Dec-2000 (Lars Duening) (3.2.9-dev.250) - (comm.c, interpret.c) More small corrections to the new telnet error reporting. - (func_spec, backend.c) Extended the efun debug_message() to take a parameter telling where to log the message (suggested by Freaky). 19-Dec-2000 (Lars Duening) (3.2.9-dev.249) - (mudlib/floattest.c) Unfortunately this code was a derivate of a copyrighted program from Numerical Recipes, and therefore can't be distributed legally. - (lex.c) On solaris, strtod() does not clear errno after a successful conversion, causion spurious warnings (reported by Coogan). - (main.c) The driver version is now also printed in the .debug.log file on startup. - (comm.c) A bug in the comm_fatal() routine caused the driver to abort on the second call to it. Oops. (reported by Gnomi and Freaky) 15-Dec-2000 (Lars Duening) (3.2.9-dev.248) - (comm, simulate.c) When the telnet machine for a player gets into an inconsistent state, the driver no longer aborts. Instead, the player's interactive structure is dumped to stderr, and then removed the hard way. - (regexp.c) Added the metacharacter '+' (match at least once' to the regexp code (provided by Terry Penn, thanks!). - (mudlib/floattest) This LPC program computes the floating point characteristics; it can be called from the test_master (provided by Croft, thanks!) 14-Dec-2000 (Lars Duening) (3.2.9-dev.247) - (lex.c) Octal numbers are now prefixed by '0o' - it's cleaner and avoids compatibility problems with existing code. 12-Dec-2000 (Lars Duening) (3.2.9-dev.246) - (interpret.c) Omitting a check_map_for_destr() in a mapping addition caused data in a wiz_info mapping to mysteriously disappear (thanks, Coogan) (b-001204). 10-Dec-2000 (Lars Duening) (3.2.9-dev.245) - (prolang.y) The result type for '++map[mixed]' was computed incorrectly (reported by Freaky). - (comm.c) If an error occurs on accept(), the driver no longer abort()s but prints a better error message. 06-Dec-2000 (Lars Duening) (3.2.9-dev.244) - (lots of files) Closures bound to destructed objects are now treated like destructed objects and zeroed out where possible. - (wiz_list) Destructed objects and closures is now removed from the 'extra' wizinfo in regular intervals. Note that this doesn't extend to subarrays and -mappings held by the 'extra' info. 06-Dec-2000 (Lars Duening) (3.2.9-dev.243) - (autoconf/configure.in, configure) If MySQL or IPv6 are not requested, the configure script won't check for the libraries; consequently they won't linked in unless needed (suggested by Coogan). - (interpret.c, prolang.y) The operators '++' and '--' can now be used on floats, too (suggested by Croft). 04-Dec-2000 (Lars Duening) - (interpret.c) sqrt(0.0) caused a 'bad argument' error - corrected that (thanks, Andre!). 30-Nov-2000 (Lars Duening) (3.2.9-dev.242) - (interpret, efuns.c) debug_info(DINFO_TRACE) returns the current call stack as either string or array. (f-000716) 29-Nov-2000 (Lars Duening) - (interpret.c) The efuns sin(), cos(), tan(), atan(), atan2(), log(), exp(), sqrt() and pow(), when given an integer argument, returned a float result typed as integer. (reported by Slava) 28-Nov-2000 (Lars Duening) (3.2.9-dev.241) - (lex.c) Oops, a last minute change to remove a warning broke the recognition of escaped characters in strings (thanks, Slava!). 27-Nov-2000 (Lars Duening) - (interpret.c) Efun pow() did not properly check its arguments. 27-Nov-2000 (Lars Duening) (3.2.9-dev.240) - (lex.c) Numbers can be specified in binary using the '0b' prefix. - (lex.c) Moved the parsing of escaped character sequences into a separate function, and implemented the specification of characters through their numeric code (expressed in decimal, octal, sedecimal and binary). (f-990203-04) 26-Nov-2000 (Lars Duening) (3.2.9-dev.239) - (closure.c) Implemented catch-nolog as '({#'catch, <body>, 'nolog })'. 21-Nov-2000 (Lars Duening) - (prolang) Folded catch() and catch_nolog() back into one catch(), distinguished by the keyword 'nolog'. 18-Nov-2000 (Lars Duening) - (func_spec, interpret.c) The efuns sin(), cos(), tan(), atan(), atan2(), log(), exp(), sqrt() and pow() now accept integers as arguments, too. The results remain float. (suggested by Croft). - (interpret.c, mapping.c) When testing for a 'mapping too large' condition on mapping addition or element assignment, it is made sure that all invalid keys have been removed. (reported by Fiona) - (func_spec, lex.c, interpret.c, simulate, prolang.y) New language construct catch_nolog() as alternative to catch(): the new statement doesn't log the error in the driver logfiles (thanks, Michael!) (p-990901). - (lots of files) The mode of the driver (plain or compat) can now be selected with the commandline options '--no-compat' resp. '--compat'. The default is still selected during configuration. (f-981229-00) 17-Nov-2000 (Lars Duening) - (mudlib/sys, doc/) Small corrections (thanks, Coogan!). 15-Nov-2000 (Lars Duening) - (mudlib/telnet_neg.txt) Added Slava's example on telnet negotiation, which includes a means of measuring lag using TMs (thanks!). 15-Nov-2000 (Lars Duening) (3.2.9-dev.238) - (efuns.c) The OIB_NAME returned by object_info(OINFO_BASIC) did not start with a '/' in plain mode. (reported by Fiona) - (mud/lp-245) Added secure/master.c as a wrapper for obj/master.c for use with plain drivers. - (func_spec, interpret.c, port) Added new efun utime() (provided by Slava) and adapted ctime() to accepts its output. - (settings/unitopia) Updated with new version provided by Freaky. - (settings/default) Clean up. 14-Nov-2000 (Lars Duening) - (strfuns, actions.c, ed.c, efuns.c, interpret.c, swap.c, wiz_list.c) New function xstrncpy() to copy a string up to a certain length, but without adding extra 0 padding if the string is shorter than the given length. 07-Nov-2000 (Lars Duening) (3.2.9-dev.237) - (efuns.c) object_info(OINFO_MEMORY) also returns the program flags NO_SHADOW, NO_CLONE and NO_INHERIT. - (pkg-mysql.c) Small bugfix (thanks, Michael!) - (settings/forgottendreams) Updated according to Michael. - (util/[x]erq/erq.h) Around dev.220 I accidentally broke compatibility with older erq binaries by renumbering the KEEP_HANDLE constant - fixed that now. (reported by Coogan) 05-Nov-2000 (Lars Duening) - (interpret.c) call_resolved() now accepts a destructed object as second argument and returns 0 instead of throwing an error. (reported by Coogan) - (settings/tubmud) 'nosave' is now on by default. 03-Nov-2000 (Lars Duening) (3.2.9-dev.236) - (interpret.c) When setting up the frame for a function, resolve a possible cross-definition first. (b-000616, b-001004) - (interpret.c, prolang.y) Right shifts of 32 or more are now properly defined. 02-Nov-2000 (Lars Duening) (3.2.9-dev.235) - (mapping.c) Adding two mappings of different width throws an error (reported by Holger and Fiona). (b-001010) 01-Nov-2000 (Lars Duening) - (regexp.c) Added a patch by Marcus Meissner to make regexp more compatible with 8-bit characters (thanks!). 23-Oct-2000 (Lars Duening) - (simulate, interpret) Moved the catch_instruction() function into simulate to keep aggressive optimizers from over-optimizing this function. I hope this finally cures the catch() related crashes. (thanks, Coogan!) (b-001010-1) 16-Oct-2000 (Lars Duening) - (simulate, array.c, call_out.c, comm.c, mapping.c) Callbacks now accept protected lvalues as arguments only if they are called just once. Reason is that the protectors don't count references: at the first return from a callback the lvalue would be freed even though it is still needed. (b-001002) 12-Oct-2000 (Lars Duening) - (main.c) New driver option '--pidfile <filename>' to write the pid of the driver into a file (suggested by Freaky). 10-Oct-2000 (Lars Duening) - (backend.c) The IGNORE_BANG option to efun input_to() didn't work (provided by Casey, thanks!). - (pkg-mysql.c) Changed some xalloc()s into pxalloc()s to protect them against the GC (provided by Michael Fedrowitz, thanks!). - (interpret.c) Cosmetics to the traceback output (thanks, Coogan). - (simulate.c) A semicolon was missing (thanks, Christian B.). 13-Sep-2000 (Lars Duening) (3.2.9-dev.234) - (func_spec, lex.c, closure.c, interpret.c, prolang.y) Added the operators '>>>' and its cousin '>>>=' for logical right shifts. 11-Sep-2000 (Lars Duening) - (configure) Removed no longer used option 'max-log-size' (reported by Malcom). 09-Sep-2000 (Lars Duening) (3.2.9-dev.233) - (simulate.c) Callbacks in efuns like filter() can now take references, too (reported by Freaky). 07-Sep-2000 (Lars Duening) (3.2.9-dev.232) - (simulate, call_out) Due to a flag inversion, call_out() was unable to take lvalues as arguments. It also gave the error message with an off argument count (reported by Thorsten). 06-Sep-2000 (Lars Duening) - (interpret) The setjmp() for a catch() instruction has been moved into its own stack frame and eval_instruction() is called recursively on the guarded code. This seems to prevent failures and cores when using gcc as compiler, and doesn't hurt on other platforms either (reported by Parsec). 05-Sep-2000 (Lars Duening) - (xerq/socket.c) A faulty comparison caused the xerq to ignore SOCKET_STDERR and send everything on stdout (thanks, Fiona!). 30-Aug-2000 (Lars Duening) (3.2.9-dev.231) - (comm.c) The driver did not fully recover from a failed INPUT_CHARMODE notification - treating the input as in charmode even after the failure. (b-000818) 29-Aug-2000 (Lars Duening) - (interpret.c) When reactivating a destructed master, the function applies failed to find static functions, even though all other master applies do (thanks, Daniel!) 28-Aug-2000 (Lars Duening) (3.2.9-dev.230) - (pkg-mysql.c, configure, Makefile.in) Made the mySQL code work again. Additionally the mysqlclient library is now part of the @PKGLIBS@ template in Makefile.in, so that utils like xerq don't have to link it in. (thanks, Michael!) 25-Aug-2000 (Lars Duening) (3.2.9-dev.229) - (xerq/lookup.c) In the IPv6 lookup, a buffer was freed twice. (found by Uwe and Fiona). - (pkg-mysql, doc/..., INSTALL) Integrated the mySQL efuns into the driver proper. They are activated on suitable hosts with the configuration switch 'use-mysql' and, when activated, define the macro '__MYSQL__' in LPC programs. - (lex.c) Added new LPC defines __VERSION_MAJOR__, ..._MINOR__, ..._MICRO__ and ..._PATCH__ do exactly what you expect. 24-Aug-2000 (Lars Duening) (3.2.9-dev.228) - (INSTALL) Added compilation notes provided by Bruce (thanks!) - (xerq/erq.c) The DEBUG part was missing the data to generate proper error messages for RLOOKUPV6 requests (reported by Fiona). - (comm.c) Removed two warnings. 23-Aug-2000 (Lars Duening) (3.2.9-dev.227) - (util/indent/, autoconf/configure.in) The Makefile is now under configure control, and I removed a couple of warnings. - (util/erq/erq.c) Oops, the check for HAVE_SYS_TIME came before machine.h was included, so it couldn't work (reported by Betram and Fiona). 22-Aug-2000 (Lars Duening) (3.2.9-dev.226) - (hash, stralloc) Changed a few parameters from 'char*' to 'const char*'. - (comm.c, erq, xerq) The IPv6 lookup was totally broken - fixed it. (reported by Holger) 20-Aug-2000 (Lars Duening) (3.2.8-dev.225 == 3.2.8) - RELEASE 3.2.8 - (func_spec, prolang.y) Efun clonep() now accepts 'mixed' arguments. (suggested by Fiona) - (interpret.c) The salt-handling of efun crypt() is now compatible with the md5 passwords of FreeBSDs libscrypt. (thanks, Gerhard!) 15-Aug-2000 (Lars Duening) (3.2.8-dev.224) - (prolang.y) All ranges '[..<op><expr>]' caused the compiler to overwrite memory (most of the time the compiled program) because the attributes of <expr> weren't promoted up by the <ob>. - (lex.c) '//' comments in the argument list of a function macro at the point of use were not ignored and caused syntax errors. It even was already a 'TODO' in the lexer... (reported by Croft and Parsec). 14-Aug-2000 (Lars Duening) (3.2.8-dev.223) - (prolang.y) The range '[..-<expr>]' caused the compiler to overwrite memory (most of the time the compiled program) because the attributes of <expr> weren't promoted up by the '-' (reported by Mammi). - (prolang.y) In the expression '++<ident>' the opcode for '++' was overwritten if the <ident> required a 2-Byte index. (b-000803, b-000810) - (comm.c) Looking up hostname first enters the IP address as string into the lookup table, to be replaced later by the ERQ. This neatly handles unresolved hostnames as well (provided by LynX). 11-Aug-2000 (Lars Duening) (3.2.8-dev.222) - (mstrings) Completed the implementation of the string module itself. 10-Aug-2000 (Lars Duening) - (func_spec, interpret.c, efuns.c) Efun terminal_colour() can now be used with no keyword mapping at all: it then acts as a simple wrapping efun. 08-Aug-2000 (Lars Duening) - (smalloc.c) Attempts to free a memory block twice are now explicitely flagged (when compiling with MALLOC_TRACE). - (lex.c) #else and #endif followed by uncommented text raise an error only in pedantic mode; otherwise just a warning is generated. - (array.c, regexp.c, backend.c) Made the handling of a regexp's memory more resistant to errors. Especially an error during a match or the first pass of a regexp compilation caused the memory for the last regexp compiled being freed unconditionally, even when that last regexp was still in use or had been freed already. 04-Aug-2000 (Lars Duening) - (lex.c) After an include, the driver did not refill its lex buffer. This had bad or even catastrophic results if the include failed and the include statement was the last in the lex buffer. (reported by Freaky) - (lex.c) The preprocessor statements #else, #endif and #define now check if they are followed by uncommented text. If yes, the former two raise an error (like real C does), if no, the latter one raises an error. (reported by Freaky) 01-Aug-2000 (Lars Duening) (3.2.8-dev.221) - (hosts/be/Makefile, .../beyacc) The driver can now be crosscompiled on x86 for PPC. - (mud/lp-245) Corrected a few simul_efun definitions. - (lex.c) '#else' was not recognized if followed by whitespace. (reported by Bardioc and Dafire) 30-Jul-2000 (Lars Duening) (3.2.8-dev.220) - (prolang.y) Redefinition of an inherited function with differing arguments was not properly differentiated from a normal redefinition, causing the redefined argument information to be lost even with #pragma save_types in effect. (reported by Bardioc) - (mstrings) First code for an improved string handler. 26-Jul-2000 (Lars Duening) (3.2.8-dev.219) - (lex.c) Preprocessor keywords weren't recognized if they were followed by a non-space character (e.g. '# if('). (reported by Fiona). - (object.c) Saving a mapping to a file wrote unsaveable keys like objects as '0' into the savefile, possibly creating multiple entries with the same key. As restore_object() didn't expect this, svalue were simply overwritten instead of properly freed. This lead to memory leaks. Now, non-saveable keys and their data are completely excluded from a saved mapping, and the restore function is more robust. (part of: b-990429, b-991123-1, b-000104) 25-Jul-2000 (Lars Duening) - (interpret.c, smalloc.c) Removed two warnings (thanks, Freaky!). 23-Jul-2000 (Lars Duening) (3.2.8-dev.218) - (main.c) The removal of MAX_SMALL_MALLOCED somehow didn't make it into the diff. 20-Jul-2000 (Lars Duening) (3.2.8-dev.217) - (regexp.c) Errors in the regular expression leaked the memory from temporary buffers (found by Freaky). 18-Jul-2000 (Lars Duening) - (smalloc.c, xalloc, main.c, settings/*, configure.in) When the system run out of memory in a small block allocation during a GC, it tended to crash because the interaction of the 'force_more' allocation and the 'max_small_malloced' parameter caused the allocator to free the reserves, but to never check the freelists. Removing the max_small_malloced parameter and slightly simplifying the force_more logic solved the problem. As a side effect, the commandline argument '--max-small-malloc' and the corresponding configuration parameter are gone. 17-Jul-2000 (Lars Duening) - (closure.c, gcollect.c, dumpstat.c, interpret.c, object.c) Revisited the places where a NULL object->variables might be used. Especially replace_program was suspectible to this. - (interpret.c) Efun to_string() now accepts lambda closures as well. sprintf() was modified to print the same information as to_string() returns. (suggested by Fiona) 06-Jul-2000 (Lars Duening) (3.2.8-dev.216) - (smalloc.c) Removed two warnings (thanks, Freaky!) - (lex.c) Improved the display of the error context at line ends or end of files. - (configure, port, machine.h.in) Ultrix (VAX) doesn't provide the strdup() function, so I added a check and a replacement function. Ironically, make_func.y already has its own strdup() for exactly this reason... (reported by Bruce). 04-Jul-2000 (Lars Duening) (3.2.8-dev.215) - (smalloc.c) Oops, the file didn't compile. 03-Jul-2000 (Lars Duening) (3.2.8-dev.214) - (comm.c) A DEBUG_TELNET statement was at the wrong place, dereferencing an uninitialized variable (b-000620). - (object.c) Forgot to delete the old ixopen() statement when making the change in 3.2.8-dev.210, causing the driver to run out of fds because every savefile was opened twice. (b-000630) - (smalloc.c) Added more magic words to accomodate the increased SMALL_BLOCK_MAX value. In addition, the magic word arrays are indexed modulo their size in case SMALL_BLOCK_MAX is increased over the size of the arrays again. - (interpret.c) Made the VM Stack error messages more informative. - (prolang.y) Provided by UNItopia: return type mismatches are by default warnings, and in pedantic mode real errors. - (Makefile.in) The generated Makefile ended the OBJ definition on sane systems with a line ending in a backslash, which the Ultrix make doesn't like. (thanks, Bruce!) 