Version 1.0.0, 5 February 1991
-- First public release.
Version 1.1.0, 8 February 1991
-- recycle(x) now calls x:recycle() before destroying the object.
-- the little-used and functionally-dubious update() function has been removed.
**** NOTE: Technically, this change should make this a major release, since
**** it can break old code that depended on update() existing. In
**** practice, however, only $prog:@update used it (in LambdaCore.db,
**** anyway). Thus, to install this server, you should first @rmverb
**** $prog:@update, since its code would otherwise be syntactically
**** illegal.
-- added renumber() and reset_max_object(), to facilitate automatically
generating new versions of LambdaCore.db from the current LambdaMOO db.
-- made it so that only wizards can set the .name property of players; code in
the database now checks that players keep unique names.
-- added set_player_flag() and boot_player() to enable registration-style
player creation and @toad-ing.
Version 1.1,1, 12 February 1991
-- Much rearrangement of #include's to come as close as possible to conformance
with ANSI C and POSIX.
-- Many changes to improve portability, especially to sites with only 'nearly'
ANSI-compliant compilers.
-- Added backtrace-printing for tasks aborted for running too long.
Version 1.1.2, 13 February 1991
-- Fixed nasty infinite recursion problem on server panic.
-- Fixed the implementation of renumber() to really do the job. It's now much
more expensive (linear in the size of the database rather than constant
time), but at least it's correct. It's only intended for use on a tiny
DB anyway.
Version 1.1.3, 11 March 1991
-- Fixed benign unparsing bug that always parenthesized unary-minus
expressions.
-- Changed move() to insert the newly-moved object as the last item in the
contents list, rather than the first.
-- Fixed a less-than-benign bug in the unparsing of conditional expressions,
where 'x ? y | (z ? w | v)' was being unparsed without the parentheses.
-- Fixed a bug whereby an extra checkpoint timer was created every time the
'.dump' command was used.
-- Added printing the value of 'this' to error backtraces whenever it is
different from the verb location.
-- Fixed bug in propagation of property ownership on object creation.
-- WARNING: I believe that there is a storage leak somewhere in the server. If
so, it is not very fast. With any luck, I'll find it soon. For
safety's sake, check on the size of your server's address space
and reboot the server every few days if it's getting large.
Version 1.2.0, 11 April 1991
-- Fixed bug in parser that made 'after' a reserved word; this was part of an
old (and never completely implemented) alternative syntax for the 'fork'
statement.
-- Added memory_usage() function, to help in tracking down any memory-leak
bugs, among other things.
-- Removed ancient restrictions on players recycling themselves.
-- Changed recycle() so that children of recycled objects are not themselves
recycled. Instead, they are adopted by their former grandparent.
-- Fixed bugs in which errors are returned from move() in certain cases.
-- Removed useless error value E_BADMOVE.
-- Fixed bug in move() that allowed the enterfunc to be called even if the
object being moved was no longer valid.
-- Changed several uses of E_INVIND to E_INVARG; E_INVIND is now only used when
an attempt is made to fetch or store the value of a property or verb from an
invalid object.
-- Changed several functions to improve consistency in the order in which
certain kinds of errors are checked for.
-- Made the allocator keep track of statistics concerning memory usage by type
of object being allocated.
-- Changed queued_tasks() to return the value of 'this' in each task as well.
This new value is returned at the end of the sublist for each task, so old
code should continue to work unless it depended upon the exact length of the
sublists.
-- Changed set_task_perms() not to return E_PERM if the argument is the same as
the current permissions of the verb (i.e., if the change in permissions
would be a no-op).
-- Made tonum() and toobj() more robust in the face of extra spaces before or
after tokens.
Version 1.3.0, 7 June 1991
-- Removed the now-useless `.help' and `.mem' commands, and all code concerning
the also-removed CHECKMALLOC option.
-- Fixed allocation bug in inheritance of 'dobjstr' and its ilk, when the
current value of the variable has the wrong type (e.g., 'dobjstr' is a list
when another verb-call happens).
-- Fixed the timers implementation not to try to free storage inside an
interrupt procedure.
-- Changed move() so that :accept is called even if it's a wizard doing the
moving; of course, the wizards gets to do the move even if :accept returns
false.
-- Added 'off/off of' as a new preposition.
-- Fixed the built-in parser to return #-1 ($nothing) for the empty string,
rather than #-2 ($ambiguous_match).
-- Fixed a few bugs in how the allocator keeps track of types.
-- Fixed two memory leaks:
1) If you returned or aborted from inside an iteration over a list,
the list was never freed.
2) If you killed a task, its associated environment was never freed.
-- Added ticks_left() and seconds_left(), for determining how much longer the
current task will be allowed to run.
-- Reduced the maximum number of seconds for each task from 60 to 15.
-- Removed restriction that the first argument to notify() and boot_player()
must be a valid player object.
-- Removed the .shutdown and .dump commands.
-- Added shutdown() and dump_database() functions to replace their respective
built-in commands. The shutdown() function takes a single string argument
that is printed to all players as their connections are closed.
-- Added output_delimiters() as a way to discover the current PREFIX and SUFFIX
strings on a given connection.
-- Completely rewrote and modularized the network interface (now in
bsd_network.c), the main loop, connection-management, and checkpointing code
(now in server.c), and the task dispatcher (now in tasks.c).
-- Eliminated the notion of `clocks' associated with tasks. All tasks,
including forked tasks, begin execution with a full complement of ticks and
seconds. To make this reasonable, the task dispatcher maintains a separate
queue per player, with command tasks and ready forked tasks intermingled.
Thus, a malicious player who forks many tasks can only clog his own queue.
-- Eliminated the `command burst' heuristics from the dispatcher; this always
seemed like just a patch for having a fair dispatcher anyway.
-- Added a timeout for closing idle un-logged-in connections.
-- Removed the QUIT command; the boot_player() function can now be used for
that purpose. boot_player() now does a normal disconnect, including running
the disfuncs. At the same time, I made it allow players to boot themselves.
-- Because clocks have been eliminated, the lists returned by queued_tasks()
have changed somewhat, though in a mostly-compatible way. The clock ID and
clock ticks fields are both constant now, with the ticks equal to 20,000.
-- ****************** All processing of commands typed by un-logged-in players
*** VITAL NOTE *** has been moved into the database. Every un-logged-in
****************** connection is assigned a unique negative player number.
The functions notify() and boot_player() can be used by wizards to send
text to and to terminate such a connection, respectively, using that player
number as the connection identifier. Each line of input on such a
connection is first parsed into words in the usual way and then these words
are passed as the arguments to #0:do_login_command(). For example, the line
connect Munchkin frebblebit
would result in the following call being made:
#0:do_login_command("connect", "Munchkin", "frebblebit")
If #0:do_login_command returns a valid player object, then the connection is
considered to have logged into that player. When the connection is first
established, the null command is automatically entered, resulting in a call
to #0:do_login_command with no arguments. This signal can be used by the
verb to print out a welcome message, for example; this service is no longer
provided by the server.
Because of this change, it will be **NECESSARY** to add an implementation
of #0:do_login_command to your database *before* beginning to use version
1.3.0 of the server. The Minimal.db database and the version of the
LambdaCore database released concurrently with 1.3.0 already provide such an
implementation. The following code can be used as a simple implementation
that preserves most of the functionality of version 1.2.0:
if (callers())
"This code should only be run as a server task.";
return;
endif
if (args == {})
args = {"help"};
endif
command = args[1];
args = args[2..length(args)];
nargs = length(args);
if (command == "connect")
if (nargs < 1)
notify(player,
"Usage: connect <existing-player-name> <password>");
else
name = args[1];
password = nargs >= 2 ? args[2] | "";
for p in (players())
if (p.name == name
&& (typeof(p.password) != STR
|| (length(p.password) >= 2
&& crypt(password, p.password[1..2]) == p.password)))
p.last_connect_time = time();
return p;
endif
endfor
notify(player, "Unknown player or wrong password.");
endif
elseif (command == "create")
if (nargs != 2)
notify(player, "Usage: create <new-player-name> <new-password>");
else
name = args[1];
password = args[2];
for p in (players())
if (p.name == name)
notify(player, "That player name is already in use.");
return 0;
endif
endfor
new = create($player_class, $nothing);
set_player_flag(new, 1);
new.name = name;
new.aliases = {name};
new.programmer = $player_class.programmer;
new.password = crypt(password);
new.last_connect_time = time();
move(new, $player_start);
return new;
endif
elseif (strcmp(command, "QUIT") == 0)
boot_player(player);
else
msg = $welcome_message;
if (typeof(msg) != LIST)
msg = {msg};
endif
for line in (msg)
if (typeof(line) == STR)
notify(player, line);
endif
endfor
endif
return 0;
Of course, the whole point of bringing this code into the database is to
enable easy experimentation with new facilities for un-logged-in users, such
as multiple Guest players, friendlier coping with similarly-named players,
provision of a `who' command, etc.
-- The network interface now copes more nicely with running out of buffer
space for output to a player. Before, excess output was simply dropped on
the floor; now, it instead attempts to push it out the network socket and
only flushes output when that doesn't work. I was able to have a single
command send me over 60,000 characters of output without flushing any.
-- The PREFIX and SUFFIX strings are no longer printed either for the .program
command or for any of the lines of program input.
-- Recycled players with active connections are more promptly noticed and their
connections closed.
-- The numeric ID associated with forked tasks is now guaranteed to be the same
as the value of task_id() in that task when it executes.
-- The create() function now invokes the :initialize verb on the newly-created
object before returning it. It is not an error if there is no :initialize
verb defined on the object.
-- Changed listinsert() and listappend() never to generate E_RANGE errors.
Instead, the following expressions are now always equivalent:
listinsert(list, element, index)
listappend(list, element, index - 1)
{@list[1..index - 1], element, @list[index..length(list)]}
-- Changed the matching algorithm for verb names so that the old behavior of
`*' matching anything is generalized to `foo*' matching anything beginning
with `foo'.
-- Changed eval() and set_verb_code() to require programmer permissions.
-- Added a timezone abbreviation onto the end of ctime()'s result.
-- Added min(), max(), and abs() functions.
-- The messages printed for running out of ticks and for running out of seconds
are now disinguishable.
Version 1.4.0, 22 August 1991
-- Fixed a bug whereby running out of seconds might be misreported as running
out of ticks.
-- Fixed a bug whereby using `@' on a non-list in a verb's argument list could
end up passing E_TYPE as the value of `args'.
-- Fixed parsing bug that would ignore an extraneous comma at the front of an
argument list.
-- Changed random() to raise E_INVARG on non-positive numerical arguments,
rather than E_TYPE.
-- Fixed bug in handling of the case where a verb disappears between a
`.program' and the corresponding `.'.
-- Made it a type error to use <, <=, >, or >= either on operands of different
types or on lists.
-- Completely rewrote the MOO-code parser, unparser, and interpreter using a
new program representation based on vectors of bytecodes instead of abstract
syntax trees. This saves space and time and also allows the interpreter to
be iterative instead of recursive. This, in turn, allows us to implement
various operations that involve saving the entire state of a running task
for later resumption.
-- Added read() and suspend() functions, using the new capabilities of the
interpreter.
-- The new interpreter counts a few more ticks than the old one; in particular,
every IF, ELSEIF, and FORK now counts a tick. In view of this, and in view
of the improved performance of the interpreter, the ticks limit for tasks
has been increased from 20,000 to 30,000.
-- Added an indexed assignment statement for updating lists stored in variables
and properties.
-- Fixed bug in numeric comparisons where `-2147483647 > 1' was false but
`-2147483647 > 2' was true.
-- Fixed database dumping to detect and recover from filesystem errors. For
normal periodic checkpoints, the dump attempt is simply abandoned. For
panic and shutdown dumps, the server retries the dump from the beginning
after waiting 60 seconds (during which time, presumably, the archwizard will
fix whatever the problem was). In all cases, explanatory messages are
printed in the log. This should help mitigate the problems that arise when
there's no more room left on the filesystem on which the server is dumping.
-- Enhanced the callers() function to return the verb-location and initial
value of `player' for each caller.
Version 1.4.5, 20 September 1991
-- Made unparser failures panic the server rather than try to press on.
-- Changed the task scheduler in two important ways:
1) Effectively gave each player's input tasks a higher priority than
their forked or suspended tasks. This should both improve
interactive response and make it possible for users to recover from
accidentally forking off many tasks; they can now run commands to
kill those tasks and avoid having those commands wait behind all of
the forked tasks.
2) The scheduler no longer uses a strictly round-robin algorithm for
choosing which player's task to run next. Instead, it keeps track
of how many seconds have been consumed by each player's tasks and
runs the next task from the queue with the least total `usage'.
Fresh queues and those queues that are empty when it becomes their
turn to run again are kept apart from non-empty queues. When an
empty queue next gets a task to run, its usage is set equal to the
least usage of already-non-empty queues. Thus, such queues go to
the head of the list but have no advantage over the queues that were
never empty. This more complex scheduling scheme is intended to
ameliorate the problems with some players having long-running tasks
constantly runnable in the background; such players will now get
only their fair share of the server.
-- Fixed queued_tasks() to list the player being read from as the owner of a
reading task. This makes it consistent with kill_task(), which only allows
that player to kill such a task.
-- Fixed a bug with my use of signal() on some POSIX-compliant machines. One
possible consequence of the bug (actually manifest on the IRIS) is that all
checkpointing processes after the first one will never be waited for,
leaving `defunct' processes lying around until the server itself terminates.
-- Fixed a small bug in my use of ANSI C: the `typedef enum Opcode Opcode;'
line in opcode.h must come *after* the definition of `enum Opcode'; unlike
struct types, enum types may not be forwardly referenced. Thanks to yduJ
and the Lucid C compiler for finding this bug.
-- Fixed a bug in the line-number-finding code; this bug was identical to one
we fixed long ago in the unparser. Since the LNF code is *derived* from the
unparsing code, this is unsurprising. What's embarassing is that we forgot
to fix the LNF code at the same time...
Version 1.5.0, 3 November 1991
-- Added connection_name(player) to allow wizards some level of reasonable
control over whom they allow to connect to their servers.
