CVS log for winex/scheduler/client.c
	ViewCVS and CVS Help

Request diff between arbitrary revisions

- fixed GetModuleFileNameA() to properly return a truncated filename error code.  GetModuleFileNameW() just calls GetModuleFileNameA(), so its error codes should be updated as well.
- fixed all the calls to GetModuleFileNameA/W() to check for small buffer and bad module errors.  Currently small buffer errors are just reported and are allowed to proceed with either an empty string or truncated string for the filename.  In the case of an error retrieving the module filename (error == 0), the calling function reports the error and returns immediately.

Refactor SHM server to handle both POSIX and SYSV IPC.
Add support for SHM server on Mac OS X.

Clarify reason for not enabling flush_key for the SHM; it's to help ensure the integrity of the .reg files

Implement RegFlushKey.
Do some minor cleanups of duplicated functions.

- disambiguate in-process tids (which all platforms have) with
  cross-process tids (which only some platforms have)
- store in-process tid in the TEB, and use that to fix thread cleanup on
  MacOS X
- modify shm server to handle platforms which don't have cross-process tids
- rename server unix_tid to unix_tid_or_pid for clarity
- implement MacOSX functionality on top of Mach to replace ptrace functionality on Linux:
   - suspend/resume thread
   - read/write process memory
   - get/set thread context

Don't bail if the USER environment variable isn't set.  Just use a default instead

Handle data payload sent with send_unix_signal, allowing it to be returned
from recv_unix_signal. Additionally, send_unix_signal (slow version) is
allowed to wake up a suspended thread; recv_unix_signal will suspend it
again once it has handled all its pending signals.

Additions to the wineserver protocol, making the send_unix_signal and
recv_unix_signal more flexible, by allowing specification of a process
as the signal destination, and the transmission of a data payload.

Fix various type confusions resulting in warnings. Initialize the process_info variable, as it could be used uninitialized if creating a process fails.

Changed the wineserver lock folder from PATH/wineserver-hostname to
PATH/wineserver-hostname-user. Also bumped the protocol version.

Added a wineserver call to set the scheduling mode for a thread, with the
choices of "normal", "internal" (using realtime priority class), or
"blocked" (exempt from scheduling).

Force Sleep() calls through the wineserver if the scheduler is enabled.

Config option for turning on the upcoming scheduler:

[Wineserver]
"Scheduler" = "Y"

Allow recv_unix_signal and send_unix_signal to be shmserver calls.

Block asynchronous signals during process startup, until signal handlers
are fully set up.

Add a wrapper for the shared_server_complete_poll() call which blocks
asynchronous signals while it's running.

Explicitly disabling the preloader must work for child processes as well.
Add an explicit check for the preloader having been disabled.

make the preloader restart itself with argv[0] being the binary it
is going to load. This gives us a sane ps output (and killall target).
Had to move the preloader's --server command argument to an env var as a result.

This adds support for prelink, execshield and a work around for the legacy va layout problem.
This is done through the implementation for a preloader.

this allows the shm wineserver to be allocated at any address, including any address that shmat gives us.
for now we've still fixed it to 0x90000000 but that will soon be configurable. set to NULL to let the kernel choose the address for us.
also changed s/int/size_t/ for shm_size in various places.

At least print the right system call so we don't have two sendmsg cases.

Portability fixes. Based off patch from Bruce M Simpson for FreeBSD.

Add profile for fast shm calls to get rough idea of cool things that are
happening in the wineserver and inter thread contention for the wineserver.

Allocate regular memory only if we cannot allocate a suitable amount
of shared memory. We guess at this amount by taking a quick guess at
the size of the registry plus a 4MB breathing room. This should resolve
issues that were reported on winex-devel.

Eject handling, take 2.  Move back to signals for requesting a client
fd close operation.  Remove EJECT APC stuff.  Add in polling of the
CD eject button, actual ejection of disks.

Check in cdrom eject handling code. This code will force the immediate
closure of all active fds on CDROM drives when the wineserver receives a
USR2 signal.  More work remains to poll the CD hardware for user eject
requests, but this patch allows several previously broken multi-cd
installers to work.

Speed up get_handle_fd operations.

Remove some client traces, make shm wineserver calls distinguishable by a
0001: vs the wineserver having 0001::
+client now completely controls if that process is to dump shm wineserver
information and +server remains exclusively for the wineserver process.