01-Jul-2000 (Lars Duening) (3.2.8-dev.213) - (main.c) DEBUG options are printed with --options, too. - (object.c) If a savefile can't be opened, the strerror() message is included into the error message. - (backend.c) If the driver runs out of fds in efun write_file(), lex_close() is called only when there is a compilation going on (before, it was called on every out-of-fds, which crashed the driver if it happend outside of a compile). (reported by Freaky) - (comm.c) h_telnet_neg() always removed two values from the stack, regardless of many actually were there. (b-000625) - (interpret.c) The handling of a caught error in catch() has been exported in a separate, non-inlineable function to thwart overzealous optimizers. (b-000606) - (doc/LPC/closure_guide) Clarified the creation of mappings using #'([. (b-000217-1) - (lex.c) lex_error_context() pays more attention to the end of the line buffer and EOF characters - before it blindly copied 14 characters after *outp. (b-000518) 29-Jun-2000 (Lars Duening) (3.2.8-dev.212) - (settings/unitopia, doc/efun/count_bits) Small changes. - (backend.c) The 'inconsistency in main loop' message is now printed onto stdout, too. - (main, backend.c) The state check (available when compiled with DEBUG) can now be controlled via the commandline option '--check-state <lvl>'. 28-Jun-2000 (Lars Duening) (3.2.8-dev.211) - (prolang.y) Removed the shift/reduce conflict in the rules for argument definitions: token L_VOID was ambiguous. - (prolang.y) The argument type check stumbled over additional arguments in a function redefinition. (reported by Bardioc) - (interpret.c) Operator += allows to add strings and floats to ints, as operator + already does. (reported by Croft) 27-Jun-2000 (Lars Duening) (3.2.8-dev.210) - (closure.c) The lambda block ({ #'return }) was compiled incorrectly: a F_RETURN0 was generated, but for the closure compiler the block returned void, necessiting the insertion of a default value by the surrounding block. The result was a 'Bad stack at F_RETURN' at the final F_RETURN[0] of the lambda closure latest, e.g. for funcall(lambda(0, ({ #'?, 1, 1, ({ #'return }) }) )). (reported by Mat@Maethos) - (object.c) Under Cygwin and other MSDOS filesystem systems, the savefiles are opened in binary mode. (thanks, Bastian!) (b-000528) 16-Jun-2000 (Lars Duening) - (pkg/mysql) Integrated the update to dev.207 (thanks, Malcom!) 14-Jun-2000 (Lars Duening) (3.2.8-dev.209) - (settings/hyperborea) added. - (comm.c) The default routine for answering telnet negotiation requests forgot to clean up the LPC stack when no hook was defined, thus eventually overwriting memory. Since this happened only with clients which actually did negotiate options, this bug remained undetected for quite some time. (b-000508) - (backend.c) Since b-000508 would have been found earlier if the LPC stackpointer had been checked for consistency, I reactivated the check_state() mechanism in DEBUG mode. - (lex.c, prolang.y, interpret.c) Corrected the linecounting in combination with inline closures (found by Zora). 12-Jun-2000 (Lars Duening) (3.2.8-dev.208) - (efuns.c) present(object) returned the environment of the object found, not the object itself. While traditional, this behaviour is rather illogical. (thanks, Freaky!) - (several files) Clarified some casts. 04-Jun-2000 (Lars Duening) (3.2.8-dev.207) - (mapping.c) Added a fatal() if the mapping size rises over 0x10000000. This is to track down b-000218. 02-Jun-2000 (Lars Duening) - (comm.c) Added a number of debug prints to debug the telnet machine. The are inactive unless DEBUG_TELNET is defined. 30-May-2000 (Lars Duening) - (doc/efun/send_imp) Added a note regarding the handling of errors on some machines. (thanks, lynX!) - (interpret.c) The driver now prints the last 200 instructions when fatal()ing in DEBUG mode. In addition, the trace also lists the absolute stack usage together with the already known relative usage. 29-May-2000 (Lars Duening) (3.2.8-dev.206) - (stralloc) Small improvements to the CHECK_STRING mechanism. Added several diagnostics regarding possible refcount over/underflows. - (stralloc) New define 'KEEP_STRINGS' prohibits shared strings from being deallocated. While memory intensive, this is a debugging measure to locate the 'free string not found' error. 28-May-2000 (Lars Duening) (3.2.8-dev.205) - (configure) Remade with autoconf 2.13 28-May-2000 (Lars Duening) (3.2.8-dev.204) - (prolang.y) Warnings are generated if the redefinition of a function loses the 'varargs' attribute (pedantic only), or if the redefinition makes an incompatible change to the return type. Assignment type error messages are more informative. And the argument type warnings/errors are now printed for the correct source line, not the one following the function. - (ed.c) New command 'M' to strip ^Ms from lineends (thanks, Dafire!) - (hosts/be/Makefile) The cross-compiled executable is provided with proper resources and file attributes (icons!). - (doc/LPC/if) Added. 27-May-2000 (Lars Duening) - (prolang.y) Misplaced prototypes are now a warning, not just a debug message. 23-May-2000 (Lars Duening) (3.2.8-dev.203) - (comm.c, util/xerq/erq.c) Corrected a type error which caused the erq to misinterpret message lengths, resulting in overwriting its global variables: Due to a missing sign, the driver did not handle a blocking write() operation properly, causing the resend of the pending data to start off at the wrong offset. In turn, the xerq did not protect itself against invalid message lengths. This bug only triggered when the mud tried to sent more data to the erq at once than the connection could handle. (Thanks, Elric!) (b-000404) 22-May-2000 (Lars Duening) - (various files) Cleaned up a couple of documentary files. 22-May-2000 (Lars Duening) (3.2.8-dev.202) - (prolang.y) Whenever a function is redefined, the argument types are checked for consistency. If they differ, a warning (error in pedantic mode) is issued. The same mechanism compares the prototype for a function with its definition. 21-May-2000 (Lars Duening) - (smalloc.c) Added a diagnostic to the 'Temporary out of mem' situation, because the driver run totally out of mem before the usual diagnostic in the allocating code is reached. 20-May-2000 (Lars Duening) - (interpret.c) foreach() makes sure that destructed objects are removed from arrays. Also, assign_svalue_no_free() assigns 0 to the destination if the source is a destructed object. 18-May-2000 (Lars Duening) (3.2.8-dev.201) - (lex, prolang.y) Pragma 'pedantic' causes the compiler to treat certain warnings as errors. Pragma 'sloppy' (the default setting) turns the pedantic behaviour off. - (prolang.y, simulate) parse_error(), and with it the master lfun log_error() received a flag argument 'warn', which is set if the passed message is a warning, not an error. 16-May-2000 (Lars Duening) (3.2.8-dev.200) - (prolang.y) Added functions yywarn() and yywarnf() to print warnings during the compile. They act like yyerror() except that they don't count up parse errors. - (prolang.y) Failure to pass enough arguments to a non-varargs simul_efun generates a warning. I'd like to make it an error, but then a lot of mudlibs would be unhappy. 14-May-2000 (Lars Duening) (3.2.8-dev.199) - (ed.c) Removed an endless loop in ed_buffer_size(). 08-May-2000 (Lars Duening) - (func_spec) Efun apply() was still listed with two arguments where only one is actually required. - (comm.c) Removed a warning. 07-May-2000 (Lars Duening) (3.2.8-dev.198) - (util/xerq/socket.c) In flush_queue(), the loop pointers weren't advanced after removing a queue element. - (doc/efun/symbol_variable) Clarified the functionality. - (interpret.c) Efun apply(), like efun funcall(), just returns the first argument if it is not a closure. - (func_spec) Efun typeof() can now accepts literal references. - (array.c) Efun transpose_array() handles mixed size subarrays better. - (lex.c, interpret.c, simulate.c, object.c, prolang.y) More tweaks to the memory statistic. - (stralloc.c) Corrected the memory statistics. 07-May-2000 (Lars Duening) (3.2.8-dev.197) - (smalloc.c, comm.c, simulate.c) Added a few more memory statistics. Some memory uses are still unaccounted for, including the memory used for mallocated strings. 05-May-2000 (Lars Duening) - (smalloc.c, efuns.c) Corrected the smalloc memory stat to count and list the wasted memory fragments. Adapted the efun debug_info() to return this information. Also safeguarded the dump_malloc_data() function against allocations during this function (this lead to confusing data). - (lex, simulate.c) Added an output of the lexer structures to 'status'. 30-Apr-2000 (Lars Duening) (3.2.8-dev.196) - (actions.c) Corrected the cleanup code in execute_command() to handle the situation if the command giver is destructed during the command. (b-000207-3) - (lex.c) By Matthew Julius: added the preprocessor macros __DIR__ and __PATH__(). (f-990203-08) - (main.c) Corrected a few typos in the help texts. - (prolang.y) Corrected the type tracking when compiling the '+' operator. - (prolang.y) '(void)' is now accepted as empty function argument list. 27-Apr-2000 (Lars Duening) - (hosts/be) Adapted the BeOS port to R5. 21-Apr-2000 (Lars Duening) (3.2.8-dev.195) - (prolang.y) Added the change required by bison. 20-Apr-2000 (Lars Duening) (3.2.9-dev.194) - (prolang.y) Local variables can now be initialized in the definition. - (random.c) Another small tweak. 19-Apr-2000 (Lars Duening) (3.2.8-dev.193) - (configure.in, configure) Oops #2, the '--with-malloc' was applied wrongly to the config.h - (random.c) Improved the handling of large ranges. - (gcollect.c) Multiple references to a memblock no longer fatal() the driver, but instead print a diagnostic. 18-Apr-2000 (Lars Duening) (3.2.8-dev.192) - (configure.in, configure) Oops, the '--with-malloc' was not really applied to the config.h - (util/xerq) Added a missing variable initialisation, and corrected the num_fd computation for select(). (thanks, Michael!) - (settings/forgottendreams) The name of the settings file should be used by the file itself, hm? 17-Apr-2000 (Lars Duening) (3.2.8-dev.191) - (random.c) Improved the fitting of the raw random number into the given range. The key is not to use the '%' operator, as the low-order bits are distressingly non-random. (b-990217-3) - (settings/unitopia) Updated. - (settings/forgottendreams) Added. - (smalloc.c) Increased the max size of a small block to 128 Bytes. This value used to crash Unitopia, but probably only because the tables weren't fully initialised. This has been corrected, and with Unitopia's new system the crash vanished anyway. - (array.c, backend.c, closure.c, comm.c, efuns.c, prolang.y, object.c, ptrtable.c, simulate.c stralloc.c, strfuns.c, sprintf.c, xalloc.h) Removed a number of warnings. (thanks, Freaky!) - (util/(x)erq/erq.c) Correct the time_stamp() function. - (util/xerq/socket.c) Corrected a type which clobbered the TCP functionality. (thanks, Michael!) - (pkg/mysql/sql.diff) Updated. (thanks, Michael!) - (object.c) Got rid of the fragile 'old_format' variable; instead format parse decisions use the read version value directly. - (interpret.c) <mapping1> -= <mapping2> now works for <mapping2> of any width, like the normal subtraction already does. 16-Apr-2000 (Lars Duening) (3.2.8-dev.190) - (lotsa files) Improved to 'out of memory' error messages to state how much memory was needed for what. 16-Apr-2000 (Lars Duening) (3.2.8-dev.189) - (array.ch, interpret.c) Array sizes are listed verbatim in the array structure. - (smalloc.c) Small improvements to the aligned allocations. 15-Apr-2000 (Lars Duening) (3.2.8-dev.188) - (doc/driver/codestyle) First version. - (xalloc.ch) New 'standardised' memory allocator module: it simply includes the source for whatever allocator was selected. As a side effect, the makefile no longer needs to know which allocator to use - the define moved into config.h . - (smalloc.ch, driver.h) Simplified the compilation of smalloc: the functions are now compiled as they are, and stubs are provided for the emulation of the clib functions if possible. - (other files) Adapted to the new memory allocator compilation, reducing the number of dependencies on MALLOC_smalloc. Especially: the GC support is now tied to the macro GC_SUPPORT, SMALLOC_(LPC_)TRACE has been renamed to MALLOC_(LPC_)TRACE, and array always store their size with the vector structure. - (configuration) The selection of the memory allocator can now be done with the '--with-malloc=<value>' option of configure. 15-Apr-2000 (Lars Duening) (3.2.8-dev.187) - (prolang.y, exec.h) Extended the bugfix from yesterday so that inherited functions are lookup up using a proper breadth search. 14-Apr-2000 (Lars Duening) (3.2.8-dev.186) - (prolang.y, exec.h) The compiler now keeps track which inherits happened at the top level. When looking for a ::function() to call, it searches until it either finds the function defined in a first level inherit, or it takes the first definition found on any other level. (b-000207-4, b-000411-1) - (lex.c) Added new pragma 'weak_types' to counter the effects of a pragma 'strict_types'. (b-000208-1) 13-Apr-2000 (Lars Duening) (3.2.8-dev.185) - (actions.c) If the notify_fail() is a closure, it is made sure that it won't be freed while it is executing (thanks, Gnomi!) - (erq, xerq) The time_stamp() tried to use the driver's get_current_time() instead of the proper time(). - (configure.in) Improved the recognition of unsupported compiler options. 11-Apr-2000 (Lars Duening) (3.2.8-dev.184) - (comm.c) Corrected an endless loop in lookup_ip_entry() (thanks, Freaky!) - (hosts/unix.h) When compiling with gcc on Amiga, the symbol AMIGA is made undefined. 07-Apr-2000 (Lars Duening) (3.2.8-dev.183) - (gcollect.c) Found another void-function casted to a value, this time it's writed(). - (smalloc.c) Increasing the SMALL_BLOCK_SIZE to 32 caused crashes in UNItopia, manifesting themselves in garbage stack data. Reduced it back to 8. - (settings/eotl) Updated. - (object.c) Corrected the versoin detection in restore_object() - the logic was reversed, but happened to work for the current generation of savefiles. (Thanks, Casey!) 06-Apr-2000 (Lars Duening) (3.2.8-dev.182) - (gcollect.c) Casting a the void returning function fatal() to a value returning function caused the AIX-xlc code generator to crash on high optimization levels. - (lots of files) Adapted for compilation on AIX 3.4 using xlc. Some of these changes may break the adaption previously done by Uwe, but now at least I have a base to work from on. 06-Apr-2000 (Lars Duening) (3.2.8-dev.181) - (lex.c, prolang.y, main.c, settings/*, configure) The keyword 'nosave' is now optional, controlled by the setting parameter 'enable-lpc-nosave'. 05-Apr-2000 (Lars Duening) - (closure.c) Forgot to remove two leftover DEBUG printfs (thanks, Coogan!). - (object.c) Corrected one error message. (thanks, Casey!) - (comm.c) Added Lynx' optimization for the iptable[] IP address lookup. Main advantage is that it scales better to large numbers of users. (thanks!) 04-Apr-2000 (Lars Duening) (3.2.8-dev.180) - (pkg/mysql) Coogan provided the updated source files - thanks! - (INSTALL) BeOS: Added a paragraph about the etc/hosts file. - (settings/eotl) Added the settings for End of the Line (thanks, Casey!). - (util/erq, util/xerq) Added timestamps plus nametag to the diagnostics of the erq demons. 03-Apr-2000 (Lars Duening) (3.2.8-dev.179) - (array.c) Freaky was correct: sort_array() no longer made a copy of the incoming array *Mateese blushes deeply*. (b-000106-3) - (sprintf.c) Trailing pad space is not added if a newline follows directly afterwards. 02-Apr-2000 (Lars Duening) (3.2.8-dev.178) - (stralloc.c) Increased the hashed length from 20 chars to 100. - (smalloc.c) Increased the number of small block sizes from 8 to 32, that is: the max small block size from 32 to 128 Bytes. - (*.h) Removed the leading '_' from the include guards as these are reserved for system defines. - (interpret.c, prolang.y, swap.c) Removed pointer-mismatches (for example char* vs. unsigned char*). - (port.c) Implemented time_stamp() using strftime(). 30-Mar-2000 (Lars Duening) (3.2.8-dev.177) - (func_spec, prolang.y, lex.c, interpret.c, closure.c, sprintf.c) Two new bytecodes F_NX_RANGE/F_RX_RANGE (and their lvalue counterparts) implement the constructs [x..] and [<x..]. The former solution of abusing other bytecodes when these constructs were used in a closure was too unstable. (b-000106-1) 29-Mar-2000 (Lars Duening) - (sprintf.c) Change the implementation so that the efun (s)printf() are now re-entrant. 27-Mar-2000 (Lars Duening) (3.2.8-dev.176) - (prolang.y) If the declaration of a new local variable in an lvalue context (foreach()) shadowed a global identifier, the lvalue was computed for the global identifier. And that meant random memory accesses! (b-000221-3, b-000105-2) - (simulate.c) execute_callback(): if an apply() fails, the apply_return_value is now properly set to 0. (b-000113-1) - (interpret.c, prolang.c) It is now made sure that no function is passed more than 255 arguments (the maximum which can be encoded in just one byte). (b-991220) - (simulate, smalloc.c) Removed the MALLOC_STAT statistic. 26-Mar-2000 (Lars Duening) (3.2.8-dev.175) - (actions.c) remove_action_sent(): Removed a segfault in the DEBUG code (of all things). Thanks, Casey! - (sprintf.c) (s)printf() didn't add padding if no data was following the format field. This is probably a space- saving feature, but seriously hinders some formatting tasks. (b-000228, b-000306-5) - (closure.c) symbol_efun(), and with it efun symbol_function(), didn't know about #'foreach. Oops. (b-000228-3) - (closure.c) compile_value(): #'++ was accidentally compiled as #'--. (b-000228-4) - (lex.c) The variable pragma_no_shadow wasn't initialized before a new compile. (b-000103) 25-Mar-2000 (Lars Duening) - (prolang.y) Added two %type statements which allow the use of bison instead of byacc. (p-000205) - (efuns.c) Efun terminal_colour() didn't handle well the combination of indentation and forced wrapping in the middle of a word. (b-000217-2) - (main.c, simulate.c) Added a 'vdebug_message()' as companion to 'debug_message()', taking a va_list argument instead of the printf() style varargs. It's needed for a proper implementation of fatal() (I'm amazed that it didn't crash earlier). 