-- Changed read() to accept an optional argument, player, specifying the player
from whose connection to read. If this argument is given, the check that
the current task is an input task is not done. The only good use for this
that I'm aware of is in conjunction with the next change...
-- The new compilation option OUTBOUND_NETWORK enables a new built-in function
called open_network_connection(). The arguments are specific to the network
module in use, but the intent is that they somehow specify a remote place to
which a network connection can be made. For the default, BSD networking
module, there are two arguments, a host-name string and a port number. The
function creates a connection to that place, if possible, and tricks the
rest of the server into believing that a normal connection just came from
that place. Open_network_connection() returns the negative object number
that is the pseudo-player object for that connection. One can then use
read() and notify() to send and receive lines of ASCII text on that
connection. Of course, if the server should ever receive a line of input
when there's no task read()ing from the connection, that input will be
treated as a command in the usual way. Also, unless the connection `logs
in' at some point, the usual five-minute timeout-if-no-input-received
connection shutdown will take place. The usual practice is to have the same
task that opens the connection go into an infinite loop read()ing from it.
As usual, boot_player() can be used to shut down the connection. This
function can only be called by wizards, of course. If the OUTBOUND_NETWORK
option is not used, then open_network_connection() still exists, but always
raises E_PERM.
-- Changed the bsd_network module to map tab to space on input, rather than
simply dropping tabs on the floor.
Version 1.5.1, 12 November 1991
-- Fixed bug in checkpointing whereby the server's command log was flushed even
if the checkpoint failed.
-- Various minor tweaks to make the server work on an IBM RS/6000.
Version 1.5.9, 6 December 1991
-- Gave different ticks/seconds limits to non-input tasks. They get only 15,000
ticks and 5 seconds.
-- The task seconds limit is now measured in server CPU seconds (as opposed to
wall-clock seconds) on systems that can support it.
-- Adding some simple object-hierarchy checking and (in very simple cases)
repair code to the DB reading process. If unfixable errors are found, the
server refuses to run.
-- Fixed minor inconsistency in the log format.
-- Fixed a few minor memory leaks.
-- Fixed a bug whereby the bottom frame of a suspended task that was read from
the database file could be given the wrong size runtime value stack.
-- Made the DB-reading code a bit pickier about the correctness of its input.
-- Tweaked create() to make it more properly initialize the new object's
location and contents.
-- Fixed bug in move() that could be tickled if an `accept' verb recycled
either the movee or the destination.
-- Fixed several bugs in recycle() that could (and did!) corrupt the contents
hierarchy in the database. The implementation is now much simpler, too!
Version 1.5.14, 3 March 1992
-- Various minor tweaks to make the server work on a Macintosh running A/UX.
-- Tweak to work around bug in isgraph() on systems that default to signed
characters.
-- Fixed bug in boot_player() that didn't allow players to boot themselves.
[I know it says under release 1.3.0 above that boot_players() already worked
this way, but as far as I can see, that was a lie!]
-- Fixed bsd_network startup sequence so that the server does not `listen' on
its port until after the DB has been read. Connections during the
DB-reading phase will most likely time out. I think this is better than the
old situation, in which those connections succeeded and then hung without
printing anything until the server finished reading in the DB.
-- Changed behavior of server when there aren't enough file descriptors to
accept a new connection. The server now arranges to stop just short of this
point and to immediately close any new connections after printing an
explanatory message.
-- Fixed properties(), verbs(), x.contents, and queued_tasks() to run in time
linear in the length of their results, rather than quadratic.
Version 1.5.20, 24 July 1992
-- Added check for an empty string as the second argument to strsub(); this now
raises E_INVARG.
-- Conditionalized the keeping of the command log; it is now controlled by the
LOG_COMMANDS symbol.
-- Removed a number of causes for GCC 2.1 to issue warnings about server code.
(Version 1.5.15)
-- Fixed an ordering bug whereby the task queue for a newly-connected player
was not fully initialized when the :confunc task(s) were run. In
particular, the new, connected queue was not marked as belonging to that
player, and certain functions like output_delimiters() would always fail
(not that that's a very interesting function to call in a :confunc, since
it's guaranteed to return {"", ""} then...). (Version 1.5.16)
-- Finally fixed the bug whereby it was possible to use chparent() to create a
situation in which an object defines a property with the same name as one
defined by one of its ancestors. Such an attempt now results in chparent()
raising E_INVARG. (Version 1.5.17)
-- Finally fixed the bug whereby add_property() allowed the definition of a
property in a parent that was already defined in a child. With this change,
it should no longer be possible to create a situation in which a parent and
child define properties with the same name. (Version 1.5.18)
-- Eliminated some minor warnings from the Dell SVR4 and HP/UX 9000 compilers.
(Version 1.5.19)
-- Added logging of all instances of x.wizard = 1.
-- The very first task run on booting the server is now a server task calling
#0:server_started() with player == #-1. This task runs even before the
server gets the value of $dump_interval in order to schedule the first
checkpoint. (Version 1.5.20)
Version 1.6.0, 13 August 1992
-- Changed MOO-code parser to generate the old version 1.3 program-
representation based on abstract syntax trees. These syntax trees are then
translated to vectors of bytecodes. This is much cleaner and easier to
understand than the old method; perhaps surprisingly, it also appears to be
significantly faster, too. (Version 1.5.21)
-- Added support for compilation on MIPS SVR3 machines.
-- Added support for a remote `checkpoint-request' signal (SIGUSR2), analogous
to the existing `shutdown-request' signal (SIGUSR1).
-- Changed unparser to first convert bytecodes to trees and then use output
routines from version 1.3 of the server. Same for line numbers.
-- Error tracebacks now include line numbers for all frames on stack.
(Version 1.5.24)
-- Cleaned up interpreter code.
-- Added sqrt() and server_log() built in functions (Version 1.5.25)
-- All built-in function registration procedures are now located in one list,
so it is easier to add new ones.
-- Added match(), rmatch(), and substitute() built-in functions to handle
regular expression searches. (Version 1.6.0)
Version 1.6.1, 14 August 1992
-- Fixed off-by-one bug in GNU regexp searching code. (*sigh*)
Version 1.7.0, 23 October 1992
-- Fixed behavior of index(), rindex(), match(), and rmatch() on empty pattern
and/or subject strings. (Version 1.6.3)
-- Added the prefix `> ' to all log messages generated by the server_log()
built-in function, so that they can be distinguished from server-generated
messages.
-- Added a compile-time option (OUT_OF_BAND_PREFIX) enabling a method of
entering commands that bypass both normal command parsing and any pending
read()ing task. If OUT_OF_BAND_PREFIX is #define'd (in config.h) as a
non-empty string, then any lines of player input that begin with that prefix
will be parsed into a list of words and those words passed as arguments in a
server task invoking #0:do_out_of_band_command. This is intended for use by
fancy MOO clients that need to send reliably-understood messages to the
server, such as window-event notifications.
-- Added output to the `.program' built-in command, stating that it is obsolete
and will be replaced `soon' (i.e., in the next release).
-- Rearranged the per-system configuration procedure for compiling the server;
now it should be easier for people to understand just what needs to be done
for their particular system. The user changes in the Makefile are now
confined to specfying details of the compiler; all other kinds of options
are handled by editing the `config.h' file, which has been radically
reorganized.
-- Renamed a few files (notably parse_command.[ch]) to fit within the
14-character limit imposed on certain systems. (Version 1.6.4)
-- Implemented reference counting on string and list values, replacing the old,
CPU-intensive deep copying method of storage management.
-- Added subrange assignment for strings and lists. Added indexed assignment
for strings. (Version 1.6.5)
-- Fixed bug whereby a read()ing task with no more input to consume was never
resumed if the connection being read was subsequently closed, from either
side. Now the read() call raises E_INVARG, just as it would if the read()
were begun when the connection was already closed and no more input was left
to consume.
-- Added TYPE_CLEAR value to properties, which cause property lookup on the
parent. Added built in functions `clear_property()' and
`is_clear_property()' to assist in TYPE_CLEAR property manipulation.
-- Added hash-lookup for properties and changed property definition
representation to be arrays instead of linked-lists, both to speed up
property lookup. (1.6.6)
-- Added support for multiple complete networking implementations. The first
use of this flexibility is a new SINGLE_USER option, which creates a version
of the server that accepts only one connection at a time and uses the
server's own standard input and output streams for it.
-- Added a new built-in property on objects, the `f' (for `fertile') bit; it
replaces the use of the `r' bit to allow children to be made of the object
by either create() or chparent(). That is, now those operations check that
the `f' bit is set and disregard the setting of the `r' bit. (Version 1.6.7)
******** Before upgrading an existing MOO to use this version of the server,
* NOTE * you should make sure that no object in your database already has a
******** property named `f'; the following MOO program, run by a wizard,
will perform this check:
for i in [0..tonum(max_object())]
o = toobj(i);
if (ticks_left() < 1000 || seconds_left() < 2)
notify(player, tostr("Checking ", o, " ..."));
suspend(0);
endif
if (valid(o) && "f" in properties(o))
notify(player, tostr("*** ", o.name, " (", o,
") has an `f' property!"));
endif
endfor
After eliminating all such properties from the database and restarting with
this version of the server, you should, as a wizard, run the following MOO
code to initialize the `f' bits of all of the objects:
for i in [0..tonum(max_object())]
o = toobj(i);
if (ticks_left() < 1000 || seconds_left() < 2)
notify(player, tostr("Fixing ", o, ".f ..."));
suspend(0);
endif
if (valid(o))
o.f = o.r;
endif
endfor
-- Reorganized the files so that all database modifying procedures were in one
of four modules. (Version 1.6.8)
-- Reorganized existing network protocol and multiplexing wait implementations
into a pluggable modular form. (Version 1.6.9)
-- Fixed bug in the interpreter that could pass a garbage program counter to
the line-number-finding code, used in printing error tracebacks. The l-n-f
code responded semi-robustly by printing a message in the server log and
returning a line number of zero. (Version 1.6.10)
-- Fixed bug whereby passing negative numbers to random() failed to evoke an
E_INVARG error.
-- Fixed a bug in MOO-code compilation that led to real nastiness if there were
more than 255 literals in a single verb.
-- Incorporated several new networking implementations, allowing for use by a
single-user (using the standard input and output streams of the server
itself), and by multiple users on either System V or BSD-style UNIX systems,
either with or without TCP/IP networking. (Version 1.6.11)
-- Fixed a bug in built-in function management that caused max() and min() to
always raise E_TYPE on some systems.
-- Added a new automatic configuration system to the server distribution, so
that people do not, in general, need to know much of anything about their
local system in order to compile the server. (Version 1.6.12)
-- Fixed a misfeature of substitute whereby it was not possible to include a
percent-sign in the string that was to survive into the output. Now, `%%'
is replaced by `%' in the output. (Version 1.7.0)
Version 1.7.2, 3 August 1993
-- Made tabs input as themselves, rather than being mapped to spaces. This is
necessary in order to communicate with Gopher servers from within the MOO.
-- Reduced the default `seconds-left' limit for task; forground tasks went from
15 to 5 seconds and background tasks from 5 to 3.
-- Fixed regular-expression matching to respect task seconds limits, aborting
the match.
-- Fixed bugs whereby both kill_task() and queued_tasks() would miss reading
tasks that had no ready input (i.e., whose associated task queues were
empty).
-- Fixed memory leak when a read() call fails with E_INVARG; the copied
interpreter stack was never being freed.
-- Fixed possible free of the null pointer during object recycling.
-- Fixed a number of places that generated warnings in newer versions of GCC.
-- Fixed inconsistency of the behavior of strcmp() between different server
machines; now it always returns one of 1, 0, or -1.
-- Added a five-second timeout to outbound connection attempts, aborting them
if the timer expires.
-- Made the error message to users about network buffers overflowing and lines
of output being flushed much more understandable.
-- Added log entry for each time the server refuses a connection due to being
full.
-- Fixed premature-free bug in preposition matching.
-- Bowing to popular pressure, removed warning from `.program' command about
its (no longer) imminent demise.
-- Added more error checks to options.h to make it easier for installers to
diagnose problems.
-- Fixed a bunch of auto-configuration problems encountered by installers.
-- Fixed a bug in substitute() where certain successful match() results were
rejected as invalid.
-- Removed restrictions that only player objects could be used in place of `x'
in the following:
object.owner = x
add_property(object, pname, value, {x, perms})
set_property_info(object, pname, {x, perms})
add_verb(object, {x, perms, names}, args)
set_verb_info(object, vname, {x, perms, names})
set_task_perms(x)
That is, non-players may now own objects, verbs, and properties, and running
tasks may take on the authority of a non-player.
-- Fixed bug where it was possible for a task to run out of seconds during a
call to set_verb_code() and have a truncated version of the program
installed as the new verb code. Now, either all of it gets installed or
none of it.
-- The log messages produced by successful open_network_connection() calls are
more useful now, containing the hostname and port of the connection.
-- Outbound network connections are no longer susceptible to connection
timeouts due to a lack of input.
-- Fixed a bug whereby verbs with numeric names screwed up DB loading, with the
wrong code getting put into some verbs and others getting no code at all.
Version 1.7.3, 4 August 1993
-- Outbound connections no longer get a spurious blank line of input as soon as
they're created.
-- Fixed stupid bug introduced in fixing bug in substitute().
-- Fixed some bugs in the Makefile and the configuration script.
Version 1.7.4, 10 August 1993
-- Fixed more configuration problems and warnings from GCC.
Version 1.7.5, 12 August 1993
-- Added more log entries during loading, to make it clearer where the time is
being spent.
-- Fixed a long-standing denial-of-service attack vulnerability, in which a
connection could keep the server from ever running any tasks at all by
*always* providing ready input.
Version 1.7.6, 11 October 1993
-- The built-in command parser now matches the (in)direct object string against
the names of objects as well as their aliases.
-- The server no longer calls :confunc or :disfunc directly when users connect
or disconnect; instead, it calls the following verbs on #0 in the
appropriate circumstances:
:user_created(USER)
When #0:do_login_command() returns USER, a valid player object
whose number is greater than the value max_object() returned
before the server called #0:do_login_command(). That is, when
a new user has been created.
