Re: Rob Shearman : msvcrt: Free memory allocated in TLS slot on module unload as well as thread exit.
On 12/14/2009 18:51, Alexandre Julliard wrote:
Module: wine Branch: master Commit: c20868e0a2f7af909cf8af2877ae8b024fa6d11e URL: http://source.winehq.org/git/wine.git/?a=commit;h=c20868e0a2f7af909cf8af2877...
Author: Rob Shearman<robertshearman(a)gmail.com> Date: Mon Dec 14 14:13:57 2009 +0000
msvcrt: Free memory allocated in TLS slot on module unload as well as thread exit.
---
dlls/msvcrt/main.c | 27 +++++++++++++++------------ 1 files changed, 15 insertions(+), 12 deletions(-)
It's a partial solution I suppose? I works for a one thread only that unloads module.
2009/12/14 Nikolay Sivov <bunglehead(a)gmail.com>:
On 12/14/2009 18:51, Alexandre Julliard wrote:
Module: wine Branch: master Commit: c20868e0a2f7af909cf8af2877ae8b024fa6d11e URL: http://source.winehq.org/git/wine.git/?a=commit;h=c20868e0a2f7af909cf8af2877...
Author: Rob Shearman<robertshearman(a)gmail.com> Date: Mon Dec 14 14:13:57 2009 +0000
msvcrt: Free memory allocated in TLS slot on module unload as well as thread exit.
---
dlls/msvcrt/main.c | 27 +++++++++++++++------------ 1 files changed, 15 insertions(+), 12 deletions(-)
It's a partial solution I suppose? I works for a one thread only that unloads module.
The real motive for the patch was to free the memory on process exit to silence valgrind warnings. In reality, I don't believe msvcrt would be loaded dynamically by an app, but if we did want to cope with this the code could be changed to maintain a linked list of the objects allocated for this purpose or to use a private heap. -- Rob Shearman
On 12/14/2009 20:35, Rob Shearman wrote:
2009/12/14 Nikolay Sivov<bunglehead(a)gmail.com>:
On 12/14/2009 18:51, Alexandre Julliard wrote:
Module: wine Branch: master Commit: c20868e0a2f7af909cf8af2877ae8b024fa6d11e URL: http://source.winehq.org/git/wine.git/?a=commit;h=c20868e0a2f7af909cf8af2877...
Author: Rob Shearman<robertshearman(a)gmail.com> Date: Mon Dec 14 14:13:57 2009 +0000
msvcrt: Free memory allocated in TLS slot on module unload as well as thread exit.
---
dlls/msvcrt/main.c | 27 +++++++++++++++------------ 1 files changed, 15 insertions(+), 12 deletions(-)
It's a partial solution I suppose? I works for a one thread only that unloads module.
The real motive for the patch was to free the memory on process exit to silence valgrind warnings. In reality, I don't believe msvcrt would be loaded dynamically by an app, but if we did want to cope with this the code could be changed to maintain a linked list of the objects allocated for this purpose or to use a private heap.
Sure, that's ok, most tests (if not all) are single threaded. Already discussed some time ago, no need in such complication here, just exit as fast as possible is preferable.
participants (2)
-
Nikolay Sivov -
Rob Shearman