23-Mar-2000 (Lars Duening) (3.2.8-dev.174) - (object.c) restore_value()/restore_object() modify the value string, which is desastrous if it's a volatile or shared string. (b-000212-1) - (doc/efun/strstr) Documented the negative start indices. (b-000106-2) - (doc/efun/sort_array) Made clear that the argument array is sorted in-place. (b-000106-3) 22-Mar-2000 (Lars Duening) (3.2.8-dev.173) - (interpret.c) call_others on arrays crashed with empty arrays, and failed to handle 0s gracefully. The crasher was caused by a premature optimization of the algorithm... (b-000221). - (closure.c) Compiling a #'foreach, <sym-array> construct didn't check if the sym-array is non-empty (b-000228-2). 21-Mar-2000 (Lars Duening) (3.2.8-dev.172) - (actions.c) When execute_command() called privilege_violation4(), it passed sp+1 instead of sp as the stackpointer to use. This lead inter_sp pointing to an uninitialized stack entry if the privilege was denied, causing a fandango on the core. (b-000110-2) - (interpret.c) set_bit(), clear_bit(), test_bit() did not check if the given bit number might be negative. (b-000112) 20-Mar-2000 (Lars Duening) (3.2.8-dev.171) - (actions.c) Revisited the 64-Bit issue and removed all casts. - (actions.c, mudlib/sys/commands.h, doc/efun/add_action) Corrected the behaviour of add_action() so that it is backwards compatible again. Added symbolic defines for the flags accepted by add_action(). Improved the manpage to include examples of how the flags affect the command parsing. (thanks, Freaky!) (b-991230) 10-Mar-2000 (Lars Duening) (3.2.8-dev.170) - (actions.c) Corrected a possible 64-Bit issue: a cast from (p_int) down to (int) might have lost bits while handling an address. - (prolang.y, lex.c) The modifier 'nosave' for variables takes over the meaning of 'static' - the latter is still recognized for variables, but its use is deprecated. Similar, 'nosave' should be used instead of 'static' as inheritance modifier. 09-Mar-2000 (Lars Duening) (3.2.8-dev.169) - (lex.c) Added the name of the macro to certain error messages issued during expansion. This also required a new function lexerrorf() with obvious purpose. 06-Mar-2000 (Lars Duening) (3.2.8-dev.168) - (doc/efun/closure_guide) Removed Tubmud'isms. - (bugs/...) New bug reports, and the EotL patches. 04-Mar-2000 (Lars Duening) (3.2.8-dev.167) - (interpret.c, doc/efun/tell_room) Doc corrections. - (interpret.c) f_traceprefix called master::valid_trace() with "trace" instead of "traceprefix". - (doc/efun/call_out) Corrections. - (bugs/...) Added all the bug reports since .164. 01-Mar-2000 (Lars Duening) (3.2.8-dev.166) - (make_func.y, exec.h, backend.c, parse.c) Removed warnings. 01-Mar-2000 (Lars Duening) (3.2.8-dev.165) - All the prvious changes since .141 are back under PRCS control. - (hosts/be/Makefile) The dependencies on generated files like stdstrings.h are now generated automatically as well (requires MkDepend 1.6). 17-Dec-1999 (Lars Duening) (3.2.8-dev.164) - (mapping.c) The callbacks for map(mapping) and filter(mapping) were told about more arguments than actually existing on the stack. Very stupid, but it took the Evermore lib to find this one. (thanks, Bardioc!) 16-Dec-1999 (Lars Duening) (3.2.8-dev.163) - (simulate.c) And of course the difference between toplevel and normal callbacks also includes the current_object :-( - (prolang.y) Typechecking for pre-increment/decrement on indexed values stumbled over mixed arrays. (thanks, DiEHARD). 15-Dec-1999 (Lars Duening) - (comm.c) Recognition of a quoted IAC in the input forgot to change the state back to TS_DATA. - (simulate.c) Freeing callbacks with just one arg used the wrong pointer and crashed. (thanks, Freaky!) - (simulate, array.c, comm.c, call_out.c, mapping.c) Argh - I have to distinguish if a callback is called from the toplevel (current_prog is invalid) or from a program (current_prog is valid). (thanks, Freaky!) 14-Dec-1999 (Lars Duening) (3.2.8-dev.162) - (pkg/mysql) Corrected a number of crashers (thanks, Bastian!) and added the HOWTO by Karm. - (actions.c) The recent changes activated a bit of legacy code, which started passing the verb as part of the argument string to NO_SPACE commands. *sigh* (thanks, Daryth!) - (simulate, call_out.c, comm.c) The callbacks can now accept lvalues as arguments. Also the index returned for faulty arguments was one too high. (thanks, Zwirch!) - (array.c) If setting up an efun callback generates an argument error, the proper, lower stack pointer is returned. (thanks, Zwirch) - (interpret.c) apply() on efun-, sefun- and operator-closures didn't properly check the number of arguments. This wasn't detected earlier because the related code has hardly ever been executed. (thanks, Zwirch!) - (comm.c) The driver telnet negotiations understands quoted IACs now. (thanks, Amylaar!) 13-Dec-1999 (Lars Duening) - (func_spec, array.c) Efun map() can now map an array through a mapping; similar to how filter() works. - (prolang.y) Indexing [..x] and [..<x] are allowed as short form of [0..x] and [0..<x]. (f-990419) - (prolang.y) Moved the parsing and compilation of the various indexing forms into two dedicated rules - this greatly simplified all the other rules which involve indexing (yes: for once a file became shorter :-). - (func_spec, prolang.y, interpret.c, closure.c, sprintf.c) The ranges [x..] and [<x..] are now mapped onto [x..<1] resp. [<x..<1]; this removes the need for F_EXTRACT2 for everything but the efun extract(). F_(PROTECTED_)EXTRACT_LVALUE are gone, too. - (sprintf.c) Made '%O' a bit more knowledgable about the various lvalues, but without the interpret.c-internal datastructures the printed data is just an approximation. 12-Dec-1999 (Lars Duening) (3.2.8-dev.161) - (array, interpret) Removed a few compilation problems. - (ed.c) If a file can't be read on editor start, the full message is printed. 12-Dec-1999 (Lars Duening) (3.2.8-dev.160) - (array, simulate, mapping, interpret.c) Modified the map_() and filter_() efuns, also sort_array() and walk_mapping(), to use the recently introduced callback_t for their implementation. This not only simplified the code (especially in array.c), but also introduces a standard way how callbacks are specified in such efuns. 11-Dec-1999 (Lars Duening) - (actions.c, sent.h) Implemented the "short verb" usage of efun add_action(). - (sent.h, ed) Completed the documentation of sentences. - (sent.h, comm, simulate, other files) The interactive_t is no longer a subclass of shadow_t; instead the latter now acts as generic "attribute" sentence for an object, linking to an independent interactive_t for interactive users. This makes handling interactive connections somewhat simpler. 10-Dec-1999 (Lars Duening) - (actions, gcollect.c) The garbage collector did not trace the globals of actions.c under the faulty impression that they are valid only during a command execution. However, e.g. notify_fail() can be used outside of a command, which led to used memory being deallocated in the GC. From there, the damage continued to spread. (at least part of b-991123-1) - (doc/LPC/functions) Rewritten to cover all the new features introduced since LPmud 2.4.5. - (interpret.c) Added check for stack overflow in the eval loop - code like funcall(lambda(0, ({ #'({ })+({1})*10000)) could otherwise easily crash the driver. (thanks, Zwirch!) 09-Dec-1999 (Lars Duening) - (Makefile.in, hosts/*/Makefiles) The manual dependencies for instrs.h were out of date. (reported by Freaky and Anders - thanks!) - (port.c) The current time_stamp() implementation omitted leading zeroes in the day field. (thanks, Freaky!) 08-Dec-1999 (Lars Duening) (3.2.8-dev.159) - (closure, gcollect.c, interpret.c) The closure compile did not handle lambdas with more than 254 constants properly: the number of values was consistently stored in the wrong place. A classic off-by-one, caused by the indexing from the end. (thanks, Zora!) 07-Dec-1999 (Lars Duening) - (pkg/mysql/) Small corrections. 06-Dec-1999 (Lars Duening) - (mud/lp-245/obj/master.c, .../simul_efun.c) valid_write() and valid_snoop() didn't work well with plain drivers due to the reliance on file_name(). The master now has its own file_name(). (thanks, Andy!) - (string_spec, make_func.y, Makefiles, sources) Tired of updated stralloc.* in three places for every change in the common strings table, I now define all these strings in string_spec and modified make_func to generate stdstrings.[ch] from these. - (interpret.c) Efun get_type_info(cl, 2) did not return the object for closure <cl> if <cl> was a lambda or bound lambda. (thanks, Freaky!) 05-Dec-1999 (Lars Duening) - (interpret.c, comm.c, simulate.c, object.c) Cosmetics. - (settings/dt2, mud/dt2/...) Created the Deeper Trouble II adaption from the old 3.2.1@141 diffs. It compiles and seems to work, but without the mudlib I can't tell for sure. 03-Dec-1999 (Lars Duening) (3.2.8-dev.158) - (interpret.c) Efun query_input_pending() deref'd the wrong object, causing a nice crash. (thanks, Michael!) 01-Dec-1999 (Lars Duening) - (doc/efun/inherit_file) Corrected typos (thanks, Freaky!). - (doc/efun/[]) Accidentally got lost in the last snapshot. - (interpret.c) Removed a sneaky memory leak in transfer_pointer_range(). (thanks, Michael!) - (func_spec) Efun functionlist() was specified to return "string*" instead of the correct "mixed*". (thanks, Freaky!) - (interpret.c) Improved the error messages of F_MAP_INDEX. 29-Nov-1999 (Lars Duening) - (func_spec, comm) New efun "set_combine_charset()" to combine characters received in charmode into strings. (f-991126-0) - (func_spec, comm.c) Efun "set_connection_charset()" now also accepts a string as charset specificator. 27-Nov-1999 (Lars Duening) (3.2.8-dev.157) - (port.h) A tentative change of VAR_PROT() and VARPROT() for AIX. - (prolang.h, prolang.y) struct s_lrvalue is now defined twice in prolang.y, to be able to cope with a wider variety of yaccs (especially the AIX yacc). - (util/xerq/(erq.c, socket.c)) Forgot to change two "struct queue_s" to "struct equeue_s". - (stralloc.c) string_dinfo_data() used an uninitialized variable. 26-Nov-1999 (Lars Duening) - (stralloc, gcollect.c, main, backend.c, Makefiles) When compiling with DEBUG and CHECK_STRINGS: The structures of the shared string table are shadows in a second table, and every string access checks the consistency of the structures. Additionally, with the commandline option "--check-strings" the whole string table is checked with every backend cycle. 25-Nov-1999 (Lars Duening) The following three changes should correct b-990430, which might have caused b-991123-0 and b-991123-1 in turn. - (comm.c) When accepting a line of non-negotiation text in charmode, interactive_t.tn_start became negative after the second char. Similar if an error occured during a very long line - here set_noecho() used wrong indices. - (comm.c) Charmode: when a client sent single CRs and there were no negotiations on the same line, the CR was re-read (almost) endlessly. - (comm.c) Should the telnet machine receive too much data, the program now makes sure that .command_end won't fall under .tn_start resp. .command_start. 24-Nov-1999 (Lars Duening) - (main.c) Commandline option "--check-refcounts" was mistakenly marked to accept a value. 23-Nov-1999 (Lars Duening) - (actions.c, backend.c, call_out.c, comm.c, ed.c, gcollect.c, interpret.c, lex.c, main.c, mapping.c, object.c, prolang.y, simul_efun.c, simulate.c, stralloc.c, swap.c, wiz_list.c) Added the timestamp to the diagnostics. error() and fatal() calls automatically log the timestamp. The whole situation of what is logged where when is still a mess :-( - (comm.c) Ups, call_function_interactive() forgot to remove its (stack located) error recovery context from the context stack. And it hid the leading "!" when falling through to IGNORE_BANG handlers. (thanks, Freaky!) - (prolang.y, stralloc) New master-apply "inherit_file()" to check and generated the full filename from an inherit statement. (suggested by Freaky) (f-990323-0) - (interpret.c) free_interpreter_temporaries() now also removes all destructed objects from the code trace (obviously only when compiled with TRACE_CODE). Since this function is called only from the GC, it has no further sideeffects. 22-Nov-1999 (Lars Duening) - (sprintf.c) sprintf("%O", object) in plain mode prints the object name with a leading "/". (thanks, Freaky!) - (lex.c) Master apply include_file() now receives the object filename in !compat with a leading slash; of the result, leading slashes are ignored. Also removed a typo. (thanks, Freaky!) - (simulate.c) Efun limited() did not give the closure the full amount of eval ticks are requested in the call. (thanks, Freaky!) - (mapping.c) Corrected the error handling in filter_mapping() and map_mapping(), which are used for the filter()/filter_indices() efuns and their map_() counterparts. (thanks, Robert!) - (port) New function time_stamp() for use in the creation of error and driver log messages. 21-Nov-1999 (Lars Duening) - (simulate, gcollect.c, comm, call_out.c) Introduced the "callback" structure, to be used for input_to()s and call_out()s. call_out.c became really simple :-) 20-Nov-1999 (Lars Duening) - (autoconf/configure.in, configure, port.h) Small tweaks. - (interpret.c) Correction to sort_array(): the GET_NUM_ARG occured too late. 19-Nov-1999 (Lars Duening) (3.2.8-dev.156) - (efuns.c, simulate.c, swap) Added new debug_info() request DINFO_DATA: DID_SWAP to retrieve the data usually printed as "status swap". - (efuns.c, smalloc) Added new debug_info() request DINFO_DATA: DID_MEMORY to retrieve the data usually printed as "status malloc". - (typedefs.h) Just a definition of strbuf_t was missing. 18-Nov-1999 (Lars Duening) - (comm.c, autoconf/configure.in, autoconf/acconfig.h, machine.h.in) Added tests if F_SETOWN and SO_OOBINLINE are actually supported by the operating system. - (port.c, autoconf/configure.in, config.h.in, settings/{default,osb}) Added the configuration option USE_SYSTEM_CRYPT. This solves the problem for OSB that the existing password files which have been created with IRIX' crypt() are invalid under HP-UX's crypt(). - (efuns.c, simulate, otable, call_out, heartbeat, stralloc, rxcache) Added new debug_info() requests DINFO_DATA, and implemented the first sub-request DID_STATUS. 17-Nov-1999 (Lars Duening) - (make_func.y, lex.c, efuns.c, util/xerq) Removed a few warnings when compiling under Solaris (some remain and are caused by the system headers); also changed the queue_t type in xerq to equeue_t because of a clash with the Solaris queue_t type. - (port.h, autoconf/configure.in, autoconf/acconfig.h, machine.h.in) Added a test for ssize_t. - (erq.c, xerq/lookup.c) Added two fixes provided by Uwe (thanks!). 16-Nov-1999 (Lars Duening) - (comm.c, settings/osb) Removed a few warnings when compiling under HP-UX; and the normal erq still doesn't compile there (but xerq does). - (func_spec, array, interpret.c) The efun sort_array() now accepts extra arguments to be passed to the comparison function. (suggested by Robert from Unholy Lands). 15-Nov-1999 (Lars Duening) - (strfuns.h) Changed "ulong" to "u_long", which should be the standard. - (mempools) Added an include of <sys/types.h> for ssize_t. - (settings/tubmud) Corrected some settings from "<foo>=<value>" to the correct "with_<foo>=<value>". (thanks, Coogan!) - (interpret.c) When apply() expanded an array, it didn't check the elements for destructed objects. (thanks, Coogan!) - (interpret.c) Before calling a vefun, inter_sp and inter_pc are updated. (thanks, Michael!) - (lex.c) Corrected an off-by-one error in the linenumbers of an inline closure. (thanks, Michael!) - (array.c, efuns.c, exec.h, interpret.c, prolang.y) The various inherit types are now distinguished by a proper "inherit_type" field in the inherit_t structure; as opposed to the former "is_extra" flag. (provided by Michael) 14-Nov-1999 (Lars Duening) - (smalloc.c) Added first hooks for use of madvise(). - (mud/anders) Added the diffs and extra files for the patched 3.2.1@141 driver of Deeper Trouble 2. - (interpret.c) A return from within a foreach() crashed because foreach's hidden "variables" messed up the stack check in F_RETURN. (thanks, Bardioc!) 11-Nov-1999 (Lars Duening) (3.2.8-dev.155) - (prolang.y, interpret.c, strfuns, make_func.y) The operators "&" and "-" (and their assignment variants "&=" and "-=") may now be applied on strings. - (comm.c) Arguments passed to an input_to() haven't been checked for destructed objects. (thanks, Anders!) - (lex.c, stralloc) News master apply include_file() to generate the full pathnames of files specified in #include directives. 10-Nov-1999 (Lars Duening) - (mud/sticklib/) Updated the diff files. 09-Nov-1999 (Lars Duening) - (bugs/p-990901) Updated the catch_no_log() patch. (thanks, Michael!) - (array.c) unique_array() on an empty array crashed. (thanks, Ralph!) 08-Nov-1999 (Lars Duening) - (simulate.c) Oops, "rename_to" has been written "rename_tp". (thanks, Elric!) - (gcollect.c) Added some DEBUG: printfs to check if the objects are counted properly. - (prolang.y) "*"::foo() now succeeds if at least one inherited object contains this function. - (prolang.y) The name table in get_type_name() was still incorrect. (thanks, Michael!) 07-Nov-1999 (Lars Duening) - (settings/default, unitopia) Updated the unitopia settings, and commented the max. hashtable values. - (object.c, closure.c, parse.c, sprintf.c) Removed a couple of compiler warnings. - (prolang.y, closure.c, switch.h) Removed a memory leak in the compilation of top-level switch() statements. 