Fix REQ_resolve_rpc_endpoint so it can be reenabled. Other rpc calls are still
incorrect even potentially for non shm wineserver case.

Should fix multidisc installers by merging the ctx fd cache and the handle
cache. Also removes limitation that only pthread systems can have sych
acceleration. An a couple of other small things for fun.

It appears that the rpc code in the server uses a non standard idiom
to send and decode a number of variable arguments that doesn't play
at all with FAST_VAR_DATA. For the time being just turn off these
particular calls off so that the FAST_VAR_DATA can remain on.
This will be fixed properly later.

If we're using pthread emulation we can't accelerate WaitForMultipleObjects
because it uses malloc.

Someone should send the cut and paste thinko police after me. Soon.
Actually connect to the fd-server rather an the wineserver twice.

It appears my previous "fix" broke process spawning. Silly me. Should be
fixed now. We were recreating a new wineserver for each process and
severing the links to the old one. Not good.

Better handle a crash in the wineserver and the sockets which can remain
hanging around.

Clean up possible race condition where server deletes poll list that client
is still using.

Initial pass at small integer thread/process IDs. Needs to be tweaked
in the future to unbreak 16bit window hooks.

Start of SHM phase 2.
multi disc installers are broken and there are several bugs/races remaining
Simple benchmarking shows 5 time throughput improvement.

REQ_destroy_window doesn't actually really work right now.

First phase of SHM wineserver. Only the simple calls are accelerated.
shm calls are output to client debug channel.

* change all occurances of *_DEBUG_CHANNEL to WINE_*_DEBUG_CHANNEL
* remove the #define's for *_DEBUG_CHANNEL

Import from ReWind CVS at Dec 17, 2003.

Finally get around to changing all uses of debugtools.h -> wine/debug.h

Some warning fixes.

Added SIGUSR1, used by x11drv for GLX context migration, to the blocking
set used for wineserver communication.

Big wacking commit. Add most of the remaining pthread support.

Merge of WineHQ CVS at Jan 16, 2002.

Import of WineHQ CVS at Jan 16, 2002.

Comment out Andrew's wineserver protocol changes until I (or someone else)
figures out its purpose.

Attempted merge of Wine release 20011226. Probably breaks lots of stuff.
(Use the wine-premerge-20011226 cvs tag if you want)

Import of Wine release 20011226.
(The tree before this import has been tagged wine-premerge-20011226)

Support for recursive server calls.
The exception handler needs some fixing.

Merge of WineHQ CVS at Oct 28, 2001.

Import of WineHQ CVS at Oct 28, 2001.

Merge back the winehq-premerge-20010824 branch.

Delete the socket if we can't connect to the server.

Merge from WineHQ CVS on at Oct 4, 2001.

Import of WineHQ CVS at Oct 4, 2001.

Merge of Wine release 20010824 (only resolved CVS conflicts so far).

Import of Wine release 20010824.

Integrate updated version of Gerard Patel's deferred trace patch.  This
allows us to do -debugmsg +whatever and not generate output until
Alt F12 is pressed.  This is extremely useful for debugging purposes,
but Alexandre doesn't want it in the WineHQ tree as he seems to want
to wait for a more sophisticated and general debug control technique
being worked on by Rob Farnum at CodeWeavers.

It should be pretty straightforwards to back this patch out once
the new system is available, if ever...

 -Gav

Import of Wine release 20010510.

Import of WineHQ CVS at Mar 18, 2001.

Import of Wine release 20010305.

Import of Wine 20010216 release (tagged with wine-20010216).

Import of Wine 20010112 (at Jan 13, 2001).

Import of WineHQ CVS at Jan 5, 2001, after Alexandre applied our
winehq-ddraw-merge patch, slightly modified.
(I tagged the tree before this import with winehq-premerge-20010105,
just in case.)

Import of WineHQ CVS at Dec 29, 2000.
OK, I'll also add tags to mark the Wine 20001222 release relative to
our tree, if we still want to make diffs relative to last release...

Import of WineHQ CVS at Dec 23, 2000 (timezone=CET).
Wow, almost Christmas...

Import of WineHQ CVS at Dec 12, 2000 (really this time).

Import of WineHQ CVS at Dec 12 2000. Lots of bugfixes in there...

Import of Nov 23 winehq.

Merging in from winehq, everything except ddraw.
Still working out kinks in winelib & dinput.

Import of WineHQ 090700

Initial revision