:user_connected(USER)
When #0:do_login_command() returns USER, a previously-existing
valid player object for which no active connection already
existed.
:user_reconnected(USER)
When #0:do_login_command() returns USER, a previously-existing
valid player object for which there was already an active
connection.
:user_disconnected(USER)
After the end of a task that called either boot_player(USER) or
set_player_flag(USER, 0).
:user_client_disconnected(USER)
When USER's client unilaterally closes its connection to the
server.
******** Before upgrading an existing MOO to use this version of the server,
* NOTE * you should install the following verbs on #0 in order to remain
******** compatible with existing code; NOTE WELL that they should be
installed with the `d' permissions bit UNSET:
#0:"user_created user_connected"
user = args[1];
fork (0)
user:confunc();
endfork
user.location:confunc(user);
#0:"user_disconnected user_client_disconnected"
user = args[1];
fork (0)
user.location:disfunc(user);
endfork
user:disfunc();
-- Non-wizard users can now use the read() built-in function when they give a
connected object they own as an argument. Thus, it is possible, for
example, for a non-wizard to prompt itself for input.
-- The match() and rmatch() built-in functions now cache some number of the
most recently used patterns, somewhat speeding up the matching process.
-- The server now calls #0:checkpoint_started() whenever it starts to dump a
checkpoint of the database and #0:checkpoint_finished(SUCCESS) whenever the
dump finishes, where SUCCESS is true iff and only if the checkpoint was
successful.
-- Fixed a bug (!) in the built-in command parser whereby in the command
="This is in quotes" and this is not
the variable `verb' would correctly be set to "=This is in quotes" but the
variable `argstr' would be "is in quotes\" and this is not" instead of the
correct "and this is not". (Reported by Shirgall.)
-- Fixed some compilation problems with the SYSV/TCP configuration on Solaris
machines.
-- The value of `argstr' when the server calls either #0:do_login_command() or
#0:do_out_of_band_command() is now the raw command line as received from the
connection, as opposed to the empty string.
Version 1.7.7, 20 December 1993
-- Added some extra configuration tests and other changes to work around bugs
in HP/UX.
-- Fixed an output-timing bug in calling the #0:checkpoint_started() hook.
-- Fixed bug in the SYSV/LOCAL client whereby some output from the server could
be lost, never being printed.
-- Added the server FIFO file name to an error message in the SYSV/LOCAL
configuration.
-- Added the first rudimentary support for in-DB command parsing. Each user
command is broken up into words, a list of which is passed as the arguments
in a call to #0:do_command(), if it exists, with `argstr' initialized to the
raw command line typed by the user. If #0:do_command() does not exist, or
if that verb-call completes normally (i.e., without suspending or aborting)
and returns a false value, then the built-in command parser is invoked as
usual to handle the command. Otherwise, it is assumed that the DB code
handled the command completely and no further action is taken by the server
for that command.
Version 1.7.8, 30 September 1994
-- Many serious bugs in the MOO-code compiler (some server-crashing) were fixed
by rewriting the code-generator portion from scratch. This should fix the
problems some folks have had with panics due to `Unparse failed!'.
-- As a side-effect of the support I wrote for debugging the new code
generator, there is a new built-in function, disassemble(OBJ, VERB), which
returns a (longish) list of strings giving the `assembly-language' listing
for the compiled code of the verb VERB on the object OBJ. The permissions
checks are the same as for verb_code(OBJ, VERB). I don't make any promises
about how that output might change in later releases and I probably won't be
writing any detailed documentation of the opcode set, etc. Think of it as a
little puzzle gift, from me to you... :-)
-- Added a number of new configuration tests, especially to aid compilation on
the DEC Alpha running OSF/1 and any machine running NeXT Mach 3.1.
-- The checkpoint process now changes how its command line shows up in the
output of the `ps' command (on systems where that's possible), to make it
clear what process it is.
-- Added the first stages of a general new facility for customizing the
operation of the server dynamically, from inside the database. At a number
of times, the server checks for whether the property $server_options exists
and has an object number as its value. If so, then the server looks for a
variety of other properties on that $server_options object and, if they
exist, uses their values to control how the server operates. There are only
a couple of uses of this facility in 1.7.8; many more are expected in the
future.
-- As the first use of the above facility, the server uses the value of
$server_options.fg_ticks and $server_options.fg_seconds (if they exist and
are numbers) as the numbers of ticks and seconds allotted to command tasks
and server tasks; the values of DEFAULT_FG_TICKS and DEFAULT_FG_SECONDS in
options.h are used if the corresponding properties either don't exist or are
not numbers. Similarly, the values of $server_options.bg_ticks and
$server_options.bg_seconds are used for forked tasks and suspended tasks,
falling back on DEFAULT_BG_TICKS and DEFAULT_BG_SECONDS. These properties
are looked up anew every time a task begins or resumes execution.
-- As the second use of the $server_options facility, on every call to a
built-in function `foo()', if the property $server_options.protect_foo
exists and is true, and the programmer is not a wizard, then E_PERM is
raised. Thus, for example, to make the `chparent()' function wiz-only,
simply set $server_options.protect_chparent to 1.
-- Fixed a bug in the MOO-code unparser where any use of a string naming a MOO
keyword as a property or verb name would be unparsed in such a way as to
make the resulting code syntactically incorrect. Thus, for example, the
expression
x.("while")
now unparses into the same code, instead of into
x.while
which won't parse properly.
-- The server now uses a pair of additional (smallish) UNIX processes to
perform all network host-name lookups. Under the new system, it can
reliably abort stalled lookups without causing problems; in particular, this
provides the real fix for the long-standing `VM exhausted' problem on NeXT
machines.
-- A few small storage leaks were fixed.
-- A number of new progress-report log messages were added to the DB loading
process.
Version 1.7.8p1, 3 October 1994
-- Fixed a file-descriptor leak in the new name-lookup code; if the server has
enough name-lookup timeouts, killing the name-lookup worker process over and
over, then the name-lookup intermediary will eventually run out of file
descriptors and fail, shutting down name-lookup services until the server is
rebooted.
-- Fixed a minor portability problem in the new code generator; believe it or
not, there are still C compiler's out there (on Ultrix, for example) that
don't implement initializers for locally-declared structures...
-- Fixed a bug in disassemble() whereby it couldn't find !x verbs.
Version 1.7.8p2, 6 October 1994
-- Fixed another little bug in disassemble(), whereby it could crash the server
if you try to disassemble an unprogrammed verb.
-- Added a hard minimum of 100 ticks and 1 second for all tasks, to prevent
disaster when somewiz somewhere blows the values of $server_options.fg_ticks
and company.
-- Fixed a bug whereby the new name-lookup code would fail to recognize even
dotted-decimal numeric addresses once the intermediary was presumed dead.
-- Fixed a null-pointer free in the disassembler that was crashing servers that
were using the GNU malloc() implementation.
-- Added Craig Horman's patch to the regex.c code for the Alpha, which avoids
occasional random failures of the matching code. I don't know why it took
me so long to apply this...
Version 1.7.8p3, 19 October 1994
-- Fixed a typo in the "hard minimum" code added in 1.7.8p2; the minimum
seconds allocation is now really 1, not 2...
-- Fixed a bug in the disassembler wherein it could produce garbage or even
possibly crash the server if the C compiler decided to evaluate certain sets
of function arguments in other than left-to-right order.
-- Fixed a bug in is_clear_property() that could attempt to access an
uninitialized variable. (Thanks to Quowong P Liu for pointing out this and
others of the bugs fixed in the various 1.7.8 patch releases.)
-- Fixed a bug whereby the server was, on some systems, always susceptible to
being shut down prematurely by the `hangup' signal. Now, the `restart'
script uses `nohup' and the server continues to ignore SIGHUP if it was
started ignoring it (i.e., by `nohup').
-- Fixed MOO-level tracebacks to include lines for built-in functions that call
verbs, such as eval(), move(), etc.
-- Added MOO-level traceback printing for all server panics that happen during
the execution of a MOO task.
Version 1.7.8p4, 2 November 1994
-- Fixed minor memory leak in disassemble().
-- Fixed bug whereby verb_info() and friends allowed an argument of "" to
always match the first verb on an object.
-- Fixed bug whereby MOO keywords (e.g., `if', `E_PERM', `return', etc.) were
case-sensitive.
-- Fixed (I hope...) the 32-bit assumptions that break on the (64-bit) Alpha.
Version 1.7.9alpha1, 25 September 1995
-- Fixed a few minor configuration and portability problems.
-- Fixed a potentially server-crashing bug in subrange assignments.
-- Renamed the internal server function `log()', to avoid name conflicts with
the logarithm function in the math library.
-- Changed the name-lookup subsystem to be more optimistic about the chances of
eventually recovering from an earlier failure to restart the lookup process.
Also bullet-proofed it against problems with its read() calls getting
interrupted by the checkpoint timer.
-- Added a paragraph to `README' explaining how to boost the limit on the
number of connections a server can support.
-- Added two new built-in functions enabling faster case-sensitive tests:
equal(X, Y) returns true iff the values X and Y are completely equal,
including the case of any strings they might contain; this is just
a case-sensitive version of the `==' expression.
is_member(X, L) is a similarly case-sensitive version of `X in L'.
-- Made the `.program' built-in command wiz-only if
$server_options.protect_set_verb_code exists and is true.
-- Added the built-in function `set_connection_option(CONN, OPTION, VALUE)',
for controlling various optional behaviors on the connection CONN. The only
allowed values for OPTION in this release are as follows:
"hold-input" -- if VALUE is true, then input received on CONN will
never be treated as a command; instead, it will remain in the queue
until retrieved by a call to read().
"client-echo" -- (NP_TCP configurations only) sends the Telnet Protocol
WON'T ECHO or WILL ECHO commands (depending on whether VALUE is
true or false, respectively). For clients that support the Telnet
Protocol, this should toggle whether or not the client echoes
locally the characters typed by the user. Note that the server
never echoes input characters under any circumstances.
-- Fixed stupid bug that let people lose by setting $server_options.fg_ticks
and company to negative values.
-- Added an optional second argument to the `read()' built-in function. If it
is provided and true, then this call to `read()' will not suspend the
calling task under any circumstances. If there is input currently
available, it will be returned immediately; otherwise, `read()' returns 0.
(As before, if no input is available and no more is coming, `read()' raises
E_INVARG as an end-of-input indicator.)
-- Added flow-control to the server's input-handling: if more than a reasonable
amount of unprocessed input accumulates for any connection, the server will
temporarily stop trying to read from that connection at all, until the
backlog drops down substantially.
-- Fixed a longstanding bug in match() that could make it return garbage in
certain circumstances. More bugs in match() almost certainly still exist.
(Thanks to Judy Anderson for finding this.)
-- Fixed a minor memory leak in the case where #0:do_command() exists and
returns a list or string value. (Thanks to Ian Macintosh for finding this.)
-- Fixed a possible race condition in the TCP networking code, where a timer
could go off before we've installed the exception handler. (Thanks to Alex
Stewart for finding this.)
-- Officially deprecated the USE_GNU_MALLOC option in options.h, since it's not
aging very well.
-- Completely replaced the regular-expression matching implementation that
underlies match() and rmatch(); it used to be the GNU `regex' package and is
now the GNU `RX' package shipped with GNU `sed'. This may not eliminate all
bugs in match(), but it almost certainly has moved them around a bit.
********** The old GNU regex package had a bug in its handling of certain
** NOTE ** patterns with parentheses in them, and it is reasonably likely
********** that many MOO programmers have, perhaps unconsciously, come to
depend upon this buggy behavior. Unfortunately for such programmers, RX
does not have this bug, so you will want to fix your regular expressions
before upgrading to this release; the fixed patterns will work correctly on
both releases.
The old bug concerns patterns of the form `%( ... %)*', that is, a starred
parenthesized sub-pattern; for example, consider the MOO expression
match("foo", "%(o%)*")
Using the old regex package, this returns
{2, 3, {{2, 3}, {0, -1}, ...}, "foo"}
which is *wrong*; the last successful match of the parenthesized sub-pattern
covered just the third character, not the second and third ones. Using the
new RX package, this expression returns the proper value:
{2, 3, {{3, 3}, {0, -1}, ...}, "foo"}
To get the effect of the old bug, you need another set of parentheses around
the whole starred sub-pattern:
match("foo", "%(%(o%)*%)")
Under both GNU regex and RX, if M is the result of this expression, we have
M[3][1] == {2, 3}
You should look carefully at your uses of match() and rmatch() before
upgrading to this release, fixing those places where your code depends on
the old, buggy behavior.
-- Added an optional third argument to the built-in function notify(); if it is
provided and true, and if there isn't enough room left in the given user's
output buffer to hold the given line, then notify() will return false and
the line will not have been queued for output. In all other circumstances,
notify() now returns true. If the new optional argument is false or not
provided, then the old behavior is invoked, in which some of the
already-queued output is discarded to make room for the new line.
-- Made it possible to change the maximum verb-call depth from inside the DB.
The MAX_VERB_DEPTH constant in options.h was replaced by
DEFAULT_MAX_STACK_DEPTH, which can be overridden by
$server_options.max_stack_depth. The maximum stack depth for any task is
set at the time that task is created and cannot be changed thereafter. This
implies that suspended tasks, even after being saved in and restored from
the DB, are not affected by later changes to
$server_options.max_stack_depth.
-- The task scheduler is now guaranteed never to assign a task_id() of zero.
-- The built-in functions notify(), connected_players(), connected_seconds(),
idle_seconds(), connection_name(), and set_connection_option() now treat
connections on which boot_player() has been called as if they did not exist.