05-Nov-1999 (Lars Duening) (3.2.8-dev.154) - (regexp.c) In one place, FALSE was returned instead of TRUE, making regexp()/regreplace() unfunctional (thanks, Zwirch!). - (main, gcollect.c) writex() renamed to write_x() to avoid a clash with AIX' writex(). - (prolang) Some more AIX adaptions: struct s_lrvalue has to be defined after the rules (to be precise: after %union), and yyerrorf() has to use VAR_PROT() in the prototype. - (pkg/mysql) Added new efun db_handles() (provided by Karm). - (strfuns.h, smalloc.c) Added the necessary changes to compile under FreeBSD (thanks, Elric!). - (configure.in, configure) Took out the test for "-xO2" and "-xO4". It gave wrong results under FreeBSD, and doesn't seem to be useful anyway. 04-Nov-1999 (Lars Duening) - (configure) Updated it with the script created on Solaris from the configure.in. - (mudlib/dhrystone.c) Added Hyp's Dhrystone implementation for test purposes (486/33MHz/Cygwin: 6.7 Dhry/s). - (port, main.c) Cygwin's rusage() only returns 0s, so I replaced it with the simulation also used for Amiga - this way we at least get some time information. - (prolang.y) "#'f == string" crashed on printing an error message due to a lacking typename for the #'symbol. - (closure.c) Numbers bigger than 255 in closures weren't compiled. 03-Nov-1999 (Lars Duening) - (doc/LPC/foreach) Improved the documentation a bit. - (make_func.y) Since Solaris doesn't know stricmp() either, and since on Cygwin it is just a define to strcasecmp(), I just changed my two uses of stricmp() to strcasecmp(). 02-Nov-1999 (Lars Duening) - (prolang.y) Added "foreach(x in y)" as alternative to "foreach(x : y)" - without making "in" a reserved word. - (interpret.c) When confronted with negative starting offsets, strstr() returned invalid results. 01-Nov-1999 (Lars Duening) (3.2.8-dev.153) - (efuns.c) terminal_colour() stumbled over strings starting with a keyword. - (closure.c) The branch-rewriting was broken in @142 - using an invalidated variable, and rewriting only two of three branch types weren't good ideas at all. - (lex.c, prolang.y) Replaced the 'in' keyword in foreach() by ':' - mudlibs already use 'in' as variable name :-( 31-Oct-1999 (Lars Duening) - (erq.c) Small fix in the IPv6 rlookup, provided by Uwe. - (prolang.y) Cleaned up the lvalue rules for variables a bit, and also removed the "dangling else" shift/reduce conflict. - (func_spec, lex.c, prolang.y, interpret.c, closure.c) Implemented the foreach() loop construct. (f-981229-18) 29-Oct-1999 (Lars Duening) - (closure.c) An include of stddef.h was missing here, too. - (backend.c) In non-DEBUG mode, the interactive player wasn't set in the backend loop. (thanks, Heiko!) - (lex, prolang.y, simulate.c, exec.h) New program flag P_NO_SHADOW, set by the new #pragma no_shadow, disables shadowing for a program (suggested by Holger). - (efuns.c) Efuns debug_info()/object_info(), xINFO_MEMORY: also print resp. return the datasize of the object. (suggested by Freaky) - (sprintf.c) Multiples of 10 were printed without the leading digit. - (simulate.c) Efun rename(): the two valid_write() calls are now passed the operation codes "rename_from" and "rename_to" instead of "do_rename" twice. (suggested by Freaky) - (interpret.c, func_spec) call_other() now accepts arrays of objects and returns the call results in another array. 28-Oct-1999 (Lars Duening) - (settings/evermore) Added. - (make_func.y) Linux doesn't know stricmp(), but strcasecmp(). Added ad-hoc define for that. - (swap.c) Added missing include stddef.h. - (Makefile.in) The high WARN setting for gcc defined -Wno-parentheses instead of -Wparentheses. - (doc/LPC/pragma) The pragmas no_clone and no_inherit weren't documented. - (smalloc.h, interpret.c) Added braces around ambigouous if-elses. - (exec.h, prolang.y, lex.c, interpret.c) Runtime errors within inline closures were reported with wrong line numbers. Reason is that the normal store_line_number_info() expects line numbers to increase and therefore ignore the #line directives used with inline closures. Solution was to add a new LI_BACK linenumber code plus creation function to decrement the current line counter. (thanks, Zwirch!) 27-Oct-1999 (Lars Duening) (3.2.8-dev.152) - (xerq/erq.c) The global variable "sockets" was not initialized. 26-Oct-1999 (Lars Duening) - (interpret.c) In call_lambda(): moved the checks for destructed objects to the beginning. - (comm.c) If after an ERQ call back the object is found to be destructed, the callback is removed even if the ERQ sent a KEEP_HANDLE reply. - (regexp, rxcache, array.c, backend.c, ed) Regular expressions used from outside ed() throw errors instead of printing error messages. (b-990523) - (actions, func_spec) Efuns query_notify_fail(), command_stack(): the information, which object did the last notify_fail(), is now stored and accessible. (f-990217-2) - (access_check.c, lex.c, smalloc.c, wiz_list.c, hosts/amiga/dice30.diff) Just for fun, and because most of the changes had been done already: applied the remaining Amiga-diffs to the main sources. 25-Oct-1999 (Lars Duening) - (lex.c) Newlines in strings in (: :) closures are recognized and rejected. (thanks, Zwirch!) 24-Oct-1999 (Lars Duening) Only two months until Christmas! - (pkg/mysql) Added the mySQL support provided by Karm (thanks!). 22-Oct-1999 (Lars Duening) - (lex.c) Includes caused an offset-by-one for the linenumbers. - (comm, gcollect.c, interpret.c) input_to()s now accept closures for functions. (f-981229-21) - (prolang.y) Improved the ->() operator: the function name may be the same as a local variable, may be a string given as "ident" or a string expression given as (<expr>). (b-991019-0, f-990510, f-990605) - (prolang.y, stralloc) ->() again: a call_other() simul-efun is recognized. - (exec.h) Removed TYPE_SPACE and TYPE_TERM - nothing used them. 21-Oct-1999 (Lars Duening) (3.2.8-dev.151) - (prolang.y) If an object is inherited virtually multiple times, a wildcarded super call ("*::") calls it only once. (b-990217-0) 20-Oct-1999 (Lars Duening) - (prolang.y) In virtual inherits, the check for duplicate virtual functions forgot the take the virtuality of the duplicate into account. (part of b-981208-3) - (prolang.y) A virtual inherit is ignored if the program in question has been inherited virtually before. (part of b-981208-3) - (prolang.y, interpret.c) Simulefun may now take "varargs" type arguments (e.g. 'x (varargs mixed arg)'). (b-990113-0) 19-Oct-1999 (Lars Duening) Worked through a handfull of reports submitted by Rodney. - (doc/efun/say) Clarified the exclusion rules. - (interpret.c) Efun printf(): prints to NPCs are now sent to catch_tell(). - (efuns.c) Prompt-closures are protected against premature deletion when replaced while they are executing. - (svalue.h, interpret.c, prolang.y) New macro addref_closure() to easily increment the refcount of a closure-svalue. 13-Oct-1999 (Lars Duening) - (efuns.c) Efun terminal_colour(): when, during a wrap, spaces were removed from a string, this fact was not considered in the memory-access sanity checks. (b-991012-1) 12-Oct-1999 (Lars Duening) - (comm.c) Added another IPv6 patch. (thanks, Uwe!) - (erq, xerq) Added the ERQ_RLOOKUPV6 request supplied by Uwe (thanks!). 07-Oct-1999 (Lars Duening) - (comm, backend.c, gcollect.c) input_to() are now be nested when a command is executed from within an input_to(). This is not a real stacked input system as outlined in f-990721-0. 06-Oct-1999 (Lars Duening) - (swap.c) Corrected a small, but fatal omission of a cast. 05-Oct-1999 (Lars Duening) (3.2.8-dev.150) - (efuns.c) Straightened the key recognition in efun terminal_colour(). - (object) Added member object_t.load_id to be able to distinguish between objects created at the same time. - (func_spec, efuns) New efun clones() to find the clones of an object. (f-990203-00, f-990203-26) 04-Oct-1999 (Lars Duening) - (make_func.y) %tefuns specified with 'void' parameters are now recognized as vefuns. (b-990320-0) - (make_func.y, comm.c, ed.c, lex.c, main.c, simulate.c, swap.c) Use of MAXPATHLEN verified. - (prolang.y) Implemented '++<map>[<x>,<y>]'. (b-990204-1) - (func_spec, object) New efuns save_value() and restore_value() to encode/decode values in strings. (f-981229-19) - (func_spec, object) Extended the efuns save_object() and restore_object() to save and restore directly to and from a string. Maybe this should be put into separate efuns? (f-990203-25) 03-Oct-1999 (Lars Duening) (3.2.8-dev.149) - (util/xerq) Commented and beautified - now that was a nice program to work on. I left the docs/ and lpc/ files alone, for the moment, though. - (prolang.y) The dummy code created for undefined functions was not aligned even with ALIGN_FUNCTIONS - the other reason for the crash in functionlist(). (b-990819-0, see 26-Sep-99 as well) 02-Oct-1999 (Lars Duening) - (doc/util/erq, util/erq) Commented and beautified the erq a bit. There could be done more, but I really know better things to do. But note that somebody (probably the folks at TubMud) has added the socket requests to the erq. (b-981212) 01-Oct-1999 (Lars Duening) - (simulate, smalloc) Added a malloc statistic, which is dumped into the file MALLOCSTAT with the special command "status mstat". - (util/overhead.c) Commented. - (util/fundesc.c, util/make_docs/) Removed - these were programs for the original LPmud environment. 30-Sep-1999 (Lars Duening) (3.2.8-dev.148) - (func_spec, simulate) New efun copy_file(). - (smalloc) Commented and beautified. 29-Sep-1999 (Lars Duening) - (simulate, efuns, other files) Commented and beautified simulate.c; moved and removed several pieces of code which didn't belong there. - (prolang.y) The compiler thought, "int * string" yields int - a really silly typo (thanks, Michael!). 28-Sep-1999 (Lars Duening) - (exec.h, closure.c, interpret.c, object.c, simul_efun.c, prolang.y) Changed FUNCTION_NAME into FUNCTION_NAMEP to make sure people (like me) don't use it directly (alignment!). - (array.c) Corrected a mis-use of FUNCTION_NAME() which could lead to Bus Errors due to misaligned data accesses (thanks, Sunblood!) (b-990819-0, but see 03-Oct-99 as well) - (sprintf) Commented, beautified, and two changes: formatting erros now throw an error(), and recursive arrays/mappings are recognized. 27-Sep-1999 (Lars Duening) - (swap) Commented and beautified. - (stralloc) Commented and beautified. 26-Sep-1999 (Lars Duening) (3.2.8-dev.147) - (simul_efun) Commented and beautified. - (wiz_list) Commented and beautified. 25-Sep-1999 (Lars Duening) - (interpret.c) Corrected a nasty crasher in the multiplication of arrays of malloced strings (thanks, Michael!). - (regexp) Beautified it, but didn't add many comments. 24-Sep-1999 (Lars Duening) (3.2.8-dev.146) - (prolang.y, lex.c, closure.c) Added #'parse_command. - (prolang.y) Corrected a rare problem with the code generated for '&(expr[x])' and '&(expr[<x])'. - (prolang.y, lex, make_func, lotsa files) Finally: the lexer/compiler tokens no longer double as machine instructions. Not only this simplifies include dependencies, it also gets rid of that dreadful F_OFFSET (and '%instrs' is gone again, too). *Lars bounces happily*. - (make_func.y, Makefile) make_func now takes arguments determining what to create. This greatly reduces compile times after changes to the LPC compiler. See the source of make_func.y for more details. 23-Sep-1999 (Lars Duening) - (prolang.y, exec.h, svalue.h) Commented and beautified the compiler. 21-Sep-1999 (Lars Duening) - (array.c, doc/efun/functionlist) Clarified that efun functionlist() does not load objects from a name, and improved the error message for this case. 20-Sep-1999 (Lars Duening) - (make_func.y) Added the numeric bytecodes to the created efun_defs.c and intrs.h - helps with debugging sometimes. - (make_func.y, func_spec, prolang.h, closure.c, dumpstat.c efuns.c, gcollect.c, lex.c, array.c, interpret.c, swap.c, simul_efun.c, sprintf.c) Added a new func_spec-type '%instrs' to distinguish the instructions used as tokens from those which aren't. This allows less files to depend on lang.h . 19-Sep-1999 (Lars Duening) (3.2.8-dev.145) - (parse.c, doc/efun/parse_command) Commented and ansified. - (parse_old.c) Beautified, but only sparsely commented. 18-Sep-1999 (Lars Duening) (3.2.8-dev.144) - (mudlib/lp-245/obj/simul_efunc.c) Adapted to the new definition of inherit_list(). - (stralloc) Added the parse_command() lfuns as predefined shared strings. 17-Sep-1999 (Lars Duening) (3.2.8-dev.144) - (mudlib/sticklib) Adapted the interpret.c.diff. - (prolang.y) Incorrect initialisation of local blocks in inline closures sometimes generated illegal code (thanks, Michael!). (also b-991026-1) - (array.c, interpret, doc/efun/inherit_list) Efun inherit_list() now prepends the returned program names with '/'es in non-compat mode. (thanks, Michael!) - (configure.in, configure) Corrected (I hope!) the tests for inet_ntoa() and ipv6 support. - (func_spec, interpret.c, doc/efun/) Efun load_name() also accepts strings, following a suggestion by Coogan. 16-Sep-1999 (Lars Duening) (3.2.8-dev.143) - (efuns.c) Documented process_string(). - (closure.c, mud/lp-245/obj/master.c) Corrected small errors. 15-Sep-1999 (Lars Duening) - (efuns, parse, func_spec, interpret.c) Moved process_string() into efuns.c and made it a tabled efun. 14-Sep-1999 (Lars Duening) (3.2.8-dev.142) - (closure, doc/LPC/closures, svalue.h) Ansified and documented. - (switch.h, prolang.y, typedefs.h) Beautified and documented the data structures used for the switch(). - (exec.h) Added macros to generate code. - (Makefile.in) The rule running make_func did not delete instrs.h first. - (configure.in, configure) Corrected the probing for optimization options. - (hosts/unix.h) Added an additional check for the newer Cygwin packages (the old 'gethostname' problem). - (strfuns.h) *sigh* ulong is not a standard type - we have to clean that up later. 19-Aug-1999 (Lars Duening) (3.2.8-dev.141) - (prolang.y) Forgot to change one 'struct lambda' into 'lambda_t'. - (comm.c) Corrected a wrong if() condition in efun exec(). 16-Aug-1999 (Lars Duening) (3.2.8-dev.140) - (actions.c) Oops, the marker_sentence was not given a valid type. 15-Aug-1999 (Lars Duening) - (gcollect.c, parse_old.c, parse.h) The {clear,count}_parse_refs() functions exist in both new and old parse_command() *duh*. - (Makefiles) Updated all dependencies. 14-Aug-1999 (Lars Duening) (3.2.8-dev.139) - (my-alloca.h) Removed a declaration conflict on BeOS/PPC. - (settings/avalon) Updated. - (comm.c) Removed a compilation warning. - (prolang.y) Redeclaration of local scopes inside of for() statements did not recover gracefully. - (lex.c) If the driver is compiled for IPv6, the macro __IPV6__ is defined. - (autoconf/configure.in, configure) Improved tests for IPv6. 13-Aug-1999 (Lars Duening) (3.2.8-dev.138) - (sent.h, actions, simulate, other files) Beautified the sentence handling. - (mud/lp-245) Added mudwho-code and explanations for non-compat masters to the obj/master.c - (mudlib/) Removed the old master.c, simul_efun.c and room.h. - (comm, port.h, autoconf/*) Integrated the AIX and experimental IPv6 support provided by Uwe (thanks!). - (object.h) Forgot to include instrs.h even though part of the struct object depends on F_SET_LIGHT *duh*. 12-Aug-1999 (Lars Duening) (3.2.8-dev.137) - (svalue.h) Created from old datatypes.h. - (typedefs.h) File holding the opaque definition of several globally used types <foo>_t (which are shortcuts for the old 'struct <foo>'s); mainly used to decouple the header files. - (other files) Adapted to use the <foo>_t types. 12-Aug-1999 (Lars Duening) (3.2.8-dev.136) - (strfuns) Increased the possible size of a strbuf from 2**16 to 2**32. - (mempools, alloca.c, Makefiles) Added mempools: specialised allocators. - (array.c, lex.c, ptrtable) Added uses of mempools where useful. 11-Aug-1999 (Lars Duening) (3.2.7-dev.135 == 3.2.7-pre.1) - (lpmud.prj, patchlevel.h, hosts/be/Makefile) Introduced new source branch '<v>-<r>-<p>.*' to hold the pre-releases during the shakedown period before the 'official' release. Features and such in these pre's are fixed, only fixes for final bugs take place. 09-Aug-1999 (Lars Duening) - (mapping.c) Efun m_reallocate() returned the old (freed) mapping, not the new one. - (interpret.c) Before every bytecode evaluation, inter_sp and inter_cp are updated. - (pkg/ip6-aix) AIX/IPv6 package added. 04-Aug-1999 (Lars Duening) (3.2.7-dev.134) - (util/xerq/) Corrected a couple of typos (I hate not being able to compile this!) (thanks, Freaky!). - (simulate.c) Cosmetic for 'status tables': the apply-cache stat lacked a divider. 31-Jul-1999 (Lars Duening) (3.2.7-dev.133) - (interpret.c) '*', '*=': inter_sp wasn't set properly before error() was called. (thanks, Michael!) - (array, swap.c) Swapping in an array ignores the current size limit. (thanks, Michael!) - (efuns.c) Efun terminal_colour(): the temporary buffer for the current line is now potentially unlimited. - (util/xerq/) Integrated patches from UNItopia, which remove a couple of warnings and works around the repeated closing of sockets. (thanks, Freaky!) 30-Jul-1999 (Lars Duening) - (efuns.c) Efun trim(): forgot a check for the string end when trimming left. This caused trim("") to core. (thanks, Freaky!) 30-Jul-1999 (Lars Duening) (3.2.7-dev.132) - (gcollect.c) Oops, forgot to change two calls to free_vector() into free_array(). (thanks, Freaky!) - (interpret.c) Efun mkmapping() could have problem with oversized arrays. 