Errm... can't remember exactly what I've done other than this, but it now
catches SIGUSR2 when the wrappers are off so you don't accidentally kill
your MW session by attempting to look at the wrappers while they're turned
off... (Coz it got annoying!)

 - Steve

Fixed more leaks in add.c

Fry

Wrappers now do rl_copy_text()
Fixed mem leak in add.c (the buffer was never free()d).
Errm - I think that's all...

 - Steve

Latest bug-fixes

Fry

I think that's all the memory bugs fixed - how about this :)
------
talk{0}-* !quit
SHUTDOWN
Current allocated memory not including memory allocated by wrappers (most
recent at top):
Total: 0 bytes
Total malloc() calls during runtime: 6085
Total calloc() calls during runtime: 605
Total realloc() calls during runtime: 0
Total free() calls during runtime: 13804
Total strdup() calls during runtime: 7093
Total outstanding memory blocks: 0

Doing garbage collection
[firefury@sucs firefury]$
------

I've put in free()s at the following locations:
new.c: lines 83 & 103
read.c: line 252

These should fix the following leaks:
read.c: line 250 - 3 bytes allocated at 0x80b5008 using strdup() at time 951783524
new.c: line 79 - 64 bytes allocated at 0x80b43c0 using malloc() at time 951783524

I have also updated the wrappers code - it now tells you when you try and
free() something it doesn't know about (this might give you a bit of a
warning about a millisecond before a SEGV <grin>).  calloc() is now
supported (this accounts for some of the excess free()s fry found).
Free(NULL) now supported - this accounts for the rest of the excess free()s.
The outstanding block counter now actually counts the number of records in
the list, instead of relying on the counters, so it is accurate.

Unless anyone can find any other leaks or bugs for the moment, I recommend
releasing this as the live version.

 - Steve

Oooodles of leak-removing fixes.

Fry

Latest Leak fixes.

Theres only script.c (and poss tscript.c - havent checked) left now, and I
think I know where at least half of those are :)

Fry

OK, I've spent *ages* tonight working on fixing leaks in mw.

I've removed *all* the leaks from rooms.c, colours.c, strings.c etc..., the
only files left to look at are init.c, and script.c (or tscript.c or
whatever).

NOTE: one leak was mine in the screen_height/width detection stuff, one was
      finnw's in the stack_msg routines or whatever, and the other hideous
      amounts were all arthurs :) - see the Devel folder in mw for details

Fry

Added and fixed the 'tidyup' command, and 'topten' command updated again
too.

Fry

Improved the readablity of ThreadState() and DoScript() in tscript.c
Removed the testing operations from ThreadState().  It now just modifies the
flags.
Restored some dependencies on bb.h in Makefile (those that caused conflicts)

Integrated the 'topten' command, and updated it from its *evil* old code to
something remotely resembling ANSI C.

Fry

also stopped carrying the devel binary in the archive.

added new #def RELEASE
only to be used when doing final compiles,
if not defined, versions are marked as development.

use as 'CFLAGS=-DRELEASE make'
(sorry for earlier 'make -DRELEASE' type confusions)

wrappers, that was a really messy #def you did there.
you can now do 'make -DWRAPPERS' rather than having to
change the makefile (also an option)

also rearranged Makefile slightly so you can now just
uncomment sections that you want turned on,
i.e. WRAPPERS, Electric fence, gdb support. etc.

Also 'static'ed the global variables in wrappers.c that are
not used anywhere outside that function, you should do this
to the approporiate functions too.

Argh, sorry guys, I forgot to add the new wrappers files <doh>

 - Steve

Written a fairly advanced set of wrappers to (hopefully) help track
memory leaks...  They wrap malloc(), realloc(), free() and strdup().
If USE_WRAPPERS is set to 1 in the Makefile, the wrappers are used, if
it's set to 0, they aren't, although they seem pretty fast, so it should
be fine to keep them turned on during development.  Information is stored
in a linked list:

malloc() and strdup() - these prepend streight to the head of the list,
so should be pretty fast all the time.

realloc() and free() - since the most recent allocations are at the head
of the list, these should be fairly fast for memory that has just been
allocated (the stuff that is mostly going to be free()d or realloc()ed.
In addition, realloc() keeps a history of past realloc()s for each memory
block.  When free() is called, this history is nuked, together with the
information about it ever being allocated.

If you send SIGUSR2 to the process, a list of currently allocated memory
blocks and any realloc() history will be printed.
Also, if you start milliways with -exitinfo, this list will be printed
just before exit (ideally, this list should be empty, showing that
everything has been free()d, but in reality, Milliways generates a few
hundred lines of information, so use this option with care).