-- A number of the messages printed to a connection by the server under various
circumstances can now be customized or eliminated from within the DB. In
each case, a property on $server_options is checked at the time the message
would be printed. If the property does not exist, the standard message is
printed. If the property exists and its value is not a string, then no
message is printed at all. Otherwise, the string is printed in place of the
standard message. The following list covers all of the newly customizable
messages, showing for each the name of the relevant property on
$server_options, the default/standard message, and the circumstances under
which the message is printed:
timeout_msg "*** Timed-out waiting for login. ***"
This in-bound network connection was idle and un-logged-in for
at least CONNECT_TIMEOUT seconds (as defined in options.h).
recycle_msg "*** Recycled ***"
The logged-in user of this connection has been recycled.
boot_msg "*** Disconnected ***"
The function boot_player() was called on this connection.
redirect_from_msg "*** Redirecting connection to new port ***"
The logged-in user of this connection has just logged in on
some other connection.
redirect_to_msg "*** Redirecting old connection to this port ***"
The user who just logged in on this connection was already
logged in on some other connection.
create_msg "*** Created ***"
The user object that just logged in on this connection did not
exist before #0:do_login_command() was called.
connect_msg "*** Connected ***"
The user object that just logged in on this connection existed
before #0:do_login_command() was called.
-- The `for VAR in [EXPR..EXPR]' looping construct can now be used with either
numbers or object numbers. That is, the construct `for o in [#0..#100]' is
now legal and does the obvious thing. NOTE that in the example `o' will
take on each of 101 object numbers in the specified range, regardless of
whether or not those object numbers are valid.
-- By popular request, added the built-in function `value_bytes(VALUE)', which
returns the number of bytes of memory required to store the given value. [I
was also asked to provide an `object_bytes(OBJ)' function, to give the total
memory required to store the given valid object, but I wanted to think
longer about possible interactions with 1.8.0's new modularity wall between
the DB implementation and the rest of the server.]
-- At long last, there is a DB-settable limit on the number of queued tasks any
single user can have at once. If $server_utils.user_task_limit exists and
is a non-negative number, then that is the `task limit' for normal users;
otherwise, the task limit is infinite. For wizards, the task limit is
controlled similarly by $server_utils.wizard_task_limit. Whenever a `fork'
statement or `suspend()' call are executed, the server checks whether or not
the current verb's owner (really, the current task perms) is already at or
above their task limit; if so, E_QUOTA is raised instead of either forking
or suspending. Reading tasks are not affected by the task limit.
-- The result of `tostr(E_QUOTA)' has been changed to the string
"Resource limit exceeded".
-- Applied Alex Stewart's pAS4 patch, which modifies the result of the
built-in function `connection_name()' on TCP networking configurations to
contain the remote port of the connection as well as the host name, in the
following format:
"99 from FOO.BAR.COM, port 9999"
As before, the first number in the result is pretty useless to MOO
programmers (it's the server's file descriptor for the connection) but can,
believe it or not, occasionally be useful to the maintainer.
******** Before upgrading an existing MOO to use this version of the server,
* NOTE * you should modify the verb $string_utils:connection_hostname_bsd as
******** follows:
@chmod $string_utils:connection_hostname_bsd -d
@program $string_utils:connection_hostname_bsd
s = args[1];
return strsub($string_utils:explode(s)[3], ",", "") || "";
.
This code should work compatibly with either version of the server.
-- Applied the key part of Alex Stewart's pAS7 patch, which fixes a problem
with the server occasionally hanging under Sun's Solaris 2.X system. [I
just removed the (useless) call to `shutdown()'; the SO_LINGER setting
didn't seem necessary or very useful.]
-- Added a new item to options.h, UNFORKED_CHECKPOINTS (off by default), that
prevents the server from forking a separate process to make checkpoints;
instead, the main server process performs the checkpoints itself, halting
all user interaction and MOO task execution for the duration.
-- Made the server's log output during the initial database load *slightly*
easier to understand.
-- Removed perhaps the last hard limit in the server; you can now have input
lines with more than 500 words on them. (Thanks to Bill Drury for sending
the message to MOO-Cows that finally got me to fix this longstanding bug.)
-- Disallowed empty verb names and those made up only of spaces.
-- Added a new built-in function `toliteral(VALUE)' such that
eval("return " + toliteral(VALUE) + ";") == {1, VALUE}
for all MOO values.
-- The `create()' built-in function will now create children of #-1.
-- Added an `emergency wizard mode' to the server's start-up sequence; if you
give an initial `-e' option on the command line, then after loading in the
database but *before* running #0:server_started(), the server will use its
standard input and output streams to allow execution of wizardly `eval'
commands and re-programming of verbs. For more details, type `help' from
inside the mode.
-- The verbs #0:user_disconnected() and #0:user_client_disconnected() are now
called for un-logged-in and outbound connections, too, just as they are for
logged-in ones.
******** Before upgrading an existing MOO to use this version of the server,
* NOTE * you should check your versions of these verbs to ensure that they
******** will work appropriately when passed negative (and therefore
invalid) object numbers. In most cases, it is probably sufficient to add
the following lines to the top of these verbs:
if (args[1] < #0)
return;
endif
This code should work compatibly with either version of the server, since
these verbs weren't being called with such object numbers before.
-- The server now saves, in the DB file, a list of all users with active
connections at the time of the checkpoint, shutdown, or panic that made the
file. Upon server start-up, if such a list is present in the DB file, a
call is made to #0:user_disconnected() for each formerly active connection.
These calls are made *before* the call to #0:server_started(). In this way,
there are no longer any discontinuities across a server reboot; from the
point of view of code in the DB, the only evidence of a reboot is that,
first, a relatively long time has passed since the last task execution,
second, all connections to the server (in-bound or out, logged-in or not)
have simultaneously been closed, and third, #0:server_started() has been
called by the server.
-- Fixed RX to do reverse-searching properly, so now rmatch() works again.
Also added a abort-check in the searching inner loop for the MOO interpreter
having run out of seconds.
Version 1.7.9, 18 October 1995
-- Fixed stupid bug in registration of toliteral(); it now shouln't raise
E_TYPE on all calls...
-- Added new built-in function `queue_info([USER])'. If USER is omitted,
returns a list of object numbers naming all users that currently have task
queues inside the server; if USER is provided, returns the number of tasks
currently queued for that user. It is guaranteed that queue_info(X) will
return zero for any X not in the result of queue_info(). In essence,
queue_info(X) is a very efficient version of
set_task_perms(X);
return length(queued_tasks());
In particular, it (a) doesn't have to allocate a large list structure, and
(b) tells you when passed no arguments the complete set of users for whom
there might actually be any queued tasks.
-- Fixed initialization bug in code to track number of queued tasks.
-- Liberalized the rules for when a call to read() without arguments will
succeed. Wizards won't get E_PERM if the current task is the one that was
last spawned by a command from the connection in question. You can assure
this in a number of ways:
1) Never suspend, but rather only call read(). This is the one way
that used to work.
2) Get lucky, go ahead and call suspend(), and have it just happen that
no commands have been read since the last time you called read(),
perhaps because the user was waiting for a prompt and no typing
ahead. This technique is *not* recommended.
3) Before suspending, call
set_connection_option(player, "hold-input", 1)
thereby ensuring that no commands will be taken from this player's
queue until you call
set_connection_option(player, "hold-input", 0)
This new third technique, suggested by Alex Stewart, was the impetus for
making this change.
-- Fixed a memory leak in the server's use of the new matcher.
-- Fixed a bug whereby the third element of a successful result of match() or
rmatch() contained 29 elements instead of the usual 9.
-- Changed the method for determining a given user's queued-task limit, which
is checked on every `fork' or `suspend()' from code running with that user's
permissions. If the current task perms are valid, and that object has a
`queued_task_limit' property, and the value of that property is a
non-negative number, then that number is the limit. Otherwise, if
$server_options.queued_task_limit exists and its value is a non-negative
number, then that's the limit. Otherwise, there is no limit. (Thanks to
Gustavo Glusman for his suggestion on how to do this.)
-- The following release note, given for the 1.7.9alpha1 release, contained an
error; the following version corrects it.
********** The old GNU regex package had a bug in its handling of certain
** NOTE ** patterns with parentheses in them, and it is reasonably likely
********** that many MOO programmers have, perhaps unconsciously, come to
depend upon this buggy behavior. Unfortunately for such programmers, RX
does not have this bug, so you will want to fix your regular expressions
before upgrading to this release; the fixed patterns will work correctly on
both releases.
The old bug concerns patterns of the form `%( ... %)*', that is, a starred
or plussed parenthesized sub-pattern; for example, consider the MOO
expression
match("foo", "%(o%)+")
Using the old regex package, this returns
{2, 3, {{2, 3}, {0, -1}, ...}, "foo"}
which is *wrong*; the last successful match of the parenthesized sub-pattern
covered just the third character, not the second and third ones. Using the
new RX package, this expression returns the proper value:
{2, 3, {{3, 3}, {0, -1}, ...}, "foo"}
To get the effect of the old bug, you need another set of parentheses around
the whole starred sub-pattern:
match("foo", "%(%(o%)+%)")
Under both GNU regex and RX, if M is the result of this expression, we have
M[3][1] == {2, 3}
You should look carefully at your uses of match() and rmatch() before
upgrading to this release, fixing those places where your code depends on
the old, buggy behavior.
-- ********** There is a severe performance bug in the new matcher, causing it
** NOTE ** to run exponentially slowly in certain cases. Fortunately, these
********** cases are usually easy to work around. If you get an `out of
seconds' traceback inside a call to `match()' or `rmatch()' (which will be
accompanied by an error message in the log giving the pattern in use), you
should probably check first for an instance of this problem.
The problem concerns starred or plussed sub-patterns inside a starred or
plussed parenthesized pattern. For example, here is a perfectly reasonable
pattern matching MOO strings:
"\"%([^\"\\]+%|\\.%)*\""
(It matches double quotes around a sequence of either (a) a cluster of
characters that don't require escaping, or (b) a single escaped character.)
Note, however, that it contains a plussed sub-pattern inside of starred
parentheses, precisely the bad case for the new matcher. Fortunately, this
pattern can be altered slightly, removing the `+', without changing the
meaning:
"\"%([^\"\\]%|\\.%)*\""
This pattern does not cause problems for the new matcher. I don't know of
any problematical patterns in LambdaCore, but there is at least one in
JHCore, in $code_utils:safe_eval, where it uses this pattern:
"^%([^\"()=]+%|\"%([^\\\"]*%|\\.%)*\"%)*$"
^ ^ ^
There are three instances of the problem here, indicated by the up-arrows;
the first two of them can be removed without changing the meaning:
"^%([^\"()=]%|\"%([^\\\"]%|\\.%)*\"%)*$"
The third instance, in my testing, does not appear to cause any problems in
practice.
I am continuing my search for a better regexp implementation, but this one
has worked well enough for us in practice that I didn't feel it was worth
holding up the release for it.
Version 1.7.9p1, 18 October 1995
-- Fixed another place in the MOO's interface to the regexp code where it was
assuming 29 registers instead of 9. I think that's all of them, now... :-(
Version 1.7.9p2, 28 October 1995
-- Fixed a bug that could crash the server if #0:do_command suspended or
aborted.
-- Fixed a minor storage leak in the emergency wizard mode.
-- Switched over to yet another implementation of regular expressions, this one
extracted from the Python-1.3 release. It's much simpler than RX, and has
seen a lot more use, so it's much more likely to be stable. It does share
some of the same inherent limitations as the old regex package used in the
1.7.8p4 release of the server (which it resembles pretty closely in many
ways), but I think I've tweaked the parameters in such a way as to render
those limitations invisible for the vast majority of MOO uses. This matcher
also doesn't have the exponential performance bug that existed in the one
used in 1.7.9p1.
-- The match() and rmatch() functions now raise E_QUOTA if they run out of
memory during matching. Previously, they simply returned {}, just as if the
pattern had not matched.
Version 1.8.0alpha1, 29 November 1995
-- Much internal cleanup, especially surrounding the interface between the DB
module and the rest of the server.
-- Added an optional argument to the built-in function callers(); if provided
and true, each element of the returned value will have a sixth element, the
currently executing line number of the corresponding verb.
-- Added new facilities for raising and handling MOO errors, as described in
the next four points.
-- Added built-in function `raise(CODE [, MSG [, VALUE]])' where CODE can be
any MOO value (*not* just one of type ERR), MSG defaults to `tostr(CODE)',
and VALUE defaults to 0. This raises CODE as an error, just like dividing
by zero raises E_DIV. If the error is not caught (by one of the other new
constructs described below), then MSG will appear on the first line of the
resulting traceback printed to the user. VALUE is accessible to an error
handler established by the new TRY-EXCEPT-ENDTRY construct; see below.
-- Added new error-catching expression:
`EXPR ! CODES => EXPR_H'
NOTE: The open- and close-quotation marks in the previous line are really
part of the syntax!
In this new expression, EXPR and EXPR_H are arbitrary expressions. CODES
is either the new keyword "ANY" or a non-empty "argument list" of
expressions; just like normal argument lists, CODES can contain "@"-marked
expressions that evaluate to lists to be spliced into the resulting list of
error codes. The "=> EXPR_H" part is optional.
First, CODES is evaluated yielding a list of error codes that should be
caught if raised; if CODES is "ANY", then it is equivalent to the list of
all possible MOO values.
Next, EXPR is evaluated. If it evaluated normally, without raising an
error, then its value becomes the value of the entire error-catching
expression. If evaluating EXPR results in an error being raised, then call
it E. If E is in the list resulting from CODES, then E is considered
"caught" by this error-catching expression. In such a case, if EXPR_H was
given, it is evaluated to get the outcome of the entire error-catching
expression; if EXPR_H was omitted, then E is the value of the entire
expression. If E is *not* in the list resulting from CODES, then this
expression does not catch the error and it continues to be raised, possibly
to be caught by some piece of code either surrounding this expression or
higher up on the verb-call stack.
Here are some examples:
`x + 1 ! E_TYPE => 0'
Returns x + 1 if x is a number, returns 0 if x is not a number,
and raises E_VARNF if x doesn't have a value.
`x.y ! E_PROPNF, E_PERM => 17'
Returns x.y if that doesn't cause an error, 17 if x doesn't
have a "y" property or that property isn't readable, and raises
some other kind of error (like E_INVIND) if x.y does.
`1 / 0 ! ANY'
Returns E_DIV.