29-Jul-1999 (Lars Duening) (3.2.7-dev.131) - (simulate.c) I am surprised nobody complained about this one: when an error occured, the runtime stack was unrolled too early and left the runtime_error() apply with no valid commandgiver. - (func_spec, simulate) New efuns set_limits(), query_limits() and limited() to handle runtime limits. (f-981229-11) - (backend, main, simulate): array, mapping, and file transfer limits are all settable from commandline and through the limit efuns. - (interpret.c, efuns.c) Code for efuns like object_info() now use the mudlib/sys/ includes to generated their results. - (interpret.c, prolang.y) The '*' and '*=' operators now allow to multiply strings/arrays with numbers. 28-Jul-1999 (Lars Duening) - (lots of files) Removed trailing spaces. - (main, backend, lex, interpret, simulate) eval_costs are now counting up from 0 and are tested against a positive limit. The limit can be 0, in which case the eval costs are still counted but can grow without limit. This is just a preparation for the oncoming code, and might even work. 27-Jul-1999 (Lars Duening) (3.2.7-dev.130) - (efuns.c) Efun trim() also accepts a string of characters to trim. - (array.h, datatypes.h, closure.h, mapping.h) Moved the definitions of non-svalues back into their associated module header files. - (datatypes.h, stralloc.h) Wrapped setting of svalues into nice put_<type>() macros. For once, the sources shrank a few lines instead of growing :-) 26-Jul-1999 (Lars Duening) (3.2.7-dev.129) - (array.h, mapping.h, object.h, and dependees) Streamline the refcount handling in macros ref_<type>(), deref_<type>() and free_<type>(). Not necessary in terms of typesafety, they make the source a bit more readable, and less likely that a ref++ is forgotten. 23-Jul-1999 (Lars Duening) (3.2.7-dev.128) - (lex.c, prolang.y) The inline function variables weren't initialized properly nor checked, leading to interesting crashes when the driver tried to freed non-existing memory. - (lex.c, strfuns) Rewrote the inline closure scanner to cope with linefills. 22-Jul-1999 (Lars Duening) - (access_check.c) A null-message could cause a random memory write. - (strfuns.c) Corrected a write beyond boundary. - (lex.c) Inline function scanning: newlines didn't trigger a buffer refill in time. 22-Jul-1999 (Lars Duening) (3.2.7-dev.127) - (actions, func_spec) New efun 'execute_command()' implements a low-level access to the command parser. - (prolang.y, exec.h) New optional driver hook H_COMMAND, which replaces the driver based command handling with its own. (obsoletes f-990204-1 and f-990203-09) - (make_func.y) Lookup failures yield an error message, not just an exit(1). - (func_spec, efuns) New efun 'trim()' to trim characters from a string. (f-981229-24) - (settings/) Small corrections to 'default', added 'wunderland'. - (pkg/ip6/) Added the experimental IPv6 patch from Uwe (thanks!). 21-Jul-1999 (Lars Duening) (3.2.7-dev.126) - (settings/) Updated the tubmud settings file, added the sticklib file. All scripts now use 'exec' to start the configure script, allowing more meaningful exit codes. - (simulate.c) Caught errors now produce a tracedump on stdout/stderr (thanks, Sunblood!). - (mud/sticklib/) Added the StickLib changes (thanks, Doomdark!) - (lex.c) 'check last char before :)' stumbled over Newline characters. - (prolang.y) Inline closures: removed an interaction between typed inline closures in untyped functions. - (actions.c, comm) Notify-fail handling is now in actions.c, this also means that notify-fail messages are saved over nested commands. - (actions.c) notify_fail is available for NPCs, too: the text will be told to the NPC. (f-990204-5) - (func_spec, actions) New efun 'query_notify_fail()' returns the current notify-fail setting. (f-990217-2) - (func_spec, actions) New efuns 'command_stack()' and 'command_stack_depth()' to query the command stack. 20-Jul-1999 (Lars Duening) (3.2.7-dev.125) - (simulate, interpret.c, backend.c) Player command execution is now stacked, meaning that after a command() the previous command giver, verb, etc are restored. This does not yet extend to notify-fails though. - (actions.ch) New source files holds the functions related to command parsing and action management. 19-Jul-1999 (Lars Duening) (3.2.7-dev.124) - (prolang.y) Removed a spurious debug printf(). - (backend, call_out.c, closure.c, interpret.c, simulate.c, main.c) Generalized the error_recovery_info stack into a runtime context stack. 18-Jul-1999 (Lars Duening) (3.2.7-dev.123) - (settings) Updated the tubmud settings file. - (autoconf/configure.in, configure) Unfortunately it's not just gcc 2.8.1 on Solaris which stumble over the float/branch test. - (prolang.y) Removed a nasty interaction between disabled local scopes and for() statements (which unfortunately didn't manifest itself in earlier tests). Also, the local_scopes pragmas are check only when entering a function. - (otable.c) The table status threatened to print some stack garbage. - (main, config.h.in, interpret.c, array.c, mapping.c, object.c, simulate.c) Array and mapping size limits are now commandline options; and a setting of 0 disables the limit checks. 16-Jul-1999 (Lars Duening) - (autoconf/configure.in, configure) Oops, a typo in configure.in really messed up the script. - (backend.c) Following a notice from Stefan Lichtenhagen, I removed the safety factor '*2' from f_regreplace() - the error it avoided was fixed in dev.28. However, if this efun should cause crashes again, this factor is one thing to try out. - (simulate.c) Stupid bug, the second: clone_object() checked the P_NO_CLONE flag before swapping in the program. (thanks, Manu!) - (lex, prolang.y) New pragma 'no_local_scopes' turns off local scoping to cater for legacy code; the second pragma 'local_scopes' turns it back on. - (settings/) Added settings file for Avalon (thanks, Thorsten!) 16-Jul-1999 (Lars Duening) (3.2.7-dev.122) - (lex, prolang.y) Added inline closures ("functionals") using the '(: <code/expression> :)' syntax MudOS invented. As a sideeffect, the lexer now knows the '#line <num>' directive, and identifier names may start with '$'. (thanks, Righ!) (p-990326) - (func_spec, interpret.c, simulate) New efun query_command() returns the full command given by the player (thanks, Ugh!). 15-Jul-1999 (Lars Duening) (3.2.7-dev.121) - (prolang.y) The first expression in a for() may define local variables which scope is the whole for() statement. - (func_spec, interpret.c) More new bitstring efuns: count_bits(), invert_bits(). 14-Jul-1999 (Lars Duening) (3.2.7-dev.120) - (func_spec, interpret.c) New bitstring efuns last_bit() and next_bit(). (thanks, Matthew!) (f-990203-14) - (access_allow, comm, mudlib/ACCESS.ALLOW) Access rules can take portnumbers into account. (idea from Gokhan, f-981211) 13-Jul-1999 (Lars Duening) - (bugs/) Cleaned up the directory a big; created summary files b-summary and f-summary. - (func_spec, interpret, object, doc/efun/) New efun object_time() returns the creation time of the object. (idea from Daniel, f-990324-2) - (lex, prolang.y, simulate.c, exec.h) New pragmes 'no_clone' to disallow clones of an object, and 'no_inherit' to disallow inheriting a program. (thanks, Daniel!) (f-990324-0, p-990417-0) - (comm.c) Error ESHUTDOWN on read yields a proper message and removes the interactive player. - (func_spec, interpret.c) New bitstring efuns and_bits(), or_bits(), xor_bits(). (idea from Matthew, f-990203-14) 12-Jul-1999 (Lars Duening) - (func_spec, interpret.c) New efuns pow(), floor(), ceil() and atan2(). - (config.h.in, configure, interpret.c) If configuration option DYNAMIC_COSTS is enabled (default is off), string additions are more expensive for long strings (this is handy against the killer-loop for (s=""; s += "a"; ) ). 12-Jul-1999 (Lars Duening) (3.2.7-dev.118, 119) - (autoconf/configure.in, configure) gcc/cc1 2.8.1 on SunOS 5.7 falls into catatonia when compiling the float/branch-incompatibility test. Inserted a workaround. - (prolang.y, lex, func_spec) Scope of variables in nested local blocks ends with the defining block as it should. The space of such variables is reused for other local blocks. (f-990203-05) - dev.119: Small correction: function local variables (depth 2) can't hide function arguments (depth 1); 11-Jul-1999 (Lars Duening) (3.2.7-dev.117) - (machine.h.in) Oops, forgot to update this. - (Makefile.in, port.h) Compile-time define NO_INLINES turns off inlining. - (interpret, stralloc, strfuns.c, smalloc) Straightened out the INLINE uses. 10-Jul-1999 (Lars Duening) (3.2.7-dev.116) - (closure.c) Removed an erroneous errormessage when creating sefun closures with large indices (thanks, Daniel!). - (configuration, several files) Mappings have now an upper size limit of MAX_MAPPING_SIZE keys (defaults to 5000). (f-990708-1) 09-Jul-1999 (Lars Duening) (3.2.7-dev.115) - (config.h.in, autoconf/configure.in, settings/default) All of of config.h is now configurable (and is it just me, or was the original value parsing really broken?). - (settings/) Added files for all Muds I had a config.h of. Oh, and the files are now scripts in themselves :-) 08-Jul-1999 (Lars Duening) (3.2.7-dev.114) - (machine.h.in, port.h) port.h now knows about the supported (working) architectures and how to find the host-specific include files. HOST_INCLUDE is still recognized, but optional. - (machine.h.in, array.h, driver.h, port.h, interpret, simulate, random): Use SIZEOF_CHAR_P instead of SIZEOF_P_INT. - (Makefile.in) Target 'distclean' also deletes the config.cache/status/log files. - (other files) Added some documentation to the new autoconfig system. 08-Jul-1999 (Lars Duening) (3.2.7-dev.113) Checked in the changes provided by Reinhard (thanks!): - (machine.h.in, config.h.in, configure, autoconf/) Autoconfiguration is now based on autoconf 2. The old configuration code is in the subdirectory config-old/ held as backup. Special features are: - rxcache and UDP port can be enabled/disabled explicitely. - Enabling ERQ defines a symbol ERQ_INCLUDE pointing to the correct include file (util/erq/erq.h resp util/xerq/erq.h). - All config.h options can be set using parameters to configure. configure --help will print the (long) list of available options. - (func_spec) Fixed a rare case of an erroneus inclusion of send_imp() UDP is disabled and thus the function f_send_imp() won't be compiled. - (comm.c) erq.h is included only when the driver is compiled for using the erq demon. 07-Jul-1999 (Lars Duening) (3.2.7-dev.112) - (strfuns.c) Corrected off-by-one mistakes, and the length of the final string. - (datatypes.h, other files) Straightened the use of STRING_VOLATILE (purely cosmetic). 07-Jul-1999 (Lars Duening) (3.2.7-dev.111) - (strfuns, efuns, simulate): Modified efun debug_info() to return the status information instead of printing it. Added "malloc" as additional second argument to the DINFO_STATUS command. - (interpret) Efun get_type_info() returns whether a string is shared or not. - (simulate, smalloc): Deactivated 'debugmalloc' and 'showsmallnewmalloced' and the associated code. - (simulate, efuns, dumpstat, interpret): Added option DINFO_DUMP to debug_info() to implement the 'dumpallobj' and 'opcprof' command. - (interpret.c) Efuns trace() and traceprefix() are now validated with the master-lfun 'valid_trace()'; the master-lfun 'query_player_level()' is thus no longer needed. - (efuns, simulate, object): Efun set_is_wizard() is now optional, and with it the object flag O_WIZARD. - (object, simulate, prolang, efun): Object flag O_APPROVED is gone. 06-Jul-1999 (Lars Duening) - (comm, simulate.c) Undid the last change. Now it is legal again for the modify_command-hook to change the command_giver; this change stays active for the execution of the command, and the original commandgiver is passed to the NOTIFY_FAIL hook. - (simulate.c) The MODIFY_COMMAND hook is now called first, even before special commands like 'status' are parsed. This allows the hook to filter out uses of these commands. - (strfuns) New file holding string functions (for now). - (simulate, otable, rxcache, stralloc, efuns, call_out, heartbeat): Rewrote the status functions to use strbufs, in anticipation to make these special commands into real efuns. 05-Jul-1999 (Lars Duening) (3.2.7-dev.110) - (comm.c) include of <stddef.h> was missing. - (simulate.c) After the modify-command hook was called, the commandgiver and current object are restored their previous setting in case the hook changed them. (thanks, Daniel!) 04-Jul-1999 (Lars Duening) - (closure.c) Corrected a nasty mistyping in compile_lvalue(). (thanks, Freaky!). - (lex.c, interpret.c) Removed two warnings. - (doc/efun) Removed the manpage for the long-gone efun query_host_name. 03-Jul-1999 (Lars Duening) (3.2.7-dev.109) - The driver compiles again with egcs. - (efuns.c) Bugfix for efun terminal_colour(): terminal_colour("bla", ([ "bla":"blu"]), 1) returned "blu" instead of "bla". (thanks, Michael!) - (efuns.c) The efun sscanf() used the wrong function to report argument errors. (thanks, Freaky!) 02-Jul-1999 (Lars Duening) - (hosts/be/be.h, .../Makefile, .../machine.h, other files): Adapted for BeOS 4.5. This involved adding numerous casts to make the now pickier compiler happy. But now egcs starts crashing... - (efuns.c) Found another shortcoming of present_clone() (and the reason why I didn't see errors before *blush*). 01-Jul-1999 (Lars Duening) - (efuns.c) Fixed two nasty bugs in efun present_clone(): make_name_sane() could return NULL, and the comparison was done against the wrong name. - (interpret.h) To be safe, macro put_number() takes the sp as argument. - (call_out.c) When setting current_prog for efun closures, the code now makes sure that the program is swapped in. 30-Jun-1999 (Lars Duening) (3.2.7-dev.108) - (mud/lp-245/obj/simul_efun.c) Duh, had a complete brainfart yesterday when adapting present_clone() to compat mode. 29-Jun-1999 (Lars Duening) (3.2.7-dev.107) - (doc/efun) Removed the files for the long-gone efuns find_living(), find_player(), set_living_name(), add_worth() and wizlist(). - (doc/efun/all_environment) Corrected a typo. - (efuns.c) Corrected a misformed condition in efun terminal_colour(). - (gcollect.c) If a swap-in fails, the GC raises a fatal error (instead of just crashing). - (mud/lp-245/) Added the old LP-245 mudlib, of course adapted to this driver. - (func_spec, efuns) New efun present_clone() to find objects with a specific blueprint (suggested by Coogan). 29-Jun-1999 (Lars Duening) (3.2.7-dev.106) - (exec.h, interpret.c, call_out.c) Efun-closures are now properly identified, and call_outs to efun-closures set current_prog to the prog of the bound object. This sources still hold some DEBUG-printfs. 28-Jun-1999 (Lars Duening) (3.2.7-dev.105) - (interpret.c) Added some argument checks to efun terminal_colour(). - (func_spec, doc/efun/environment) Documented that efun environment() can also take a string as argument. - (interpret.c) 'Indexing on illegal type' no longer dumps trace (what for did that do that any way?). - (main, backend.c, object.c) TIME_TO_RESET and TIME_TO_CLEAN_UP are now defaults for the runtime options '--reset-time' and '--cleanup-time'. - (interpret.c) Reactivating a destructed didn't update the number of listed objects. 26-Jun-1999 (Lars Duening) - (main.c, interpret.c, smalloc.c) Cosmetics. 25-Jun-1999 (Lars Duening) (3.2.7-dev.104) - (lotsa files) NATIVE_MODE is gone - the only real difference it made was the restriction of command() to the current object only, and that can be handled by a simul-efun. - (interpret.c) Oops, privilege_violation() slightly misinterpreted the result from the master apply. 25-Jun-1999 (Lars Duening) (3.2.7-dev.103) - (lotsa files) EUIDS mode is now automatic since most of the code was there in any case. New runtime option '--[no-]strict-euids' which enables/disables the enforcement of euids known from the native mode. When enabled, the LPC macro "__STRICT_EUIDS__" is defined. The default setting is defined in config.h by STRICT_EUIDS; use of an old config.h with NATIVE_MODE defined has (for now) the same effect. 24-Jun-1999 (Lars Duening) (3.2.7-dev.102) - (backend, efuns.c, heartbeat, interpret.c, object, simulate, all config.h) OLD_RESET is gone. 24-Jun-1999 (Lars Duening) (3.2.7-dev.101) - (mudlib/sys/{trace.h,input_to.h,files.h,debug_info.h}: Added these includes with definitions for the efuns trace(), input_to(), file_size(), get_dir(), and debug_info(). - (interpret, efuns, func_spec, exec.h) Finished commenting the interpreter. 20-Jun-1999 (Lars Duening) (3.2.7-dev.100) - (interpret) to_string(object) in compat mode added a leading slash. program_time() freed its argument before querying it. And recombine_free_lists() is gone for good. - (interpret, func_spec, exec.h) After a long day: more comments added (17155 lines done, 3150 to go). In particular, eval_instruction() is commented *phew* (but it's still big and ugly). 18-Jun-1999 (Lars Duening) (3.2.7-dev.99) - (interpret, func_spec, exec.h) Something new: more comments. (12558 lines done, 5800 to go). 16-Jun-1999 (Lars Duening) - (interpret.c) The operators +=, -=, *=, /= now accept a float left and an int right. - (interpret, efuns) Efun terminal_colour() moved out from the interpreter into efuns.c. And it received comments during this move. 15-Jun-1999 (Lars Duening) - (interpret.c) Binding an unbound lambda to two objects and destructing the originating object freed the lambda. (thanks, Kirnan & Michael) 14-Jun-1999 (Lars Duening) - (driver.h, port.h, all files commented so far) New type 'Bool' and its smaller sibling 'CBool' introduced. - (autoconf/configure.in, configure, port.h, regexp, ptrtable) Replaced CHARBITS by CHAR_BIT, adapted the generation of CHARBIT_MASK. - (datatypes.h, func_spec, exec.h, interpret) Something new: more comments (8700 lines done, 9000 to go). Actually I started wrapping the bytecode interpretation into its own types and macros, which may increase portability, and at least make the critical places easier to find. 13-Jun-1999 (Lars Duening) - (array.c) Efun inherit_list() just copies the program names, but doesn't make them shared. Since they aren't shared anyway, this makes the function much faster. (thanks, Michael!) - (object.c) If the H_RESET hook is a closure, a result < 0 will disable reset for this object. (thanks, Michael!) 