If the wrappers are turned on, a garbage-colleciton routine will clean
up when the process exits - not really necessary under Linux, but it's a bit
nicer than just leaving the OS to do it.

These wrappers are effective in anything that #includes "wrappers.h"
(currently everything except red.c).  If you create a new c file, it might
be wise to #include "wrappers.h".

 - Steve (bugger, that's the longest CVS message ever :)

Latest memory leak fixes :)

Fry

Umm.. stuff about tab-completion - fixed a memory leak and updated it abit

Fry

Made 'ExecEvent' return a value incase function to exec doesnt exist...

Well... it may be useful...

Fry

Made a couple more changes to incoming.c

Finnw> for now I've remarked the partial message thingy, cos its irritating,
and ignores the rest of it anyway (only sends max of 1024 chars now anyway)

Fry

Fixed hideous amounts of buffer overruns, and stuff.

Added a constant MAXTEXTLENGTH in bb.h for any function to use, so I dont
have to keep changing 1024 etc.. if it needs changing. Recommend using this
if calling any functions that 'fiddle' with strings, eg, apply_gag,
quotetext, and the send routines, talk_send, broadcast etc.. - reason being
is that *THEY* all use it, so if you dont it'll get truncated anyway if too
big! :P

Fixed other wacky stuff, all to do with sprintf, strcpy's etc.. with dont
check string sizes.

Now seems to work faster, with less memory leaks, and I haven't been able to
crash it yet (.e crosses fingers :)

Fry

Fixed a couple of bugs in the makefile, and removed firefury's snapshot
thingy (firefury - just do a 'tar' like a normal person).

Plus, various buffer overruns fixed, if I haven't already submitted them.

Fry

Made following changes to the Makefile:

Removed fry's "beep" rule and integrated it into the "bb" and "red" rules so
it now works in a parallelised make.

Added "srcsnapshot" rule - this takes a snapshot of the src directory (i.e.
the "." directory and all subdirectories).

Added "snapshot" rule - this takes a snapshot of the whole BBS (i.e. the
".." directory and all subdirectories).

The snapshots are tgz files with timestamped filenames.  The base dir is
defined by $(SNAPSHOTBASE), and shouldn't be inside the devtree (ATM it's
set to "../../", which is safe).

 - Steve

Improved the speed of the threaded version

Added an extra buffer overrun check in incoming.c

Fixed the buffer overrun problems in incoming.c

Fry

I've fixed the beeping if you had beeps on, and looked at 'ignorelist'.

God knows whatelse I've done - if any files are altered, its probably minor
spelling fixes :)

Fry

Fixed the 'sayto' log to log who you said it to.

Fry

Corrected makefile *again* (tnarg/firefury, plz stop doing this!)

Fry

Updated the login banner back to the original version rather than FireFury's
oooooooh-so manky version :P

Fry

update sayto command so the recver knows that the message was only sent to
him

Optimised idle() and ScriptStep() a bit in tbbs - they can still be
improved though.  eval_arg() takes up the most time now (~20% of total
run time)

Fixed 'whisper' and 'sayto' commands, so they detect if user is also logged
on, and not just whether they exist.

Changed the 'summon' help text to something more recent.

Fry

Ermm.. added '.sayto' command in talker (needs 'raw' priv)

Changed force commands in incoming.c to cope with finnws new proper message
stack, rather than single buffer (muliple stacked forces now available
anyone? now all we need is file access, and firefury can write a script to
multiple force ppl to write a new script, load it, and execute it ->
'chainmail' scripts :)

Fry

Added the 'incoming.o' make definition into 'Makefile'

Anything else included is just minor/touched - is it just me, or has CVS
started showing *all* files in the 'CVS: Modified Files:' bit, rather than
just the modified ones? 'cos its difficult to remember what I changed unless
I see the filenames there :/

Fry

Added strings.h - now essential for compiling strings.c and new.c

split handle_mesg (formerly part of incoming_mesg) into 10 separate
functions.

Fixed a bug in Parse.c that segfaulted if you typed a single '.' on
a line

Extended the 'stack_str' stack.  Its now a real stack not just a
single-line buffer.  Also added a flags field.  I plan to implement a
safer 'boardexec' with it in the threaded version.

modified the alarm system slightly.  It used to occasionally deadlock
during 'sleep' (I'm not sure why).  Now it puts a limit on the waiting
time.

Threaded version should now compile OK.  Some instructions (eg input,
boardexec) are still not thread-safe.  For the moment do 'busy on'
before using them.

Added 'sayto' function finally.

Allows perfect 'say' to a specific user. If user has global on, then shows
room the sender is in.

Needs 'raw' priv to use.

Fry