-- Added new error-catching statement:
TRY
statements_0
EXCEPT id_1 (codes_1)
statements_1
EXCEPT id_2 (codes_2)
statements_2
ENDTRY
The IDs are optional, CODES has the same syntax as above, and there can be
anywhere from 1 to 255 EXCEPT clauses.
First, each CODES is evaluated yielding a list of error codes that should
be caught if raised; if any CODES is "ANY", then it is equivalent to the
list of all possible MOO values.
Next, STATEMENTS_0 is executed; if it doesn't raise an error, then that's
all that happens for the entire TRY statement. Otherwise, let E be the
error it raises. From top to bottom, E is searched for in the lists
resulting from the various CODES_i; if it isn't found in any of them, then
it continues to be raised, possibly to be caught by some piece of code
either surrounding this TRY statement or higher up on the verb-call stack.
If E is found first in CODES_i, then ID_i (if provided) assigned a value
containing information about the error being raised and STATEMENTS_i is
executed. The value assigned to ID_i list a list of four elements:
{CODE, MSG, VALUE, TRACEBACK}
where CODE is E, the error being raised, MSG and VALUE are as provided by
the code that raised the error (built-in errors, such as division by zero,
currently act as if the MSG and VALUE arguments to raise() were omitted),
and TRACEBACK is a list like that returned by callers(), including line
numbers. The TRACEBACK list contains entries for every verb from the one
that raised the error through the one containing this TRY statement.
Here is an example:
try
result = object:(command)(@arguments);
player:tell("=> ", toliteral(result));
except v (ANY)
tb = v[4];
if (length(tb) == 1)
player:tell("** Illegal command: ", v[2]);
else
top = tb[1];
tb[1..1] = {};
player:tell(top[1], ":", top[2], ", line ", top[6], ":",
v[2]);
for fr in (tb)
player:tell("... called from ", fr[1], ":", fr[2],
", line ", fr[6]);
endfor
player:tell("(End of traceback)");
endif
endtry
-- Added new error-cleanup statement:
TRY
statements_0
FINALLY
statements_1
ENDTRY
STATEMENTS_0 is executed; if it completes without raising an error or
returning from this verb, then STATEMENTS_1 is executed and that's
all that happens for the entire TRY statement. Otherwise, the process of
raising the error past this point or returning from this verb (as
appropriate) is interrupted and STATEMENTS_1 is executed. If STATEMENTS_1
itself completes without raising an error or returning from this verb, then
the interrupted raising or returning process is resumed. If STATEMENTS_1
does return or raise an error, then the interrupted raising or returning
process is simply forgotten in favor of the new one.
In short, this statement ensures that STATEMENTS_1 is executed after
control leaves STATEMENTS_0 for whatever reason; it can thus be used to make
sure that some piece of cleanup code is run even if STATEMENTS_0 doesn't
simply run normally to completion.
Here's an example:
try
start = time();
object:(command)(@arguments);
finally
end = time();
this:charge_user_for_seconds(player, end - start);
endtry
-- Completely rewrote the MOO-code decompiler, restructuring it to be able to
cope with the new exception-handling constructs.
-- Fixed bug in handling of EOF in emergency wizard mode; it is now treated as
equivalent to the `quit' command.
-- Added internal interfaces allowing built-in function implementations (a) to
be notified when given file descriptors are readable/writable (see file
net_multi.h), (b) to resume tasks that they earlier caused to suspend (see
resume_task() in file tasks.h), and (c) to make it possible for their
suspended tasks to be listed by queued_tasks() and killed by kill_task().
-- Added a new, essentially empty module `extensions.c', intended to be easily
replaced by users with a file of their own MOO extensions. The file also
contains some examples of extensions using all of the new internal
interfaces.
-- Clarification to change made in 1.7.9alpha1: $server_options.max_stack_depth
can only be used to override DEFAULT_MAX_STACK_DEPTH if it *increases* the
value. This is good because (a) there probably aren't any good reasons to
want to lower the limit, and (b) you could get good and screwed if you set
it too low.
-- Fixed match() and rmatch() to treat unrecognized escape sequences in
patterns as if the `%' were not there. Thus, `%X' in a pattern is now
equivalent to simply `X', for all X not explicitly mentioned in the
programmer's manual.
Version 1.8.0alpha2, 11 December 1995
-- Added an optional argument to `connected_players()'; if provided and true,
this now includes *all* active connections in the returned list, including
outbound or un-logged-in ones.
-- Fixed checkpoint timing to test for overflow when adding $dump_interval to
the current time; when this happens, it pretends that $dump_interval = 3600.
This should fix the problem someone found where setting $dump_interval to
$maxint caused continuous checkpointing.
-- Fixed `make clean' to remove a few more derived files that aren't part of
the distribution, including `Makefile' itself.
-- Added two new DB-settable options:
$server_options.name_lookup_timeout
is the maximum number of seconds the server will wait for a response
to a hostname <--> IP address lookup before giving up.
$server_options.outbound_connect_timeout
is the maximum number of seconds that a server will hang waiting for
an outbound network connection attempt to succeed; after that, it
gives up.
-- Added new built-in function `resume(TASK_ID [, VALUE])', where TASK_ID is
the task ID of an existing suspended task and VALUE defaults to zero. If
the programmer is a wizard or owns the specified task, the task is made
immediately runnable; when it runs, its call to `suspend()' will return
VALUE. Raises E_PERM if the programmer does not own the specified task and
E_INVARG if TASK_ID does not specify an existing suspended task.
-- Made the argument to `suspend()' be optional; if omitted, the suspension is
effectively forever. Such eternally suspended tasks can only be made
runnable again with the new `resume()' function.
-- Verbs that have not yet been programmed are now completely equivalent to
those with the empty program. In particular, they are no longer invisible
to normal verb-calls.
-- Added new built-in function `task_stack(TASK_ID [, LINE_NUMBERS_TOO])',
where TASK_ID is the task ID of an existing suspended task and
LINE_NUMBERS_TOO defaults to false. Returns a value like that returned by
`callers()', but covering all of the stack frames of the specified task.
-- It is now possible to make specific built-in properties readable only by
wizards. For a built-in property `foo', if $server_options.protect_foo
exists and has a true value, then any attempt by a non-wizard to read the
value of the `foo' property of any object will raise E_PERM. For example,
if $server_options.protect_location existed with the value 1, then the
`location' property of all objects would only be readable by wizards.
-- Added OUTPUTPREFIX and OUTPUTSUFFIX as synonyms for the built-in PREFIX and
SUFFIX commands. I'm told that this would make the MOO compatible with some
other servers, allowing the same somewhat fancy clients to be used with
both.
-- Added new built-in function `object_bytes(OBJECT)', where OBJECT is a valid
object and the result is the number of bytes of memory required to represent
that object and all of its verbs and properties. Raises E_PERM if called by
a non-wizard. [There's a surprising amount of difference between this very
accurate count and the one returned by the rough-guess :object_bytes() verb
I wrote long ago, which is still used as the basis of most byte-based quota
implementations. In particular, the vast majority of objects are, in fact,
smaller than the rough guess, sometimes by substantial amounts. For
example, the $player object in LambdaCore-1Oct94.db is farthest off, being
43,260 bytes smaller than :object_bytes() guessed. The majority though,
including Mr. Spell, are within about 5,000 bytes of the guess, which isn't
too bad, eh? It would appear that the guesses were most wrong for verb
code, which isn't very surprising.]
-- Added the syntax `$vname(...)' as an abbreviation for `#0:vname(...)', by
analogy to the `$' notation for properties on #0.
-- Fixed a server-crashing bug in which the resumption value of suspended tasks
was not properly saved and restored after reboot. (Thanks to Jin-su Ahn for
pointing this out.)
-- Fixed a bug in which forked tasks were written out incorrectly in checkpoint
files. (Thanks to Jin-su Ahn for noticing the problem.)
-- Fixed a server-crashing bug in the new decompiler. (Thanks to Jin-su Ahn
for discovering the bug.)
-- Made it possible to rename properties. If the third argument to
set_property_info(OBJECT, NAME, INFO) has a third element that is a string,
NEW, and the property NAME is defined on OBJECT (not merely inherited), and
NEW names neither a built-in property nor an existing property on OBJECT or
any of its ancestors or descendants, then "OBJECT.NAME" is renamed to
"OBJECT.NEW". This more-or-less parallels the way verb renaming works.
-- Traceback lists, returned by callers() and task_stack() and also made
available in the try-except-endtry construct, now include entries for
built-in functions that have made MOO verb calls. Such entries have the
same form as normal entries, but the THIS, PROGRAMMER, and VERB-LOC fields
(elements 1, 3, and 4) are all set to #-1, the VERB-NAME field (element 2)
is set to the name of the built-in function, and the LINE-NUMBER field
(element 6), if any, is set to an index used internally to determine the
current state of the built-in function. The simplest safe test for a
built-in function entry is
(VERB-LOC == #-1 && PROGRAMMER == #-1 && VERB-NAME != "")
This combination was not previously possible. I leave it to the interested
reader to discover the previously possible circumstances under which any
weaker test might fail...
Version 1.8.0alpha3, 27 December 1995
-- Added new built-in function `call_function(FNAME, @ARGS)' where FNAME is a
string naming a built-in function to call and ARGS are the arguments to pass
to it. This allows you to compute the name of the function to call and, in
particular, allows you to write a call to a built-in function that may or
may not exist in the particular version of the server you're using.
-- Changed MOO-code parser to just issue a warning during DB loading if a call
to an unknown built-in function is encountered; in such a case, the call is
replaced by an equivalent one to `call_function()'.
-- Changed parser error and warning messages during DB loading to include
information about where the program in question is located.
-- Fixed registration of `suspend()' to allow for it being called with no
arguments. (Thanks to Jin-su Ahn for discovering the bug.)
-- Fixed a nasty memory leak in the new match() implementation. (Thanks to
Bill Garrett and Judy Anderson for tracking down a test case.)
-- Fixed a bug in the line-number-finding code in the new decompiler. (Thanks
to Jin-su Ahn for noticing the problem.)
-- Fixed a new bug preventing the loading of suspended tasks from the DB file.
(Thanks once again to Jin-su Ahn for pointing this out.)
-- Fixed a new memory leak in the handling of errors raised from built-in
functions called from !d verbs.
-- Fixed a new memory leak in command parsing.
-- Fixed a new memory leak in built-in property reference.
-- Fixed a serious memory leak in the new decompiler, triggered every time the
server needed to determine a line number in a verb program.
-- Fixed a minor memory leak in the loading of forked tasks from the DB file.
-- Fixed bug that would cause prepositions not to be properly matched in the
very first user command processed after booting. Beware C functions that
return pointers to static private data... :-/
-- Added new built-in function `db_disk_size()', which returns the total size,
in bytes, of the most recent full representation of the database as one or
more disk files. It raises E_QUOTA if, for some reason, no such on-disk
representation is currently available.
-- Changed network input handler to accept any of CR, LF, or CRLF as an end-of-
line indicator. This should improve robustness in the face of broken telnet
clients that assume local EOL conventions will work across the net.
-- Added progress messages to the dumping/checkpointing process just like those
printed during the loading process.
-- Added some extra configuration help for NeXT machines.
-- Fixed server-crashing bug in case where a raised error is not handled by the
innermost handler. (Thanks to Nate Massey for finding the problem.)
-- Added support for unambiguous reference to specific verbs in the built-in
functions `delete_verb()', `verb_info()', `set_verb_info()', `verb_args()',
`set_verb_args()', `verb_code()', `set_verb_code()', and `disassemble(). In
all cases, if the second argument (the verb name) is a positive number
instead of a string, it is treated as a 1-based index into the list of all
verbs defined on the given object.
********** The old behavior, in which numeric strings in these argument
** NOTE ** positions (and in the built-in `.program' command) were sometimes
********** treated as 0-based indices into the list of defined verbs, has
now been disabled by default. To re-enable it, define the property
`$server_options.support_numeric_verbname_strings' and give it a true value.
Most existing databases will require this re-enabling, at least for a period
of transition; a certain amount of important old code critically depends on
this old behavior.
I chose to do things this way in order to emphasize the strong incentive
for sites to update the affected old code to the new, more reliable
mechanism; with the old mechanism enabled, it is possible to create
situations where some verbs (with numeric names) simply cannot be referred
to by name.
-- Fixed bug where errors raised by built-in functions called from !d verbs are
not properly squelched if the built-in has itself previously called a MOO
verb. (Thanks to Brian Buchanan for noticing this bug.)
Version 1.8.0alpha4, 30 December 1995
-- Fixed foolish C-syntax bug in the delete_verb() built-in. (Thanks to Brian
Buchanan and Nate Massey for finding this.)
-- Fixed configuration problem on FreeBSD and perhaps other platforms. (Thanks
again to Brian Buchanan for pointing this out.)
-- Fixed compilation warning in net_bsd_tcp.c when OUTBOUND_NETWORK was
defined. (Thanks to Martian and others for reporting this.)
-- Fixed nasty occasional memory smash in delete_property(). (Thanks to Kai
Storbeck for reporting it.)
-- All DB-configurable messages (e.g., $server_options.timeout_msg, etc.) can
now be given as either a string or a list of strings; in the latter case,
each string will be printed on a separate line.
-- New DB-configurable message $server_options.server_full_msg is printed to
any connection that arrives when the server really can't accept any more
connections; after printing the message, the temporarily-accepted connection
is immediately closed. The default for this message is as was formerly
hardwired into the server:
*** Sorry, but the server cannot accept any more connections right now.
*** Please try again later.
-- The DB-configurable messages are no longer printed on outbound connections.