12-Jun-1999 (Lars Duening) - (interpret.c, mapping.c) Both m_indices() and map()/map_mapping() called check_map_for_destr() twice, wasting time. (thanks, Michael!) 12-Jun-1999 (Lars Duening) (3.2.7-dev.98) - (datatypes.h, func_spec, exec.h, interpret) Oh boy: more comments (6700 lines done, 10000 to go). 10-Jun-1999 (Lars Duening) - (call_out.c) Adding callouts from within a callout resulted in the new callouts inserted in the wrong place when the active callout had an odd delay (and thus a negative 'time left' entry). If the new call_out had a MAX_INT delay, the current callout list was effectively disabled. (b-990214-0) - (backend.c) Added Michaels improvement to the swapping algorithm (thanks!). 05-Jun-1999 (Lars Duening) - (smalloc.c) As I feared: trying to print more information about gc'ed objects may crash the driver *sigh*. Patches provided by Michael (thanks!): - (backend, interpret, simulate) Improved the handling of closure hooks: now setting a driver hook from within the hook works well, too. - (closure.c) symbol_variable() properly recognizes objects with pending program replacement. - (interpret.c) dump_trace() recognizes alien lfuns properly. - (interpret.c) Calls to alien lfun closures didn't set extern_call. - (interpret.c) Calls to F_UNDEF closures (closures whose object was destructed) generate a proper error message. - (swap.c) If a closure variable was changed to F_UNDEF by the garbage collector, the swapper now duly writes this new info into the swap file. - (mapping.c) During a GC, mappings didn't check for changed alien lfuns. Some more suggestions by Michael: - (interpret.c) to_object() on F_UNDEF closures returns 0. - (object.c) free_object() miscounted the amount of memory used by objects. - (simulate.c) Memory display in 'status' for objects and programs now both display '<total> (<swapped>)' (before, only the objects status displayed that, programs displayed '<resident> (<swapped>)' ). - (swap.c) If variables could not be unswapped, the swapper didn't clear them properly. 24-May-1999 (Lars Duening) (3.2.7-dev.97) - (interpret, exec.h, heartbeat.c, comm.c, make_func.y) More comments, really. 22-May-1999 (Lars Duening) (3.2.7-dev.96) - (config.h, interpret.c, closure.c) Corrected some typos. 22-May-1999 (Lars Duening) (3.2.7-dev.95) - (interpret.c) Guess what: more comments. - (interpret.c, sprintf.c) to_string() and sprintf() stumbled over variable closures left dangling after a replace_program(). (thanks, Michael!) - (prolang.y) insert_inherited: the loop for the case '*::f' tended to clobber its own control variable 'ip', causing the omission of function calls. (thanks, Michael!) 18-May-1999 (Lars Duening) (3.2.7-dev.94) - (config.h, hosts/be/config.h) Increased the default for MAX_LOCAL to 50. - (interpret.c) Even more comments. 17-May-1999 (Lars Duening) (3.2.7-dev.93) - (doc/efun/filter_{mapping,indices}) Corrected a wrong phrase. (thanks, Zwirch!) - (interpret.c) Some more comments. 16-May-1999 (Lars Duening) (3.2.7-dev.92) - (object, backend.c, dumpstat.c, efuns.c, interpret.c, simulate.c, sprintf.c, ../mudlib/sys/objectinfo.h, doc/efun/object_info): New object flag O_REPLACED for objects with replaced programs, so that clonep() stops flagging such blueprints as clones. (suggested by Michael) - (datatypes.h, interpret) More comments. - (simulate.c) Streamlined the namehandling in clone_object(). 14-May-1999 (Lars Duening) (3.2.7-dev.91) - (backend.c) Removed one 'var might be clobbered' warning. - (datatypes.h, interpret) More comments. - (object.c) When restoring a mapping of form '([:<width>])', the <width> is made sure to be positive. (thanks, Michael!) 13-May-1999 (Lars Duening) (3.2.7-dev.90) - (datatypes.h, interpret) More comments. 12-May-1999 (Lars Duening) (3.2.7-dev.89) - (interpret.c, doc/efun/...) Efun symbol_function() won't create closures from protected lfuns of other objects. (thanks, Michael!) - (func_spec, comm, interpret.c, config.h, driver.h, main) query_mud_port() did not accept int arguments as documented; also, the multiport feature is no longer optional. - (smalloc.c) write_lpc_trace() tries to write as much information about an object as possible, not just 'Doesn't exist anymore'. - (interpret.c) destruct() now accepts destructed objects as arguments. (thanks, Largo!). 11-May-1999 (Lars Duening) (3.2.7-dev.88) - (object.c) Fixed some memory leaks on errors during a restore. (thanks, Michael!) - (interpret.c) Some more comments. 10-May-1999 (Lars Duening) (3.2.7-dev.87) - (interpret.c) Removed a crasher in call_lambda() on closures of not-inherited variables. (thanks, Michael!) - (closure.c) Creating a symbol_variable() now protects it against replace_program()s. (thanks, Michael!) - (array.c, efuns.c, exec.h, interpret.c, prolang.y) Duplicate virtual inherits are no longer listed in the result of inheritlist() and the various statistical functions. (thanks, Michael!) 09-May-1999 (Lars Duening) (3.2.7-dev.86) More patches from Michael: - (backend.c, interpret, gcollect.c) interpret:call_lambda() no longer protects closure with an extra refcount; instead old, replaced driver hooks are stored in an array and freed by the backend (or the garbage collector). - (backend.c) When computing a 'real soon' swapping time for objects with no reset, the new swap time became negative, effectively disabling swapping for this object. - (interpret.c) Improvement of efun terminal_colour(), reducing the number of created shared strings. - (simulate.c) If the loading of objects fails, the name of the object is printed with the error message. - (swap.c) If an object WILL_CLEAN_UP on swap in, the swapper doesn't bother to check for the function 'clean_up'. - (interpret.c, prolang.y) Added a couple of missing updates of time_of_ref. - (interpret.c) Fixed a crasher when using to_string() on an lfun closure for a swapped program. - (simulate.c) lookfor_object() no longer swaps in the object, instead the caller does it when needed. As this is rather rare, the effect should be reduced disk activity. - (ed.c, heartbeat.c) Removed two compilation warnings for printf() args. 04-May-1999 (Lars Duening) - (smalloc.c) _large_malloc() didn't compile in SMALLOC_LPC_TRACE due to a missing 'const' specifier. (thanks, Freaky!) 02-May-1999 (Lars Duening) (3.2.7-dev.85) - (backend.c) time_of_ref is fudged only after a reset of a swapped object, not after a cleanup. Also, if the cleanup hook is a closure, the time_of_ref is updated manually to make sure that the cleanup won't occur in every backend loop. - (simulate.c) lookfor_object() sets the time_of_ref of an object if it had to swap it in. This is useful for blueprints for inheritance and cloning, which would thrash otherwise. (thanks, Michael!) - (lex.c) The test against recursive calls in _expand_define() crashed the driver if an error occured during the macro expansion and the error handling itself called macro expansion (e.g. through the _expand_define_() efun). For now, non-reentrant recursive calls are allowed again. 30-Apr-1999 (Lars Duening) (3.2.7-dev.84) - (backend.c) If reset or cleanup cause an object to be swapped in, the time_of_ref is fudged so that the object will swap out again pretty soon, but not immediately. - (hosts/be/Makefile) Re-introduced cross-compiling support for x86 on PowerPC. 29-Apr-1999 (Lars Duening) - (doc/efun/caller_stack_depth, HISTORY) Corrected two typos (thanks, Zora!). 28-Apr-1999 (Lars Duening) (3.2.7-dev.83) - (backend.h, closure.c, interpret.h, efuns.c) Move error_recover_info definition into backend.h, adapting the other files, and added a missing include to efuns.c - (interpret, exec.h, datatypes.h) Began commenting. 27-Apr-1999 (Lars Duening) (3.2.7-dev.82) - (heartbeat.c) 'status tables' could cause a division by 0. Integrated a patch from Michael: - (backend.c) process_objects() now checks if a swap was successful before setting did_swap (since some objects won't swap). - (interpret.c) Added some ASSIGN_EVAL_COST in critical efuns. - (interpret.c) call_lambda() did not update time_of_ref/O_RESET_STATE on calls. Furthermore, objects bound to (s)efun closures were not swapped in, which caused segfaults in dump_trace(). 25-Apr-1999 (Lars Duening) (3.2.7-dev.81) - (backend.c, swap.c) Swapped-out objects are again swapped in for reset. Skipping the call and just setting a due-time in the swapper resulted in too few calls to reset(), especially with small swap times. (thanks, Michael!) - (backend.c) Removed the RESET_DEBUG code sequences. - (interpret.c) Calls to tefuns and vefuns assign the eval costs first. - (sprintf.c) If possible, file:line of an offending sprintf() statement are appended to all error messages, not just to the one printed on stderr. - (etc/icon-*.ppm) The Icon for the Gamedriver. - (hosts/be/{Makefile, icon-*.raw, driver.r, driver.rscr) Added the icon to the compiled resources. - (heartbeat.c) Fixed a crasher which occured when a heart_beat() function managed to remove all remaining objects from the hb list. 19-Apr-1999 (Lars Duening) - (mapping.c) When adding mappings, the second summand was freed before the actual addition (thanks, Daniel!). - (doc/LPC/modifiers) added from TubMud (thanks, Fiona!). 17-Apr-1999 (Lars Duening) (3.2.7-dev.80) - (doc/) Small changes, improved the docs to switch, references and alists (thanks, Freaky!). - (lex.c) Bugfix from Daniel: Variables in closure may start with '_'. - (closure.c) Bugfix from Daniel: #'[<] indexing corrected. - (interpret.c) unmkmapping() also accepts key-only mappings (thanks, Holger!). - (func_spec) Efun caller_stack() was defined to return 'int' instead of the correct 'object *'. (thanks, Robb!). - (driver.h, smalloc, main) Added some 'const' to 'char *' where useful. - (main.c, lex.c) The list of predefs copied from the commandline is now deallocated after the defines have been entered into the table. This removes a spurious 'small block freed' warning on garbage collections. 04-Apr-1999 (Lars Duening) (3.2.7-dev.78-79) - Small changes, just cosmetics. 01-Apr-1999 (Lars Duening) (3.2.7-dev.77) - Manually copied all changes since dev.76 into the Re-Release of 3.2.6. - (call_out.c) Undid the change that callout can call protected functions: this inability was intentional for security reasons. - (backend.c) The consistency check of the object list was still active. Now the code is still there, but deactivated. - (doc/LPC/functions, doc/efun/call_out, doc/efun/call_other) Added a bit documentation regarding protected functions (thanks, Macbeth!). - (interpret.c, doc/LPC/operators) Subtraction allows any width of the subtrahend. 31-Mar-1999 (Lars Duening) - (HISTORY, hosts/be/Makefile) Small corrections. 31-Mar-1999 (Lars Duening) -- Release 3.2.6 - Froze 3.2.6-dev.76 for release. 31-Mar-1999 (Lars Duening) (3.2.6-dev.75) - Changed the directory structure so that the sources are no longer in the top directory. But there's still a long way to go before a simple 'build' in the top directory does everything. - (doc/...) Integrated the 'driverdocs', previously maintained by Stefan Petri (thanks!). 31-Mar-1999 (Lars Duening) (3.2.6-dev.74) - (doc/efun/...) Typo corrections (thanks, Coogan and Freaky!) - (array.c) map_array() did not check if objects in the array were destructed during the execution (thanks, Righ!). (b-990329-0) - (util/make_docs/make_docs.c) Linux-adaption was not complete enough. 30-Mar-1999 (Lars Duening) (3.2.6-dev.73) - (lex.c) *duh* the last change tested for COMPAT, not the correct COMPAT_MODE define. 29-Mar-1999 (Lars Duening) (3.2.6-dev.72) - (lex.c) __FILE__ and __MASTER_OBJECT__ return names without leading slash in compat mode. - (backend.c, interpret.c) Fixed some warnings. - (doc/efun/{sizeof, widthof, load_name}) Cosmetics. - (comm) Undefining ERQ_DEMON did not compile (thanks, Bardioc!). - (prolang.y, object.c) The __INIT function is now protected instead of global (thanks, Freaky!). - (call_out.c) Efun call_out(string,...) can now call protected functions as well (thanks, Freaky!). 28-Mar-1999 (Lars Duening) (3.2.6-dev.71) - (object, simulate, backend.c, interpret.c, efun_defs.c, dumpstat.c smalloc.c) Changed the object list to a double-linked list, allowing destructed objects to be removed immediately. In itself not a big deal, but it allows future, more sorted incarnations of the object list. - (heartbeat.c,h) New module holding the heartbeat related code. - (backend, interpret, simulate) Adapted for using the new heartbeat code. Especially the backend loop now gives preference to the heartbeats over resets/swaps - the latter are slowed down to one reset/swap per cycle if there is not enough time. 26-Mar-1999 (Lars Duening) (3.2.6-dev.70) - (object, simulate, backend.c, interpret.c, efun_defs.c, dumpstat.c smalloc.c) Changed the object list to a double-linked list, allowing destructed objects to be removed immediately. In itself not a big deal, but it allows future, more sorted incarnations of the object list. 26-Mar-1999 (Lars Duening) (3.2.6-dev.69) - (backend.c) The logic in the computation of the average heartbeat stats was wired backwards. - (call_out.c, closure.c, comm.c, interpret.c, lex.c, regexp.c, simulate.c) Removed a couple of warnings when compiling under BeOS/Intel. - (port.c) get_current_time(): 'noted_alarms' is kept up to speed with 'total_alarms'. (thanks, Michael!) - (backend, main, lex.c, interpret.c, simulate.c, func_spec, doc/...) After the current implementation of the free-running alarm, which is needed by the MAX_TIME mechanism, proved unstable on some platforms, the axe fell on this feature altogether. What started out as a nice idea was on the best way to grow into a maze of special cases, all ugly. Sorry, folks. 25-Mar-1999 (Lars Duening) (3.2.6-dev.68) - (backend.c) Corrected a possible division-by-0 in the heart beat status output (thanks, Freaky!) - (efuns.c) Efuns copy() and deep_copy() did copy quoted arrays. (thanks, Michael!). - (simulate.c, otable.c) Warnings in printf again... - (doc/efun/previous_object) Clarified the difference between previous_object() and previous_object(0). 24-Mar-1999 (Lars Duening) (3.2.6-dev.67) - (efuns.c) Efun deep_copy() leaked memory, and also has to copy arrays/mappings with just one ref (thanks, Michael!) - (backend.c) Oops, major fault in the changed efun heart_beat_info(), writing over array limits. - (func_spec, interpret.c, backend.c, config.h, doc/...) The hard time limit is now by default turned off (but can be activated on the command line). The efun get_eval_time() was renamed to remaining_eval_time() to avoid a clash with some mud's driver patches. See f-981229-11 and f-990324-3 for further discussion. - (main.c, simulate.c, port.h) Fixes related to MSDOS_FS: Compiling with Cygwin under/for Windows didn't set the define. legal_path() was a bit too restrictive. Added the define to the list of options printed with '--options'. 24-Mar-1999 (Lars Duening) (3.2.6-dev.66) - (backend) Moved the heartbeat handling into process_objects() as well, which also got rid of the clumsy heartbeat table handling. Only the efun heart_beat_info() is slow now. 24-Mar-1999 (Lars Duening) (3.2.6-dev.65) - (backend, gcollect.c, interpret.c, object, otable, simulate.c) Modified the list of objects so that the head and the end are accessible through pointers. This allowed a better handling of resets/cleanups/swapping such that the processing can be aborted after ALARM_TIME and continued in a later cycle. This should get rid of reset/...-induced lags. As a result, the reset table introduced in dev.62 is no longer necessary, nor is the config.h constant RESET_GRANULARITY. backend.c still contains a sanity check in the main loop, just in case. (f-990323-1) - (otable.c) Removed another printf-related warning. - (main.c, simulate) The name of the master object is assured to be sensible (no leading '/' etc) (thanks, Freaky!). 23-Mar-1999 (Lars Duening) (3.2.6-dev.64) - (comm.c, efuns.c, interpret.c, otable.c) Removed a couple of possible mismatches between format-strings and passed arguments. - (hosts/unix.h, util/*) Several small tweaks for compilation under Solaris 2. Still, the util/Makefiles need manual changes in order to build correctly *sigh* - (otable.c) It is now legal to call rtable_remove() with an object which is not in the reset table - it happens when an object is destructed during its own reset (thanks, Zwirch!). - (interpret.c) If m_allocate() runs out of memory, a NULL 'mapping' svalue was left on the stack -> BOOM. - (prolang.y, simul_efun.c) Replaced bzero() by memset(). - (main.c) The memory-reserve options had one extra 'd' in the long form (thanks, Freaky!). - (backend.c, object.c, otable.c) Made the reset handling resistant against errors during the evaluation of reset(). 22-Mar-1999 (Lars Duening) (3.2.6-dev.63) - (backend.c, otable.c) Small bugfixes. Deactivated, but not deleted the DEBUG printfs. 