-- Added a version of the MPL (multiple-port listening) extension first
implemented by Ivan Judson. This involves the addition of three new
built-in functions:
listen(OBJ, DESC [, PRINT_MESSAGES]) => CANON_DESC
Create a new point at which the server will listen for network
connections, just as it does normally. OBJ is the object whose
verbs :do_login_command, :do_command, :do_out_of_band_command,
:user_connected, :user_created, :user_reconnected,
:user_disconnected, and :user_client_disconnected will be called at
appropriate points, just like these verbs are called on #0 for
normal connections. (If a user reconnects and the user's old and
new connections are on two different listening points being handled
by different objects, then :user_client_disconnected is called for
the old connection and :user_connected for the new one.) DESC is a
network-configuration- specific parameter describing the listening
point. For the TCP configurations, DESC is a port number on which
to listen. If PRINT_MESSAGES is provided and true, then the
various DB- configurable messages will be printed on connections
received at the new listening point. CANON_DESC is a
`canonicalized' version of DESC, with any configuration-specific
defaulting or aliasing accounted for. For the TCP configurations,
CANON_DESC is equal to DESC unless DESC is zero, in which case
CANON_DESC is a port number assigned by the operating system. This
raises E_PERM if the programmer is not a wizard, E_INVARG if OBJ is
invalid or there is already a listening point described by DESC,
and E_QUOTA if some network-configuration-specific error occurred.
unlisten(DESC)
Stop listening for connections on the point described by DESC,
which should be a value returned by some successful call to
listen(). Raises E_PERM if the programmer is not a wizard and
E_INVARG if there does not exist a listener with that description.
listeners() => {{OBJ, DESC, PRINT_MESSAGES}, ...}
Returns a list of all existing listeners, including the default one
set up by the server at boot time.
I believe that this implementation is upward-compatible with the existing
uses of Ivan's package. As far as I know, the only differences are that his
package did not have the PRINT_MESSAGES argument to listen(), did not work
for the non-BSD/TCP networking configurations, did not call :do_command or
:do_out_of_band_command, did not distinguish the case of a user reconnecting
on a different listening point, and did not have the listeners() function.
Please note that there is now nothing special about the initial listening
point created by the server at boot time; you can use unlisten() on it just
as if it had been created by listen(). This can be useful; for example, you
might start up your server on some obscure port, say 12345, connect to it by
yourself for a while, and then open it up to normal users with the command
;;unlisten(12345); listen(#0, 7777, 1)
-- Changed the result of connection_name() to expose information about the
listening point on which the connection was accepted. For the TCP
networking configurations, the format is now:
"port 7777 from foo.bar.com, port 3456"
where 7777 is the server port on which the connection arrived and the rest
is as before.
********** Before upgrading an existing LambdaCore-based MOO to use this
** NOTE ** version of the server, you should modify the verb
********** $string_utils:connection_hostname_bsd as follows:
@program $string_utils:connection_hostname_bsd
s = args[1];
m = match(s, "^.* %(from%|to%) %([^, ]+%)");
return m ? substitute("%2", m) | "";
.
This code should work compatibly with any version of the server since 1.6.0.
-- Added a new kind of expression, allowed only within the indexing/subranging
brackets `[...]'; in this context, the expression `$' means the length of
the value being indexed or subranged. This allows, for example, expressions
like `x[2..$]' to get the `rest' of a list after the first element or
`x[random($)]' to get a random element of a list. You can also use this on
the left-hand side of an assignment; for example `x[$] = 7' sets the last
element of a list, and `x[$+1..$] = {y}' adds y onto the end of the list x.
My favorite abuse of the new syntax is `x[l = $]' as a substitute for the
verbose and now old-fashioned `l = length(x)'.
-- Fixed longstanding bug in `client_sysv', the client for the SYSV/LOCAL
networking configuration; it never worked to specify a server connection
file on the command line!
Version 1.8.0alpha5, 11 January 1996
-- Removed test that caused commands from disconnected users to be discarded.
(Thanks to Nate Massey for asking the question that led to this.)
-- If $do_command() returns false without suspending (signalling that the
server should carry on with its normal command parsing and dispatch), the
resulting input task will have the same task_id() value as the $do_command()
task did. (Thanks to Nate Massey for the suggestion.)
-- Added configuration code to cope with the C function crypt() not existing;
in such a case, the crypt() built-in function simply returns the first
argument, unchanged. (Thanks again to Nate Massey for the suggestion.)
-- Fixed bogus-memory freeing bug in add_verb(). (Thanks to Mark Blanchard for
reporting this and to Purify for finding it.)
-- Fixed missing declaration of strlen() in parser.y. (Thanks again to Mark
Blanchard for the report.)
-- Fixed bug where the wrong permissions are used by built-in functions who ask
after they've already called some MOO function. (Thanks to Brian Buchanan
for finding this.)
-- Added new built-in function `buffered_output_length([CONN])', where CONN is
an active connection and the result is the number of bytes currently
buffered up for output. If CONN is not provided, this returns the maximum
number of bytes that will be buffered up for output on any connection.
(Thanks to Martian for the impetus for this and to Gustavo Glusman and Jaime
Prilusky for inspiration on the interface.)
-- Made it easier for server code to get the result of a suspended task, by
adding an `result' argument to resume_from_previous_vm(). (Thanks to Jin-su
Ahn for pointing out this irregularity in the code.)
-- Fixed problem with non-flushing notify() calls when the string to be output
is larger than MAX_QUEUED_OUTPUT; this always failed, even when there wasn't
any other output in the buffer. Now, if the buffer is currently empty, any
line of output, no matter how large, will be accepted.
-- Added facility for doing fully-general binary I/O on connections. If you
call `set_connection_option(CONN, "binary", 1)', then both input from and
output to that connection can contain arbitrary bytes. On input, any byte
that isn't an ASCII printing character, the space character, or the TAB
character will be represented as the four-character substring "~NNN", where
NNN is the octal representation of the byte; the input character `~' is
represented by the two-character substring "~~". Call this representation
of bytes "binary strings". Input from a connection in binary mode is *not*
broken into lines at all; it is delivered to either the read() function or
the built-in command parser in whatever size chunks come back from the
operating system. For output to a connection in binary mode, the second
argument to `notify()' must be a binary string; if it is malformed, E_INVARG
is raised. Of course, the call `set_connection_option(CONN, "binary", 0)'
sets the connection back into the normal MOO I/O mode.
-- For convenience in manipulating binary strings, two new built-in functions
are provided:
decode_binary(BIN_STRING [, FULLY]) => LIST
Returns a list of strings and/or numbers representing the bytes in
BIN_STRING in order. If FULLY is false or omitted, the list
contains a number only for each non-printing, non-space, non-tab
byte; all other characters are grouped into the longest possible
contiguous substrings. If FULLY is provided and true, the list
contains only numbers, one for each byte represented in BIN_STRING.
Raises E_INVARG if BIN_STRING is malformed.
Examples:
decode_binary("foo") => {"foo"}
decode_binary("~~foo") => {"~foo"}
decode_binary("foo~015~012") => {"foo", 13, 10}
decode_binary("foo~012bar~012baz")
=> {"foo", 10, "bar", 10, "baz}
decode_binary("foo~015~012", 1) => {102, 111, 111, 13, 10}
encode_binary(@ARGS) => BIN_STRING
ARGS must be a list of numbers between 0 and 255, strings, and
lists of this same form. Translates each number and string in turn
into its binary string equivalent, returning the concatenation of
all these substrings into a single binary string.
Examples:
encode_binary("~foo") => "~~foo"
encode_binary({"foo", 13, 10}, {"bar", 13, 10})
=> "foo~015~012bar~015~012"
encode_binary("foo", 13, 10, "bar", 13, 10)
=> "foo~015~012bar~015~012"
-- Fixed two places missed in a change made in 1.7.9alpha1;
$user_disconnected() is now called for un-logged-in connections that either
time out or are the target of a call to boot_player(). (Thanks to Brian
Buchanan for pointing out the discrepancy.)
-- Fixed memory-smash bug in the renumber() built-in function. (Thanks again
to Brian Buchanan for reporting this and to Purify for making it easy to
find the problem.)
Version 1.8.0alpha6, 16 January 1996
-- Replaced panic with raising E_TYPE in the case where `x[$]' is applied to a
non-string, non-list `x'. (Thanks to Kipp the Kid for reporting this.)
-- Fixed (yet another) bug in the stack-unwinding code; this one stuck the
return information for each of a built-in's second and subsequent verb calls
on successively higher stack frames, causing all manner of havoc, the least
of which was somewhat surprising results from the (correct!) callers()
built-in. (Thanks once again to Brian Buchanan for reporting this bug; this
marks the *third* bug in stack-unwinding that Brian has uncovered!)
-- Changed the rules for binary strings slightly so that (a) TAB is no longer
treated as a normal, printing character (i.e., it gets converted to `~011'
on input in binary mode), and (b) `~' is treated just like all of the
non-printing characters (i.e., it gets converted to `~176' on input in
binary mode and the sequence `~~' is illegal in binary strings). Change (a)
removes one more source of TAB characters from the MOO (which can only be a
good thing), and change (b) both simplifies the specification and makes it
easier to compute the true length of a binary string (i.e., it's the length
of the string itself minus three times the number of tildes). (Thanks to
James Deikun for these suggestions.)
-- Fixed uninitialized-variable bug in the `program' command in emergency
wizard mode. (Thanks to Kipp the Kid for reporting it.)
-- Fixed potential memory smash bug if #100 (or #200 or #400 or #800 or ...)
was a recycled object in the DB file being loaded. (Thanks to Dave Van
Buren for sending me a DB file that tickled the bug and to Purify for
pinpointing it.)
-- Fixed bug that caused *all* settings of the .wizard property to be logged,
instead of just those that set it to a true value. (Thanks again to Dave
Van Buren for reporting this.)
-- Fixed bug accuracy of the top line number in tracebacks resulting from the
setting of a wizard bit. (Thanks one last time to DVB for sending me a
traceback that contained such an error.)
-- Added new built-in function `function_info([NAME])' which returns
descriptions of built-in functions available on the server. If NAME is
provided, only the description of the function with that name is returned;
if NAME is omitted, a list of descriptions is returned, one for each
function available on the server. Each description is a list of the
following form:
{NAME, MIN_ARGS, MAX_ARGS, TYPES}
NAME is the name of the built-in function, MIN_ARGS is the minimum number of
arguments that must be provided to the function, MAX_ARGS is the maximum
number of arguments that can be provided to the function or -1 if there is
no maximum, and TYPES is a list of MAX_ARGS numbers (or MIN_ARGS if MAX_ARGS
is -1), each of which represents the type of argument required in the
corresponding position. Each type number is as would be returned from the
typeof() built-in function except that -1 indicates that any type of value
is acceptable. For example, here are several entries from the list:
{"listdelete", 2, 2, {4, 0}}
{"suspend", 0, 1, {0}}
{"server_log", 1, 2, {2, -1}}
{"tostr", 0, -1, {}}
`listdelete()' takes exactly 2 arguments, of which the first must be a list
(LIST == 4) and the second must be a number (NUM == 0). `suspend()' has one
optional argument that, if provided, must be a number. `server_log()' has
one required argument that must be a string (STR == 2) and one optional
argument that, if provided, may be of any type. `tostr()' takes any number
of arguments and it can't be determined from this description which argument
types would be acceptable in which positions.
It should be noted that, in a feat of synchronicity, this new built-in
brings the total number supported in the official server release to exactly
100! (Thanks to some-user-whose-message-I've-lost [feel free to mail me
again so that I can patch this thank-you note] for suggesting such a
built-in function.)
-- Fixed the definitions of the built-in functions match(), rmatch(), index(),
rindex(), and strsub() to accept any type of value for their final,
`case-matters' flag argument. Also fixed `verb_code()' to allow any type of
value for either of its final two flag arguments.
-- Added `scattering assignment' expression, allowing the elements of a list to
be spread among multiple variables simultaneously; this could be used, for
example, to get at the arguments to a verb in a more convenient form than
the list `args'.
A scattering assignment expression looks like this:
{ TARGETS } = EXPR
where TARGETS is a comma-separated list of places to store elements of the
list that results from evaluating EXPR. A target has one of the following
forms:
VARIABLE
This is the simplest target, just a simple variable; the list
element in the corresponding position is assigned to the variable.
I call this a `required' target, since the assignment is required
to put one of the list elements into the variable.
? VARIABLE
I call this an `optional' target, since it doesn't always get
assigned an element. If there are any list elements left over
after all of the required targets (and all of the other optionals
to the left of this one) have been accounted for, then this
variable is treated like a required one and the list element in the
corresponding position is assigned to the variable. If there
aren't enough elements to assign one to this target, then no
assignment is made to this variable, leaving it with whatever its
previous value was.
? VARIABLE = EXPR
This is also an optional target, but if there aren't enough list
elements available to assign one to this target, the result of
evaluating EXPR is assigned to it instead. Thus, EXPR is a kind of
`default value' for the variable. The default value expressions
are evaluated and assigned working from left to right *after* all
of the other assignments have been performed.
@ VARIABLE
Analogously to MOO argument lists, this variable is assigned a list
of all of the `leftover' list elements in this part of the list
after all of the other targets have been filled in. It will be
assigned the empty list, if there aren't any elements left over. I
call this a `rest' target, since it gets the rest of the elements.
There may be at most one rest target in TARGETS.
If there aren't enough list elements to fill all of the required targets, or
if there are more than enough to fill all of the required and optional
targets but there isn't a rest target to take the leftover ones, then E_ARGS
is raised.
Here are some examples of how this works. Assume first that the verb
`me:foo()' contains the following code:
b = c = e = 17;
{a, ?b, ?c = 8, @d, ?e = 9, f} = args;
return {a, b, c, d, e, f};
Then the following calls return the given values:
`me:foo(1) ! ANY' => E_ARGS
me:foo(1, 2) => {1, 17, 8, {}, 9, 2}
me:foo(1, 2, 3) => {1, 2, 8, {}, 9, 3}
me:foo(1, 2, 3, 4) => {1, 2, 3, {}, 9, 4}
me:foo(1, 2, 3, 4, 5) => {1, 2, 3, {}, 4, 5}
me:foo(1, 2, 3, 4, 5, 6) => {1, 2, 3, {4}, 5, 6}
me:foo(1, 2, 3, 4, 5, 6, 7) => {1, 2, 3, {4, 5}, 6, 7}
me:foo(1, 2, 3, 4, 5, 6, 7, 8) => {1, 2, 3, {4, 5, 6}, 7, 8}
Finally MOO has a convenient mechanism for naming verb arguments, checking
for there being exactly the right number of arguments, handling optional and
`rest' arguments, etc. I intend to start every new MOO verb of mine with a
scattering assignment of `args', and I encourage other MOO programmers to do
the same.
Version 1.8.0beta1, 7 February 1996
-- Fixed bug where scattering assignment was not checking for the right-hand
side value being a list, and a severe bug where errors in a scattering
assignment in a !d verb caused the interpreter to execute many of the
operands of the EOP_SCATTER instruction as if they were opcodes. (Thanks to
Kipp the Kid for finding this problem.)
-- Changed several routines to panic the server instead of simply logging an
error message; these were places where such an error indicated a memory
smash or some other very serious error had occurred; it made no sense to try
to press on with normal operations. This eliminates the "Impossible var
type" log messages sometimes seen from FREE_VAR, VAR_REF, and VAR_DUP.
-- Added support for in-DB handling of all tracebacks, of which there are two
kinds: unhandled errors and tasks that have timed out.
If an error is raised and not caught, then the verb-call
#0:handle_uncaught_error(CODE, MSG, VALUE, TRACEBACK, FORMATTED)
is made, where CODE, MSG, VALUE, and TRACEBACK are the values that would
have been passed to a `try-except-endtry' handler for the error and
FORMATTED is a list of strings being the lines of traceback output that
will be printed to the player.
If a task runs out of ticks or seconds, then the verb-call
#0:handle_task_timeout(RESOURCE, TRACEBACK, FORMATTED)
is made, where RESOURCE is the appropriate one of the strings "ticks" or
"seconds", and TRACEBACK and FORMATTED are as above.
In both situations, the indicated verb call is made with the same
task_id() as the task that caused the traceback. If the handler verb call
either suspends or returns a true value, then that code is considered to
have handled the traceback and no further processing will be done by the
server. On the other hand, if the appropriate handler verb does not exist,
or returns a false value without suspending, or itself causes a traceback,
the original traceback (i.e., FORMATTED) will be printed to the player as in
earlier versions of the server.
Note that, if the handler verb-call itself causes a traceback, no
`nested' handler call is made; its traceback is simply printed to the player
without further processing. This prevents what might otherwise be quite a
nasty vicious cycle.
(Thanks to ThwartedEfforts for suggesting such a feature.)
-- Added a way to flush all pending input on a given connection, mostly for use
by users who change their minds about having typed something and can react
before the server has processed it. Each connection may have a defined
`flush command'; if a raw line of input is equal to that connection's flush
command, then all pending input on the connection is flushed and a message
is printed back to the connection describing what happened. By default,
each connection's flush command is `.flush'; you can change this default by
setting $server_options.default_flush_command either to a non-empty string
(the new default) or something else (a default of `no defined flush
command'). On any given connection, you can redefine the flush command with
set_connection_option(CONN, "flush-command", VALUE)
Again, if VALUE is a non-empty string, it becomes the new flush command for
CONN; otherwise, CONN is set to have no defined flush command.
********** This could confuse things for certain kinds of unusual server
** NOTE ** connections, such as outbound ones or ones to non-MOO servers
********** running in the database (e.g., HTTP servers). You may want to
set $server_options.default_flush_command to the empty string (to disable
flush commands by default) and use set_connection_option() to change this
just for appropriate connections (e.g., in #0:do_login_command).
(Thanks to Kent Pitman and others for help in designing this feature.)
-- Added the `connection_options(CONN)' built-in function, which returns a list
of {NAME, VALUE} pairs describing the current settings of all of the allowed
options for the connection CONN. (Thanks to Brian Buchanan for suggesting
this.)
-- Added `list' and `disassemble' commands to emergency wizard mode. (Thanks
to H. Peter Anvin for writing the first versions of these.)
-- Added a new `named' form of the `while' loop:
WHILE id (expression)
statements
ENDWHILE
This behaves exactly like the statement
WHILE (id = (expression))
statements
ENDWHILE
This was added solely to provide a way to give a name to a `while' loop, for
use in the new `break' and `continue' statements, described below.
-- Added new MOO statements `break' and `continue', similar to the ones in C or
Java. The syntax is
BREAK [id];
CONTINUE [id];
A `break' statement causes your program to exit an enclosing `for' or
`while' loop; a `continue' statement causes your program to skip ahead to
the begining of the next iteration of an enclosing loop. If provided, the
ID in the `break' or `continue' statement specifies which enclosing loop is
meant; ID should be the variable name appearing directly after the `for' or
`while' keyword in the desired loop. If no ID is provided, the innermost
enclosing loop is indicated. If a `break' or `continue' statement causes
control to leave the main body of a `try - finally - entry' statement, the
`finally' part will be executed first, just as with a `return' statement.
Here's an example:
x = 0;
for i in [1..5]
notify(player, "top");
try
if (!x)
x = 1;
notify(player, "continuing");
continue;
endif
x = x + 1;
finally
notify(player, "finally");
endtry
notify(player, "after");
if (x > 1)
break;
endif
endfor
notify(player, "done");
This verb produces the following output:
top
continuing
finally
top
finally
after
done
I don't claim that this is a useful verb, mind you, but it does illustrate
all of the possible interactions.
-- By popular demand, I added the new built-in function
force_input(CONN, LINE [, AT_FRONT])
which inserts the string LINE as an input task in the queue for the
connection CONN, just as if it had arrived as input over the network. If
AT_FRONT is provided and true, then the new line of input is put at the
front of CONN's queue, so that it will be the very next line of input
processed even if there is already some other input in that queue.
-- Fixed a bug whereby the very most common case of resuming a suspended task
with a new value failed to work. (Thanks to Brian Buchanan for reporting
this.)
-- Changed the server to make a log entry whenever the value of a wizard bit
changes (as opposed to just when it goes from false to true). (Thanks to
Marc <marc@got.net> for this very sensible suggestion.)
-- Fixed a bug where the server thought it sometimes advisable to parenthesize
the single-character `$' expression...
-- Fixed stack-overflow memory-smash bug that could occur if, in a !d verb, the
`$' expression got a type error. (Thanks to Kipp the Kid for reporting
this.)
-- Fixed a typo in the registration of the functions decode_binary() and
encode_binary(). (Thanks to Richard Connamacher and H. Peter Anvin for
finding this.)
-- Added floating-point numbers as a new MOO value type; this involves a number
of changes to the behavior of existing MOO primitives, described in the
following several items. The representation of these values is in the local
C compiler's type `double', which is IEEE double precision on almost all
modern systems. IEEE infinite and NaN values are not allowed in MOO; the
new error code E_FLOAT is raised whenever one of these values would
otherwise be computed.
(Enormous thanks go to H. Peter Anvin, without whose great
efforts, understanding, and persistence these floating-point
facilities would not have made it into any release of the
server on my watch. I have not taken his patches without
modification, but nearly every change I made in adding these
features was patterned closely on what he had done. Of course,
any bugs that still remain in the server are solely my
responsibility and should not be taken to reflect badly on HPA
in any way.)
-- MOO numeric literals now have the following syntax:
digit+ [. digit+] [{e | E} [+ | -] digit+]
The number is represented in floating-point if and only if either a decimal
point or a scientific-notation marker (`e' or `E') appears in the
literal.
-- Added new built-in variables `INT' (with the same meaning as the old `NUM'
variable) and `FLOAT' (the result of typeof() applied to a floating-point
number).
-- Floating-point numbers not equal to 0.0 are treated as `true' in MOO
conditionals.
-- Both tostr() and toliteral() display floating-point numbers in the fullest
available precision, with 15 decimal digits on most machines.
-- The new built-in function `floatstr(FLOAT, PRECISION [, USE_SCI_NOTATION])'
can be used to get more control over the conversion of floating-point
numbers to strings. In this function, FLOAT is a floating-point number and
PRECISION is the number of digits to appear to the right of the decimal
point (at most the maximum available precision, 15 digits on most machines).
If USE_SCI_NOTATION is false or not provided, the result is a string in the
form "MMMMMMM.DDDDDD", preceded by a minus sign if FLOAT is negative. If
USE_SCI_NOTATION is provided and true, the result is a string in the form
"M.DDDDDDe+EEE", again preceded by a minus sign if FLOAT is negative.
-- The following operators now work in the obvious way (analogously to the
integer case) if X and Y are both floating-point numbers:
-X
X + Y X - Y X * Y X / Y X % Y
X == Y X != Y
X < Y X <= Y X > Y X >= Y
If one of X or Y is an integer and the other is a floating-point number,
then most of these operators raise E_TYPE; there are no automatic coercions
of integers to floating-point numbers. The expression (X == Y) is always
false and (X != Y) always true if X and Y do not have the same type.
(This is the most major place where I decided to depart from
HPA's patches; I was persuaded by the discussion on the
MOO-Cows list that the potential dangers posed by automatic
coercions in MOO's ubiquitously persistent world outweighed
their added convenience in some kinds of programs. It's my
guess that this decision will generate more dialog on the list,
and I welcome the input; it's always possible to extend the
server upward-compatibly later to allow such coercions.)
-- The following operations all raise E_TYPE if either X or Y is a
floating-point number:
Z[X] Z[X] = E Z[X..Y] Z[X..Y] = E
List and string indices must be integers.
-- Added built-in functions `toint()' (a synonym for `tonum()') and
`tofloat()'. The former can be used to convert a floating-point number to
an integer by truncation toward zero. The latter can be used to convert an
integer, a floating-point number, an object number, a string containing a
floating-point literal, or an error value into a floating-point number.
-- The functions `min()', `max()', and `abs()' now work analogously on
floating-point numbers. If `min()' or `max()' are passed some integers and
some floating-point numbers in the same call, they raise E_TYPE.
-- ************ The function `sqrt()' no longer accepts integer arguments; its
*** NOTE *** argument must now be a floating-point number and its result
************ will always be such a number. The old and nearly useless
behavior of a call to `sqrt(X)' can be simulated with the new expression
`toint(sqrt(tofloat(x)))'. (I did this because it made no sense for the
various new math functions, like `sin()' and `exp()', to map integer
arguments to integer results and it seemed important to keep all of the math
functions consistent. It's my guess that there's very little existing code
that uses the old `sqrt()' function, so that this will not represent much of
an upgrading burden; I'm sure you'll let me know if I'm wrong...)
-- Added a new expression type `X ^ Y', which returns X raised to the power of
Y. If X is an integer, then Y must be an integer as well. If X is a
floating-point number, then Y may be either integer or floating-point.
(Yes, maybe this is inconsistent with the complete lack of
coercions described above; feel free to try to argue me around
to a position you think is better.)
-- Added the following new functions:
sin(X) sine of X
cos(X) cosine of X
tan(X) tangent of X
asin(X) arc-sine (inverse sine) of X in range [-pi/2, pi/2],
for X in range [-1, 1]
asin(X) arc-cosine (inverse cosine) of X in range [0, pi], for
X in range [-1, 1]
atan(X [, Y]) arc-tangent (inverse tangent) of X in range
[-pi/2, pi/2] if Y is not provided, or of Y/X in range
[-pi, pi] is Y is provided
sinh(X) hyperbolic sine of X
cosh(X) hyperbolic cosine of X
tanh(X) hyperbolic tangent of X
exp(X) exponential function e^X
log(X) natural logarithm ln(X), for X > 0
log10(X) base 10 logarithm of X, for X > 0
ceil(X) smallest integer not less than X, as a floating-point
number
floor(X) largest integer not greater than X, as a floating-point
number
All of these functions take only floating-point arguments and return
floating-point results. They raise E_INVARG if their argument is out of
range or E_FLOAT if the result overflows. On underflow, they return zero.
[Incredibly, some systems also print an error message on standard error if
the argument is out of range; there isn't anything I can do to stop it, so
just ignore such messages in the log.]
(This ends the floating-point changes.)
-- Changed the function `random()' to allow calls with no arguments; this is
effectively the same as passing in the largest MOO integer.
-- Fixed a long-neglected loophole in tick-counting; the following constructs
all newly take one tick now:
-- exception-handling expression: `expr ! codes'
-- exception-handling statement: try ... except (expr) ... endtry
-- cleanup statement: try ... finally ... endtry
-- scattering assignment: {A, B, @C} = X
-- Fixed a bug in the decompiler that could panic the server if a WHILE loop
was the first thing inside the ELSE part of an IF statement. (Thanks to Ron
Stanions for reporting this.)
-- Added documentation of what's required in order to add a new MOO value type
to the server; see the new file AddingNewMOOTypes.txt.
Version 1.8.0beta2, 10 February 1996
-- Made the source of standard messages printed on connections dependent on the
object acting as listener for connection. Connections received on a
listening point handled by an object OID print messages taken from
properties on OID.server_options or, if that does not exist, then from
$server_options. Thus, for example, after doing `listen(#987, LP, 1)', a
connection that arrives on the listening point LP and then is disconnected
will have the text of #987.server_options.boot_msg printed to it; if #987
does not have a property named "server_options", then
$server_options.boot_msg will be used instead, just as before. (Thanks to
Michael Brundage for persistently reminding me to add this feature.)
-- Fixed bug in clear_property(OID, PNAME) where it could crash the server if
the object OID defines the property PNAME. (Thanks to Don Schwarz for
reporting this.)
-- Fixed bug in floating-point number parsing where literals with exponents but
no decimal point (e.g., `1E6') were treated as integers. (Thanks to
H. Peter Anvin for pointing this out to me.)
-- Slightly liberalized parsing of floating-point numbers not to require any
digits before the decimal point if there are digits after it. Thus, `.02'
is now parsed identically to `0.02'. (Thanks to Kipp the Kid for suggesting
this.)
-- Fixed bug where `kill_task(task_id())' could panic the server. (Thanks to
Brian Buchanan for reporting this.)
-- Fixed handling of floating-point exceptions on systems that support the
SVID3 `matherr()' handler function. (Thanks to H. Peter Anvin for reminding
me to do this.)
-- Added built-in function `trunc(X)' which returns the floating-point number
obtained by truncating the floating-point number X at the decimal point.
For X < 0.0, this is equivalent to `ceil(X)'; otherwise, it is equivalent to
`floor(X)'. (Thanks to HPA again for suggesting this.)