22-Mar-1999 (Lars Duening) (3.2.6-dev.62) - (config.h, backend.c, otable, interpret.c, simulate.c, swap.c) Implemented new, hopefully less disruptive version of the reset handling. To get the old handling, define OLD_RESET in config.h . Under the new implementation, the objects are sorted by their reset times (but grouped according to RESET_GRANULARITY) which allows easy retrieval of the due objects. Additionally, every backend cycle checks for due objects, but limits the number of calls to whatever can be done in two seconds. After all, resets are not supposed to be exact. For now, there are a number of DEBUG printfs in the modules. (f-990318-0) 21-Mar-1999 (Lars Duening) (3.2.6-dev.61) - (ptrtable.c,h) New module holding a generalized version of the pointer table formerly located in object.c - (dumpstat.c, interpret.c, simul_efun.c, object.c) Modified to use the new pointer table. - (func_spec, efuns, doc/efun/...) New efun copy() for shallow copies, and deep_copy() for deep copies. (f-981229-12) 20-Mar-1999 (Lars Duening) (3.2.6-dev.60) - (func_spec, interpret.c) Efun clonep() now also accepts strings. - (port.h, main.c, comm.c, ed.c, simulate.c) Made sure that MAXPATHLEN is defined and used on all hosts for file name buffers (thanks, Freaky!). - (parse_old.c, prolang.y) Removed a couple of 'unwanted assignment?' warnings. - (lex.c) Reallocations of the defbuf are traced only when '-c' was given on the commandline. - (hosts/be/Makefile) Corrected a small problem in the dependency generation which would lead to incomplete builds. - (pkg/diff) Added the diff code from Ugh and Alfe. - (func_spec, efuns, doc/efun/...) New efun object_info(). - (interpret.c, doc/efun/...) Efun set_next_reset() just returns the time until the next reset when passed 0 as argument. 19-Mar-1999 (Lars Duening) - (alloca.c) Corrected some ugly formatting. - (func_spec, interpret.c, doc/efun/...) Nobody was really fond of 'nameof()', so it's back to 'object_name()'. Well, it was worth a try. - (stralloc, sprintf.c, mudlib/master_skeleton.c doc/master/...) Renamed master apply 'object_name()' to 'printf_obj_name()' due to the clash with the new efun object_name(). 18-Mar-1999 (Lars Duening) (3.2.6-dev.59) - (comm.c, hosts/unix.h) Fixed the broken alarm() and UDP socket handling for Cygwin. - (func_spec, backend, interpret.c, lex.c, config.h, simulate.c, doc/efun/...) Added a time-based watchdog to the interpreter: its default setting is given by MAX_TIME in config.h, it can be specified on the cmdline with '--max-time', the actual value is available through the macro __MAX_EVAL_TIME__, the remaining time can be queried with the efun get_eval_time(). - (func_spec, interpret.c) Efun notify_fail() returns 0, not void. (f-990203-18) - (backend.c) look_for_objects_to_swap(): The call granularity was computed from the config.h macros, not from values set on the command line. - (func_spec, simulate.c, doc/efun/...) Efun set_this_player() also accepts 0 as argument (and then simply clears this_player()). 17-Mar-1999 (Lars Duening) (3.2.6-dev.58) - (mapping.c, otable.c, efuns.c, comm.c) Fixed a couple of warnings sent in by Freaky. - (func_spec, doc/efun/...) clonep() and blueprint() take this_object() as default argument (suggested by Freaky). - (doc/efun/{clonep, assoc, insert_alist}) Small corrections, sent in by Freaky. - (lex.c) Added __MAX_EVAL_COST__ as predefined macro (suggested by Chameloid). - (interpret.c, simulate.c) Added the file to the 'Failed to load' error messages (suggested by Nathan). - (mapping.c, doc/efun/...) Modified map() and filter() so that for 0-width mappings the value 0 is passed as 'data' value. - (func_spec, interpret.c, doc/efun/...) Renamed efuns {filter,map}_mapping() to {filter,map}_indices(). The old names are still available as aliases. - (func_spec, interpret.c) Renamed blueprint() to load_name(), renamed file_name() to nameof() (the old name is still available as alias) and introduced program_name(). 16-Mar-1999 (Lars Duening) (3.2.6-dev.57) - (mudlib/master.c, wk/diff.{c,man}) Wrong lineends. Ooops. 16-Mar-1999 (Lars Duening) (3.2.6-dev.56) - (mapping) Improved the implementation of efun m_rewidth(). - (func_spec, interpret.c, mapping, doc/efun/filter) New efun filter() to filter mappings and arrays. (sort of f-981229-16) - (func_spec, mapping, doc/efun/...) Renamed m_rewidth() to m_reallocate() - this name comes closest. - (func_spec, interpret.c, doc/efun/...) Renamed mapping_contains() to m_contains(). The old name is available as alias. - (util/xerq/socket.c) Changed a wrong datasize in erq_accept(), this should improve handling of TCP sockets (thanks, Sieni!). - (array.c) Added an improvement to the handling of special cases in subtract_array() (thanks, Sunblood!). (p-990316) 15-Mar-1999 (Lars Duening) - (func_spec, mapping, doc/efun/m_rewidth) New efun m_rewidth() to change the width of a mapping. (f-981229-08) - (func_spec, mapping, interpret.c, doc/efun/map) New efun map() to map arrays and full-width mappings. (sort of f-981229-16) 15-Mar-1999 (Lars Duening) (3.2.6-dev.55) - (func_spec, interpret.c, doc/efun/m_values) Changed efun m_values() to return any column of a mapping. (f-981229-07) - (func_spec, interpret.c, doc/efun/widthof) New efun widthof() to return the width of a mapping. (mentioned in f-981228-08) - (func_spec, interpret.c, doc/efun/m_allocate) Renamed efun allocate_mapping() to m_allocate(). - (func_spec, interpret.c, doc/efun/unmkmapping) New efun unmkmapping() to deconstruct a mapping. 14-Mar-1999 (Lars Duening) (3.2.6-dev.54) - (interpret.c, efuns.c) Efuns lower_case(), upper_case(), and capitalize() try to reuse the passed string if possible (suggested by Sunblood). - (interpret.c, func_spec, doc/efun/caller_stack) New efun caller_stack() (suggested by TubMud). 12-Mar-1999 (Lars Duening) - (otable) Commented and ansified. - (port.h, comm.c, simulate.c) __CYGWIN32__ replaced by __CYGWIN__. (b-990204-4) 12-Mar-1999 (Lars Duening) (3.2.6-dev.53) - (object, interpret.c) Finished commenting this module. - (func_spec, interpret.c, efuns) New efuns abs() (thanks, Holger!), upper_case(), all_environment(). (f-990203-20, f-990217-1) 11-Mar-1999 (Lars Duening) (3.2.6-dev.52) - (comm, lex.c) Increased the size of the buffer for the hostname in gethostname() from 20 to MAXHOSTNAMELEN+1 chars. The domainname is now determined once at startup from the hostname, removing the need for getdomainname() (which on some machines returned the NIS domain instead of the DNS domain anyway) (thanks, Freaky!). (p-990309-1,2). - (configure, autoconf/configure.in, all machine.h) Added test and define for <sys/param.h>. In return, the test for getdomainname is gone. - (interpret.c) F_CATCH/F_END_CATCH added the reserve to the eval_cost, but not the assigned_eval_cost. This could lead to amazingly high assigned_eval_cost afterwards. (thanks, Sunblood!) - (func_spec, interpret.c) efun clonep() returns just an int. The name of the blueprint can now be retrieved with the efun blueprint(). - (HISTORY) Change the format a bit. - (efuns.{c,h}) Preliminary file holding all the new efuns. - (func_spec, efuns) New efun make_shared_string() (adapted from TubMud). 10-Mar-1999 (Lars Duening) (3.2.6-dev.51) - (object, dumpstat.c, interpret.c, port.h, simulate.c, exec.h) Started commenting the object module, and adapted the other files. - (doc/efun/*) Improved the formatting of some files. - (prolang.y) Better error message if a local variable is declared again. - (mudlib/master.c) Forgot to change efun308() into set_environment(). - (config.h) SUPPLY_PARSE_COMMAND is on again by default. 07-Mar-1999 (Lars Duening) (3.2.6-dev.50) - (filestat.h) Oops, some compiler really don't like spurious ' even in inactive parts of the source. 07-Mar-1999 (Lars Duening) (3.2.6-dev.49) - (backend.c, swap.c) Added missing #include "random.h" statements. - (parse.c) old_explode_string() was missing when only F_PROCESS_STRING, but not SUPPLY_PARSE_COMMAND was defined. Oops. - (filestat.h) Added #error statement in case somebody defines FILE_STAT. - (config.h) SUPPLY_PARSE_COMMAND is off by default. - (CHANGELOG) Corrected the entry for dev.46. - (interpret.c) m_indices() was found to lack the code for checking the mapping for destructed objects. Somehow it vanished from 3.2-dev.46. - (prolang.y) Cleared a variable-shadowing warning. - (closure.c, comm.c, gcollect.c, interpret, lex.c, make_func.y, mapping.c, object.c, prolang.y, simulate.c, swap.c) Added pragmas for Metrowerks compiler to handle unused variables and intentional assignments in conditions. - (hosts/be/Makefile) Now compiles on highest warning level. - (interpret.h, instrs.h, exec.h, make_func.y) Broke a recursive include loop. 06-Mar-1999 (Lars Duening) (3.2.6-dev.48) - (func_spec, interpret.c, doc/efun/load_object) New efun load_object() to load blueprints. - (func_spec, interpret.c simulate, doc/efun/clone_object) Modified efun clone_object() to accept objects. (f-990305-1) 06-Mar-1999 (Lars Duening) (3.2.6-dev.47) - (simulate.c) With logging of compilations activated, if the compilation of an object is aborted because an inherit is missing, the driver logs 'needs inherit' instead of 'done' on stderr. - (main.c, object, simulate) Extended the object structure with the field 'load_name' which gives (in !compat with leading slash) the filename from which the object resp. a clone's blueprint was compiled. This field is used by the clonep() efun. - (func_spec, interpret.c, doc/efun/clonep) New efun clonep() added. I just don't know how to handle virtual object clones with it. (f-990203-27, sort of) - (main.c, hosts/be/Makefile) Added comments regarding OPCPROF. - (array, interpret.c, parse.c) Since the old explode_string() is now static in parse.c, it was renamed to old_explode_string(). This allows to rename new_explode_string() in array.c to explode_string(). 05-Mar-1999 (Lars Duening) (3.2.6-dev.46) - (mapping, datatyps.h, interpret, array.c) Finished commenting the mappings module. - (parse, interpret.c) F_DESCRIBE was never implemented, so the stubs are gone now, too. - (func_spec, simulate, doc/set_environment) Renamed efun308() to set_environment(). The old name is still available for now (and I think Hyp was the first to propose the new name). (f-981229) - (func_spec, interpret.c, doc/set_next_reset) New efun set_next_reset(). (adapted from Morgengrauen) 03-Mar-1999 (Lars Duening) (3.2.6-dev.45) - (mapping, datatypes.h, array.c, main.c, interpret.c, object.c, swap.c) More comments and cleanup of the the mapping code. - (backend.c) No prompt is printed to interactive users if they are already closing down. This can happen if the current command caused the disconnection, but not destruction of the player object. - (array, interpret.c) Applied Sunbloods optimisation of subtract_array(), which improves the cases 'array - ({})' and 'array - ({ x })' by avoiding copies and alist operations. (p-990303) - (util/Makefile, util/xerq/Makefile) Cosmetics. - (util/erq.c) Applied patches from TubMud, which also introduces two new service types 'ERQ_ACCEPT' and 'ERQ_LOOKUP'. 02-Mar-1999 (Lars Duening) (3.2.6-dev.44) - (mapping, datatypes.h) More comments. - (util/xerq) Added Brian Gersts 'x-erq' demon (formerly 'tw-erq') as an alternative to Amylaar's original erq. So far it's a quick and dirty adaption. Thanks, Brian! - (main.c, simulate.c, gcollect.c) Removed the last FILE_STAT extra code. - (lotsa files): The following features are no longer optional resp. available: (f-981229-01) OLD_PREVIOUS_OBJECT_BEHAVIOUR: gone OLD_EXPLODE_BEHAVIOUR: gone FLOATS: always available TRANSCENDENT_FUNCTIONS: always available NO_XVARARGS: gone MAPPINGS: always available SUPPLY_PARSE_COMMAND is still there... for now. 01-Mar-1999 (Lars Duening) - (prolang.y) Undid most of the patch applied yesterday, because, as Sunblood and Chameloid pointed out, removing duplicate virtual inherits collides with p-990217 and results in sporadic losses of inherited functions *sigh*. However, the bugfix itself is ok and does stay. 28-Feb-1999 (Lars Duening) (3.2.6-dev.43) - (prolang.y) Implemented the patch for the initialisation of virtually inherited variables (thanks, Sunblade!). (p-990227) 28-Feb-1999 (Lars Duening) (3.2.6-dev.42) - (main.c) '--options' didn't print absolute swap paths correctly. - (dumpstat.c, object.h, interpret.c, simulate.c) The evalcosts are now also counted per object (in the variables ticks and gigaticks), and of course listed in the OBJ_DUMP and the debug_info(). (adapted from Morgengrauen) - (filestat.c, all config.h and Makefiles) Deactived the FILE_STAT accounting, but left the filestat.h in place for now - it might come in handy later. - (prolang.y) Variables in variable-only objects are now properly initialized by __INIT() (thanks, Sunblade!). 24-Feb-1999 (Lars Duening) (3.2.6-dev.40) - (random) New implementation using the Mersenne Twister. 22-Feb-1999 (Lars Duening) (3.2.6-dev.39) - (sprintf.c) Increased the size of the buffer for the string to be created from 10KBytes to 64KBytes (suggested by Woody). - (backend.c, swap.c) reset() is not called in objects which are swapped out or which haven't been touched since the last reset(). Also, if an object is swapped in after its due reset time, a new reset time is set. (adapted from Morgengrauen) - (prolang.y) Too big mapping aggregates generate a compile time error. (adapted from Morgengrauen) - (interpret.c) Illegal offsets in get_line_number() no longer crash the driver during a tracedump, but 'just' generate a diagnostic. (adapted from Morgengrauen) - (comm.c, telnet.h) Added telnet options for X Display location and authentification. (adapted from Morgengrauen) 18-Feb-1999 (Lars Duening) - (interpret.c) Bad types to [x..] ranges produce a meaningful error message. 17-Feb-1999 (Lars Duening) (3.2.6-dev.38) - (interpret.c, prolang.y) Applied p-990217 from Sunblood (thanks!), which fixes b-981201 (calls to virtually inherited function lose track where the parents variables are) and b-981208-4 (redefinitions of virtually inherited functions). As a sideeffect, "*":: may now call the same function repeatedly if it is virtually inherited through different parent lines. I don't like this behaviour, but right now I don't grok the compiler well enough to change this. Consider it a bug! - (interpret.c) sizeof(mapping) checks the mapping for destructed objects first (thanks, Holger!) (b-990217-2) 16-Feb-1999 (Lars Duening) (3.2.6-dev.37) - (be/Makefile) the generated executable is tagged with the proper version number (using setversion). - (interpret) sapply_int() and apply_low() take a flag argument, allowing to override any protection by static/protected keywords. This is used in apply_master_ob() to allow the master applies to be static; the earlier solution of setting current_object messed up the previous_object() relation (*doh*!), causing security systems to malfunction. sapply() itself is now a mere #define for sapply_int(). (b-981110-0) 15-Feb-1999 (Lars Duening) (3.2.6-dev.36) - (INSTALL) A few small changes (thanks, Patrick!). - (my-rusage.h) Added an extra check for HPUX; I hope it uses the correct definition of timeval now for RUSAGE_TIME (thanks, Gum!). (b-990203-3) - (prolang.y) Corrected wrap-around problem when programs larger than 64K are compiled (thanks, Daniel!). - (call_out.c, interpret.c) Whenever a callout on an efun closure occurs, current_prog is set to NULL. Should an error occur, the dump_trace() will recognize this value and not try to derefence it (what caused the bugs p-990203-1 and b-990207 in the first place). The previous fix to this problem didn't quite work because it ignored that ob->prog might have been swapped out. However, should more places crop up where current_prog has to be tested for NULL, introducing a fake struct program for efun closures might be the better solution in the long run (as suggested by p-990203-1, second patch). - (interpret.c) Added new function push_valid_ob(object) which pushes the given object onto the vm stack, unless the object is destructed in which case the number 0 is put onto the stack. This function is used in various places to protect apply_master_ob() against destructed objects (esp. calls to privilege_violation(), heart_beat_error(), valid_read() and valid_write()) (thanks, Sunblood!) (b-981226-0). - (interpret.c) apply_master_ob() temporarily sets current_object to the master object, allowing the master functions to be static. 10-Feb-1999 (Lars Duening) (3.2.6-dev.35) - (call_out.c) Callouts did not set up current_prog, this caused the tracedump to crash should an error occur in a callout to an efun closure (thanks, Sunblood and Logic!) (p-990203-1, b-990207). - (call_out.c) Callouts no longer confuse 'command_giver destructed meanwhile' with 'no command_giver to begin with' (thanks, Sissi and Tmm!) (b-990106-0, b-990204-0) - (closure.c) Made #'++/#'-- work (thanks, Zwirch!) (b-990204-3). - (comm.c) During logon, current_interactive is set to the object itself (thanks, Daniel!) (p-990203-3). - (comm.c) When doing I/O, no output is sent to destructed snoopers (thanks, Zwirch!) (b-990204-2). - (Makefile) Updated dependency generation for mkdepend 1.5. - (interpret.c) terminal_colour: Fixed a memory leak of the stack arguments and auxiliary strings - on Tubmud the leak was severe enough to make the GC last several hours (thanks, Sunblood!) (p-990203-0, b-990203-1) - (interpret.c) If adding to an array caused it to hit the upper limit, its memory was freed nevertheless. Subsequent accesses caused all sorts of interesting effects (thanks, Daniel and Zora!) (p-990203-2, b-990203-5). - (interpret.c) terminal_colour() correctly wraps strings of one less the wrap limit (thanks, Chameloid!) (b-990210). - (interpret.c) A call_lambda() on a CLOSURE_IDENTIFIER makes sure that the variable is swapped in (thanks, Sunblood and Matthew!) (b-990203-0, b-990203-6). - (func_spec, interpret.c) to_int(), to_object() and to_float() accept their target data type as argument (thanks, Zora and Zwirch!) (b-980922-0, f-980928-0). - (prolang.y) Applied a patch fixing a potential crasher (thanks, Daniel!) (p-990203-5). - (sprintf.c) Added guards against values < 1.0 to the use of log10 when estimating the number of digits to print (thanks, Chameloid!) 12-Dec-1998 (Lars Duening) (3.2.6-dev.34) - (make_func.y, Makefiles) The file efun_defs.c is written directly my make_func, not just dumped to stdout. The generated lang.y now bears a notice that the file is generated automatically. - (mapping) First comments added. 10-Dec-1998 (Lars Duening) -- Release 3.2.5 - Froze 3.2.5-dev.33 for release. 