Version 1.8.0beta3, 18 February 1996
-- Once more liberalized parsing of floating-point numbers, this time not to
require any digits *after* the decimal point if there are digits *before*
it. Thus, `32.' is now parsed identically to `32.0' and `32.e1' is parsed
identically to `32.0e1'. (Thanks to Brian Buchanan for pointing out this
remaining inconsistency with the behavior of tofloat().)
-- Added built-in function `value_hash(X)', which returns a 32-character
hexadecimal string equal to `MD5(toliteral(X))', where MD5 is a well-known
cryptographically secure hash function. In essence, if
`equals(value_hash(X), value_hash(Y))' then you can be pretty damned sure
that `equals(X, Y)', even if the two hashes were computed on different
machines. Also added `string_hash(X)', which just computes the MD5 hash of
the contents of the string X, and `binary_hash(X)', which does the same for
the binary string X.
-- Fixed `toint(X)', where X is a string denoting a floating-point number, to
return the same as `toint(tofloat(X))'. (Thanks to Kai Storbeck for
suggesting this.)
-- Fixed bug in `floatstr(X, PREC)' where PREC was allowed to be negative.
(Thanks to `slayer' for reporting this bug.)
-- Changed parser to replace calls of the form `call_function("foo", ...)'
with simply `foo(...)' if `foo' is a known function. (Thanks to Don Schwarz
for suggesting this.)
Version 1.8.0, 9 March 1996
-- Verbs defined on #0 are no longer subject to the wiz-only permissions check
on built-in functions generated by defining $server_options.protect_FOO with
a true value. Thus, you can now write a `wrapper' for a built-in function
without having to re-implement all of the server's built-in permissions
checks for that function. (Thanks to ThwartedEfforts for bringing this idea
back to my attention.)
-- If a built-in function FOO has been made wiz-only (by defining
$server_options.protect_FOO with a true value) and a call is made to that
function from a non-wiz verb not defined on #0 (that is, if the server is
about to raise E_PERM), the server first checks to see if the verb #0:bf_FOO
exists. If so, it calls it instead of raising E_PERM and returns or raises
whatever it returns or raises. (Thanks to yduJ and ThwartedEfforts for
putting this idea together.)
-- Added the source code for `pgperf' to a subdirectory of the release `tar'
file. You need `pgperf' in order to recreate the keyword-lookup table used
in the MOO-code parser should you ever change the set of keywords accepted
by the parser. Check out the `pgperf/README' file for (slightly) more
information.
-- Backed out the change in 1.8.0beta3 where the parser was replacing calls of
the form `call_function("foo", ...)' with simply `foo(...)' if `foo' was a
known function. While this change made sense for porting whole databases
between server versions, it made it harder to port single verbs or objects
from place to place. (Thanks to Richard Connamacher for being persistent
enough to finally make me see this point.)
-- Added new built-in function `connection_option(CONN, OPTION)', which returns
the current setting of the option OPTION on the connection CONN. It raises
E_INVARG if OPTION is not a valid option for that connection. (Thanks to
Don Schwarz for pointing out how inconvenient it could be to have to use
`connection_options(CONN)' for this.)
-- Changed the format of binary strings to have non-printing characters
represented in hexadecimal instead of octal, for improved concision and
consistency with the output of the `*_hash()' functions. (Thanks to
ThwartedEfforts for suggesting this.)
-- Made optional the argument to `shutdown()'. (Thanks to Richard Godard for
suggesting this.)
-- Moved the DEFAULT_PORT and DEFAULT_CONNECT_FILE options into `options.h'.
(Thanks to Richard Godard for suggesting this, too.)
-- Fixed line numbers reported in load-time warnings about renaming old uses of
new keywords. (Thanks to Ron Stanions and others for noticing this bug.)
Version 1.8.0p1, 11 March 1996
-- Fixed bug in the transition of binary strings from octal to hex. (Thanks to
Adrian Irving-Beer for reporting this.)
-- Fixed longer-standing bug in decode_binary() where bytes over 127 would be
decoded as negative numbers on some systems.
Version 1.8.0p2, 18 March 1996
-- Fixed off-by-one bug in set_verb_code() that failed to type-check the last
element of the CODE list. (Thanks to Kipp the Kid for reporting this.)
-- Reordered printing of start-up messages and the registration of built-in
functions to allow function registrars to more aesthetically print out their
own such messages. (Thanks to Richard Godard for suggesting this.)
-- Minor configuration change to improve chances of finding an ANSI C compiler
on SGI IRIX 5.2 systems. (Thanks to John Wilson for the suggestion here.)
-- Changed the behavior of calls to built-in functions made wiz-only via
$server_options; for a call to a protected function FOO from an object other
than #0, if $bf_FOO() exists, it is always called instead of the built-in,
*even for wizards*. (Thanks to Roger Crew for helping me beat this horse
thoroughly dead.)
-- Fixed conversions of floating-point numbers into strings. The default
conversion, used in tostr() and toliteral(), uses the exact number of
decimal digits of precision supported in the underlying representation
(i.e., DBL_DIG in ANSI C). The writing of the database file uses four more
digits to be certain of extracting all of the available information. The
floatstr() function allows specification of any non-negative number of
digits up to four more than DBL_DIG. (Thanks to Brian Buchanan for
reporting the problem and to H. Peter Anvin for suggesting the appropriate
fix.)
-- Fixed task_stack() to test for the given task ID being valid (i.e., naming a
suspended task) *before* testing that the caller's permissions are
sufficient. (Thanks to Judy Anderson for reporting this.)
-- The idle un-logged-in connection timeout can now be set in the database.
For a connection accepted by a listening object L, the server checks for the
property `connect_timeout' being defined on either L.server_options or, if
L.server_options is not defined, on $server_options. If the
`connect_timeout' property exists and is a positive integer, that is the
number of seconds to use for the timeout. If `connect_timeout' exists but
is not a positive integer, then no timeout is used. Finally, if
`connect_timeout' does not exist, the old default of 300 seconds is used as
the timeout. (Thanks to Richard Godard for prodding me into this.)
-- Setting the .programmer or .wizard bit on an object no longer raises
E_INVARG if the object is not a player. (Thanks to Roger Crew for
suggesting this.)
-- Fixed `;' command in emergency wizard mode so that such tasks now have
legitimate task IDs.
Version 1.8.0p3, 7 April 1996
-- Fixed panic when a line of input existed from a logged-in but invalid object
(as is made possible by force_input()).
-- Fixed `usage' message to include the new `-e' flag.
-- Fixed bug where it didn't work to rename a property changing only the case
of the letters. (Thanks to Brack for reporting this.)
-- Exported `get_server_option()' to `server.h', for possible use by server
extensions. (Thanks to Richard Godard for suggesting this.)
-- Added `-l LOGFILE' command-line option, to allow specifying the destination
for logging messages separately from redirecting the standard-error stream.
This is most useful when intending to `continue' from emergency wizard mode.
(Thanks to Seth Rich for suggesting this.)
-- Moved test for a built-in function being `protected' and the possible call
to $bf_FOO() before the test for the correct number and type of arguments.
This allows a $bf_FOO() overriding verb to take different arguments from the
built-in it's overriding. (Thanks to Brian Buchanan for suggesting this.)
-- Added built-in function `flush_input(CONN [, SHOW_MESSAGES])' that does the
same thing as the user CONN typing their defined `flush' command. That is,
it removes all pending lines of input from CONN's queue and, if
SHOW_MESSAGES is provided and true, prints a message to CONN listing the
flushed lines, if any. This makes it possible to flush an input queue from
within verb code. (Thanks to Seth Rich for pointing out this issue.)
-- An un-logged-in connection can now only `log in' (via $do_login_command()
returning a valid player object) if the network connection is actually still
open. In particular, the following cases, which used to cause panics, now
have no effect aside from the obvious calls to $do_login_command():
1) Something opens a connection, sends input that would log it in, and
then closes the connection before the server has had time to run the
$do_login_command() task that would log it in.
2) Someone executes `force_input(CONN, LINE)' where CONN is negative
and LINE would log in an un-logged-in connection.
(Thanks to Dave Kormann for reporting this bug.)
-- Changed the definition of value_bytes() to make it more intuitive to users.
Now, value_bytes() always returns a non-zero value, since *every* MOO value
requires *some* memory in order to be represented. (Thanks to several
members of the MOO-Cows mailing list for the discussion that led to this.)
-- Fixed disassemble() to properly print the mnemonic for the PUSH_LABEL
instruction. (Thanks to Brian Buchanan for reporting this bug.)
Version 1.8.0p4, 18 April 1996
-- Fixed potential memory-smash bug in the parsing of a misplaced `$'
expression. (Thanks to Brack for reporting this.)
-- Fixed odd choice of errors raised by chparent(A, B); it used to raise E_PERM
when A was not valid or B was neither valid nor equal to #-1. It now raises
E_INVARG in these cases.
-- Fixed *really* nasty bug in the way the $bf_FOO() overrides for protected
built-in functions were implemented. The first (and least nasty!) effect
was a potential memory smash and/or server panic if you killed a task that
was in the middle of a call to some $bf_FOO(). (Thanks to slayer@kaiwan.com
for reporting this effect of the bug and thereby ruining the rest of my day,
spent tracking down and trying to cope with the *rest* of the effects.) The
more subtle and terrifying effect is given below.
********** There was a serious bug in versions 1.8.0 through 1.8.0p3 of the
** NOTE ** server that could cause a database written by such a server to be
********** read back in incorrectly and, in some cases, *undetectably*,
causing one or more bad effects outlined below. One of the potential
effects, perhaps the most serious one, *is* detectable and fixable; release
1.8.0p4 of the server detects and fixes this problem during loading. If it
discovers/fixes an instance of the problem, it also prints a warning message
into the log.
----------------------------------------------------------------------------
NOTE PLEASE: As described below, most databases will NOT be at risk from ANY
of the potential problems. I only describe them in this much detail to make
sure that all POSSIBLE cases are disclosed.
----------------------------------------------------------------------------
In a nutshell, the problem is that the server can write out a database
file that, on reloading, makes it look as if an overridden built-in function
(i.e., one that is made wiz-only by a $server_options.protect_FOO property
and then overridden by a $bf_FOO() verb) *made a verb call* to $bf_FOO()
instead of being *replaced* by that verb. This would be in the saved state
of some task that was suspended at the time the database file was written.
This means that, after the task resumes and the call to $bf_FOO() returns,
the built-in function implementation could be re-entered in a very confused
state. This confusion could have the following effects:
1) A function that never actually calls a verb (i.e., almost any of the
current built-in functions) could mistake $bf_FOO()'s returned value
for an argument list and smash memory all over the server.
(Fortunately, this is the detectable and fixable case mentioned
above; therefore, this effect cannot happen under 1.8.0p4.)
2) The eval() function could end up wrapping an extra {1, ...} around
the correct returned value. That is, instead of returning {F, V} as
it should, it would return {1, {F, V}} in this case.
3) The create() function could return a different object number from
the correct one; in particular, it could return an invalid or even
negative object number.
4) The recycle() function could recycle the wrong object, without
checking permissions and without calling that object's :recycle
verb, and/or leave the correct object unrecycled.
5) The move() function could move the wrong object to the wrong
destination and/or leave the correct object where it was.
If you reboot your server under 1.8.0p4, I believe that effect (1) above is
completely prevented and that, in effects (4) and (5), the function in
question is extremely unlikely to operate on the wrong object. If you are
actually at risk from any of these effects (see below), then by far the most
likely cases are as follows:
1) Not possible.
2) The eval() function will have the full effect given above.
3) The create() function will have the full effect given above.
4) The correct object will not actually be recycled, even though its
:recycle verb will have been called.
5) The correct object will not actually be moved, even though the
destination's :accept verb will have been called.
>>> NOT ALL DATABASES ARE AT RISK FROM THESE EFFECTS <<<
For a database to be at risk, all of the following must be true:
a) One or more of the functions eval(), create(), recycle(), or move()
must have been made wiz-only via the $server_options.protect_FOO
property, for the appropriate FOO.
b) Such a function must have been overridden by a $bf_FOO() verb.
c) It must have been possible for the code of the $bf_FOO() verb (or
any code it calls) to call suspend().
d) The database file must have been written to disk (i.e., either via a
checkpoint or a shutdown or a panic) during the time that (a) and
(b) were true and while the call to $bf_FOO() was suspended.
e) You must be restarting your server from the database mentioned in
point (d).
Even in a database that is at risk, it is only at risk for the effects of
the specific function(s) for which all of points (a) through (e) are true.
For example, if you have never protected and overridden the move() function,
then your database is not at risk for effect (5).
These facts imply that, if your server might be at risk but is still
running under 1.8.0p3 or earlier, you may be able to remove the risk before
shutting down, by making one or more of points (a) through (e) false. For
example, you might be able to stop overriding one of the functions in
question and then wait until there are no longer any suspended tasks inside
calls to the appropriate $bf_FOO() verb.
>>> WARNING: Do NOT attempt to kill such a suspended task in order to
remove the risk from your system; this would trigger the bug
mentioned at the top of this release note, quite possibly
panicking your server.
Naturally, I am dismayed at both the existence of this bug and at the fact
that I cannot guarantee even those not yet bitten by it that they can avoid
eventually losing. This is the first time I can recall this sort of thing
happening in the entire time LambdaMOO has existed. Just my luck, just as
I'm about to retire... :-(
Version 1.8.0p5, 12 May 1996
-- Fixed two memory leaks, one triggered every time `match()' was called with a
malformed pattern and the other every time `add_verb()' was called at all.
(Thanks to Gustavo Glusman for putting me onto the trail of the first one.)
-- Fixed mis-initialization of the value of the "client-echo" connection
option. (Thanks to Wisquatuk for reporting this.)
-- Fixed backward-compatibility problem in 1.8.0 where the semantically
erroneous but syntactically correct expression
-"a string"
stopped even compiling, let alone running. (Thanks to Seth Rich for
pointing this out.)
-- Fixed another backward-compatibility problem in which the (again)
semantically erroneous but syntactically correct expression
36789.foo
(an understandable typo for #6789.foo) stopped compiling when found in a
database made by a pre-1.8.0 server.