09-Dec-1998 (Lars Duening) (3.2.5-dev.32) - (make_func.y, func_spec) Commented and ansified. 08-Dec-1998 (Lars Duening) - (lex.*) Commented and ansified. - (comm.c) At the end of a session the driver does no longer send 'Closing down' to the socket. 07-Dec-1998 (Lars Duening) - (simulate) New function check_no_parentdirs() checks a pathname for parent dir specifications (e.g. '/../'). - (sprintf) This is actually an old change, but I forgot to log it at that time: increased the size of an internal buffer used to print float numbers to prevent crashes with big floats. Additionally the code tries to estimate and adjust the size of the output before it prints the float, and tests for a buffer overrun and aborts the driver with a fatal() if it happens (b-980420). 06-Dec-1998 (Lars Duening) (3.2.5-dev.29) - (port.h) When compiling with the GNU compiler, _GNU_SOURCE is defined. - (swap.c) A seek error when trying to write a (new) block to the swap file is no longer a fatal() error. - (Makefile.in, util/Makefile) A normal make in util/ no longer attempts to install the binaries. To do that, use the target 'install' in util/ or 'install-utils' in the main directory. 04-Dec-1998 (Lars Duening) - (ed.c) Oops, had an embedded \0 character in a help text. - (main.c) The version number printed on --version and --options now distinguishes real releases and development versions. - (simulate.c) legal_path() uses a refined test for '..'-directory specifications, now being able to tell these apart from harmless uses like '..file' or '...' (thanks, Fiona!). 03-Dec-1998 (Lars Duening) (3.2.5-dev.28) - (backend.c) f_regreplace(): reallocation-condition was 'space < 0', instead of 'space <= 0' which could cause memory writes just one byte beyond the allocated block (b-981202-2). - (interpret.c) dump_trace() and get_line_number() lacked the handling of ALIEN_LFUN_CLOSURE calls, signified by a 0-valued pc on the control stack. This caused runtime errors in funcalls to symbol_function()s to crash in the error handling. 02-Dec-1998 (Lars Duening) (3.2.5-dev.27) - (lotsa files) Replaced tab characters with spaces. - (comm.c) Under Linux, a special handler is used to ignore SIGPIPEs, since the provided SIG_IGN is just a one-shot handler (thanks, Blackcat!). 01-Dec-1998 (Lars Duening) (3.2.5-dev.26) - (main.c, ed.c, port.h) Cosmetics. - (simulate.c) Efun command() now returns 0 when the target object or the current object are destructed. Similar, the handling for a string H_MODIFY_COMMAND hook did not check if the command_giver is destructed. If the gamedriver was compiled with -DDEBUG, this caused a fatal error 'apply() on destructed object'. - (interpret.c) call_lambda() increments the refcount of lambda closures while executing their code. This is important for driver hook closures which may first destruct and then reload the master. Without the refcounting, this would delete the bytecode while it is executed (thanks, Sunblade!). - (patchlevel.h) Correcte the automatic generation of release dates. 30-Nov-1998 (Lars Duening) - (closure.c) Compensated a (propable) optimizer error for Linux/gcc. See C_COMPILER_BUGS for details. - (port.h) The replacement define for strrchr() was faulty. 27-Nov-1998 (Lars Duening) (3.2.5-dev.25) - (simulate.c) One line was lost in error(), messing up the command_giver in a heart_beat error. - (interpret.c) Experimented with the TubMud patch to mark efun closures by clearing csp->prog. However, the driver interprets this value as 'bottom of cs stack' and thus gets confused in F_RETURN and propably other places, too. 25-Nov-1998 (Lars Duening) (3.2.5-dev.24) - (comm.*) Commented and converted to standard C. Also removed the portal fragments - they didn't work anyway. - (call_out.*, dumpstat.*, ed.*, gcollect.*) Commented and converted to standard C. Also improved a few of the helpfiles of the editor. - (my-rusage.h, port.*, comm.*, hosts/be/be.h) Resolved some minor issues under BeOS/x86. - (main.c) Fixed a small problem in the option parsing. - (object) The program renumber moved in here from gcollect. 21-Nov-1998 (Lars Duening) - (hosts/fcrypt) Removed. - (comm.c) Renamed from comm1.c 20-Nov-1998 (Lars Duening) - (lotsa files) MSDOS support removed. Parts of the driver wouldn't compile under MSDOS anyway. However, the support for the MSDOS filesystem is still there. 19-Nov-1998 (Lars Duening) - (comm1.c, config.h) The ERQ send buffer size is now configurable in config.h, too. Also, the default size is now 1KByte for better throughput. - (lex.c, main.c) New defines __ERQ_MAX_SEND__ and __ERQ_MAX_REPLY__ if if compiler is defined for ERQ support. Both changes are taken from the x-erq distribution of Brian Gerst. - (driver.h) If CATCH_UDP_PORT is not defined, UDP_SEND is made undefined, too. 18-Nov-1998 (Lars Duening) - (lex.c) Compat mode drivers define __COMPAT_MODE__ (meant to replace COMPAT_FLAG eventually); native mode drivers define __NATIVE_MODE__. Similar, a driver using euids defines __EUIDS__. - (access_check.c) Sometimes the "Can't stat ACCESS_FILE" message was generated for no reason. - (mudlib/test_master.c) Extended the master to use normal add_actions(). 15-Nov-1998 (Lars Duening) - (comm1, sent.h, simulate) Began ansification of comm1.c . Moved the definition of the input_to structure from sent.h to comm1.c - (Copyright, CREDITS) Some cleanup. 11-Nov-1998 (Lars Duening) - (smalloc.c) Ups, srealloc() must not be static when compiling with SBRK_OK, as it implements the realloc() function then. - (interpret.c, simulate.c) Added some #ifdef DEBUG...#endif statements where appropriate. - (gcollect) Variable time_last_gc was not visible when compiling with MALLOC_sysmalloc. 10-Nov-1998 (Lars Duening) - (interpret.c, my-rusage.h, port.h) A few changes to accomodate SunOS4. - (lex.c) Only regular files can be used as include files; before it was possible to include directories. 09-Nov-1998 (Lars Duening) - (lex.c) Replace the alloca()d buffers in _expand_define() with once-xalloc()ed ones. In DEBUG mode, recursive calls to _expand_define() are caught. - (lex.c, patchlevel.h) Non-release versions have a version number of the form '3.2.5-dev.16'. - (interpret.c) Corrected an off-by-one error in the stack handling of mapping_contains() (b-980922-1). 08-Nov-1998 (Lars Duening) - (my-alloca.h) If alloca.c is used, the system alloca files are not included - in most cases they define alloca() to something which is not covered by alloca.c, kind of spoiling the desired effect. - (lex.c) If alloca.c is used, a call to alloca(0) is made after each use of _expand_define(), to keep the memory usage low. - (hosts/be/machine.h, .../Makefile) alloca.c is no longer used. 04-Nov-1998 (Lars Duening) - (access_check.c) If the ACCESS_FILE can't be read, an error message is printed to stderr. - (comm1.c) Sorted out the BeOS blocking socket problem. 03-Nov-1998 (Lars Duening) - (func_spec, interpret.c, prolang.y) F_SWAP and F_BREAK_POINT are compiled in only when the efuns are actually defined. By default, they aren't. - (simulate.c) Removed the remaining debug outputs. - (closure.c) The branch rewriting code, responsible for changing BRANCHes into LBRANCHes when a closure grows big in size, did not correct 16-Bit offsets for the new target addresses of new code. The applied fix solves the problem for UNItopia (using <> as #include specifier), but might not be sufficient for all cases. Therefore, two DEBUG: messages remain to warn if this feature is used. - (hosts/be/*) Adapted for R4/x86. - (*.c) Removed most of the 'Possibly unwanted assigment' warnings. 02-Nov-1998 (Lars Duening) - (simulate) Corrected bugs in make_name_sane() and load_object(). The debug outputs are still in place for now (b-981029-0). 01-Nov-1998 (Lars Duening) - (hosts/be/Makefile) Adapted to BeOS R4. - (simulate) The functions find_object() and find_object2() are merged into one function lookfor_object(). For the old calls the macros get_object() and find_object() are available (note the subtle shift in naming!). This reduces the number of calls to make_name_sane(). The function load_object() is no longer exported. - (simulate) Rewrote the load_object() to use a loop for the iterative compilation process instead of tail recursion. - (patchlevel.h) Automatized the versioning to use PRCS keywords. - (comm1.c) After processing a received UDP packet, the evalution costs are cleared (thanks, Tucita!) (p-981027-0). - (main.c) The argument parser printed the wrong option in one error message (b-980925-1). - (config.h, hosts/*/config.h) Increased the limit for bit strings to 6144 Bits (using 1KB memory). - (interpret.c) Out-of-bounds error message for the bit efuns printed garbage instead of the faulty bitnumber (b-980925-0). 09-Oct-1998 (Lars Duening) - (array.c, backend.c) Added some checks if memory allocations actually succeed. 23-Sep-1998 (Lars Duening) -- Release 3.2.4 - (simulate.c) load_object() The code cleaning up the filename did not check if an object with the 'sane' filename already exists. This lead to duplicate entries in the object table and thus fatal driver aborts. Furthermore, the code carelessly modified the string passed, causing disruptions in the shared string table. - (doc/efun/terminal_colour, doc/efun/query_mud_port) added. 22-Sep-1998 (Lars Duening) - (filestat.*) Oops, the one statistic necessary to detect multiple uses of a file in one command was missing. Corrected that. 21-Sep-1998 (Lars Duening) -- Release 3.2.3 - (array.c) Changed order_alist() to use standard conformant pointer arithmetics. - (backend) Added the statistic 'total_player_commands'. - (filestat.*) Statistics module counting the number and type of file usages (read, write, delete, read for compilation, read for inclusion, read for restore_object, write for save_object). The statistic is enabled by defining FILE_STAT in config.h. The statistic is appended to the file /FILESTAT when the special command 'status files' is given. - (access_check.c, backend.c, comm1.c, dumpstat.c, main.c, gcollect.c, lex.c, main.c, object.c, simulate.c, various Makefiles) Adapted for filestat. - (main.c) Beautified the --options output. - (swap.*, main.c) New commandline option '--swap-compact' to put the swapper in the "compact mode". In this mode, free space in the swapfile is reused immediately; in contrast to the default mode in which the reuse is determined by low/high watermarks. 20-Sep-1998 (Lars Duening) - (interpret.c) F_MAPPING_CONTAINS returned destructed objects since it didn't check the results of get_map_lvalue() properly (thanks, Fiona and Zora!). - (backend.c) Reallocation of the heartbeat array now conforms more to the standard. 14-Sep-1998 (Lars Duening) -- Release 3.2.2 - (array.c) One statement was at the wrong place in order_alist(), causing a malfunction in explode(). - (closure.c) store_case_labels(): another crasher removed, this time two loops assumed that the variables local to their body keep their values between two iterations. - (interpret.c) As in closure.c, loop body variables are moved outside their loops if they are meant to be persistant. - (interpret.c) A misplaced cast broke sscanf(). - (backend.*, hash.*) Beautified and commented. - (main.*) Marked the boolean commandline parameters as such. The gamedriver announces itself as 'LDMud'. - (patchlevel.h) Changed version number from 03.02.1@<patchlevel> to 3.2.<patchlevel> (note the missing leading zeroes!). - (CHANGELOG) renamed from CHANGELOG.OSB. - (HISTORY) new file for a user-oriented summary of the changes. 10-Sep-1998 (Lars Duening) - (alloca.c) Beautified. - (simulate.c) load_object: Extended the sanity checks for filenames a bit: multiple '/' are collapsed into just one (thanks for the idea, Zora!). 05-Sep-1998 (Lars Duening) - (main.c) Oops, small typo in options() (thanks, Woody!). And the short command options are now case sensitive as they should be. - (closure.c) Omission of a seemlingly harmless cast caused the driver to core when compiling a switch. Unfortunately, my simple test_master.c didn't catch this. *sigh* We need a proper testsuite... - (ed.c, interpret.c, lex.c, mapping.c, parse.c, prolang.y, sprintf.c) Checked for and corrected similar casts. - (swap.c) Re-added Macbeth's watermark handling for the reusage of free-space, though for now it's deactivated . - (*.c, *.y) Added some parentheses as suggested by gcc to make the code clearer. - (comm.*, interpret.c) Replace the define MESSAGE_FLUSH by the variable message_flush, keeping gcc happier. - (prolang.y) Bugfix from mubo@silicondream: If a program defined __INIT() but didn't implement it, later inheriting of this program crashed the driver. - (port.h, interpret.*, prolang.*, stralloc.*) New macro LOCAL_INLINE for compilers which can inline a function locally while at the same time offering it for external linkage. 02-Sep-1998 (Lars Duening) - (array.*) Commented and converted function defs to STD-C style. - (array.c) Small improvements to unique_array() (boy, that code is really old!). - (main.c) Some warnings about unused variables prevented. 18-Aug-1998 (Lars Duening) - (datatypes.h) New include file declaring all commonly used datatypes of the virtual machine (svalue, vector, etc). interpret.h keeps the interpreter related declarations only. This also unclutters the struct svalue/smalloc.h relationship a bit. 17-Aug-1998 (Lars Duening) - (main.c) --options prints all the compilation options, --version just the version. --help shortened to just list all commandline options, --longhelp prints an elaborate description of these. 16-Aug-1998 (Lars Duening) - (access_check.*) Commented, converted function defs to ansi style. - The old access control using the ACCESS.DENY file is gone. - (patchlevel.h) The GAME_VERSION define moved from config.h into here. - (main.*) Commented, converted function defs to ansi style. Rewrote the commandline argument parser. Most options are now available in a short and long version; --version prints all sorts of compilation options as well. New option --help. - (driver/invocation) New documentation: driver invocation. 13-Aug-1998 (Lars Duening) - Updated the hosts/*/config.h files. - (hosts/be/be.h) The host dependent init makes sure that the erq is not started automatically, even if the option to do so is set in config.h - (my-rusage.h) new, a wrapper for the rusage related stuff. - (hosts/unix.h) received all the Unix specific portability definitions from driver.h (this includes the cygnus win32 environment). (port.h) received the remaining host dependent definitions, so that (driver.h) can be dedicated to provide global definitions. - (port.*, random.*) Commented, converted function defs to ansi style. 12-Aug-1998 (Lars Duening) - (driver.h) Created from the old lint.h, it defines some macros and includes config.h, machine.h and port.h. The system-dependent stuff will move into port.h eventually. All sources changed to use driver.h instead of lint.h. - (my-alloca.h) new, a wrapper for <alloca.h>. - (cruft/) New directory for code snippets which are no longer useful, but interesting enough to keep around. - (hosts/be/Makefile) New target 'dependall' updates the dependencies in Makefile, Makefile.in, and the Makefiles for Be, Amiga and OS/2. 11-Aug-1998 (Lars Duening) - Compiled the sources with gcc on a high warning level and removed as many warnings as possible. Added some patches for compiling under Win32 with Cygwin32. 09-Aug-1998 (Lars Duening) - 'Modularized' the sources: the exports of every .c file are listed in an associated .h file, and only there. The liberal sprinkling of 'extern' statements is gone. 10-Jul-1998 (Lars Duening) - Release 3.2.1@141-osb.16 - (prolang.y) Fixed a pointer mixup which caused '"*"::' to end in an endless loop. Bugfix provided by Daniel von Dincklage (<vonDincklage@ozet.de> aka Sunblade@sd.mud.de) - Updated the BeOS support: small tweaks to machine.h, the Makefile now supports cross-compilation, a small hack in comm1.c. 04-Jul-1998 (Lars Duening) - (prolang.y) In case of a 'Program too large' compilation error, random memory was freed due to an uninitalized variable. - (mudlib/test_master.c) Added this minimal master for simple driver tests. 17-May-1998 (Lars Duening) - Release 3.2.1@141-osb.14 - (interpret.c) The apply cache is initialized on driver startup, not statically at compilation time. This allows compilers with less capable preprocessors (line length limitations) to use big caches as well. 16-May-1998 and before (Lars Duening) - added rxcache.[ch]: cache for compiled regular expression patterns. - added support for BeOS. - (backend.c) Automatic garbage collections are kept at least 5 minutes apart. Every garbage collection is logged on stdout. - (backend.c) Variables are swapped independantly from program code. - (comm1.c, main.c) Multiple login ports may be used in parallel. The port numbers are given on the commandline, efun query_mud_port() was extended to return the appropriate port number. - (interpret.c) MudOS-efun terminal_colour() added by Mica. - (interpret.c) this_object() in destructed objects does not print a diagnostic anymore. - (lex.c) more informative diagnostics about defbuf allocations. Too big defbufs are deallocated again to conserve memory. - (main.c) reallocation of system reserves prints diagnostics. - (regexp.ch) NSUBEXP may take values different from 10 (OSB uses 50). - (swap.c) Free space is reused immediately. - (stralloc.ch, interpret.c) All the names of the lfuns applied by the driver are stored in the shared string table on startup. Together with some modifications to the apply cache for faster processing of applys on shared strings, this should make a mud a wee bit faster. Also added some comments in these areas (with no guarantee for correctness!). - (simulate.c) 'status tables' prints the number of cache hits both absolute and as percentage.