FlushFileBuffers:
- In spite of what MSDN pretends, it appears that FlushFileBuffers flushes console input buffer when it's called with a console handle. Then, do it here as well.
- Get rid of SetLastErrorByStatus here...
svn path=/trunk/; revision=50980
- RtlpInsertUnCommittedPages: Don't rely on linked list data in an UCR Descriptor after destroying it.
No functionality change because the linked list data was still "as expected", but not something we want to rely on.
- RtlpCreateUnCommittedRange: Fix a typo that caused the head of UCR Segment list of the Heap to be treated as an UCR Segment header.
Side effects of the typo: When this list wasn't empty, the (fake) UCR Segment it described was grown to contain more UCR descriptors.
Due to the data involved, this always happened to be the initial Heap Segment, which contains the Heap Header.
Writing the new UCR descriptors caused the Heap Header and trailing Heap Entries to be partially corrupted.
svn path=/trunk/; revision=50978
- Use one instead of two critical sections to synchronize service start and control actions.
- Fail, if an already running service is started again.
svn path=/trunk/; revision=50977
- Fix a buffer overflow (overread) when adding a locale key to the registry. The history of this bug is funny:
1. Eric wrote the code, which sets a key of REG_SZ type, as 4 widechars plus terminating zero, but passes 8 as the bytesize of the buffer. It's not fully correct (a terminating zero is absent from the bytesize of the buffer, but MSDN doesn't specify if it should be added or not, and hardcoding "8" is not the best idea too) but not dramatic. That was revision 9596, 7 years ago.
2. Lentin notices something is not right in this code, and decides to "fix" it by multiplying that same hardcoded value by.... guess what? sizeof(PWCHAR)! That is, size of a pointer, which on an x86 would be 4 bytes. Massive out of bounds access obviously happens. That was revision 31642, 3 years ago.
3. Very soon Colin reshuffles and improves the code based on patch #2635, however the problem still goes unnoticed (r31655+).
See issue #5810 for more details.
svn path=/trunk/; revision=50968
- Missed the clearing of the clip region for the window at the end of of its life. The mouse is looking for the window, doing so, scanning through all of them, even the ones that are at deaths door.
svn path=/trunk/; revision=50962
- Fix the select of the clip region for a window. Apologies to Rafal Harabien select works like select, also don't
assume the original author is always right.
svn path=/trunk/; revision=50961
Revert part of r50941. First its architecturally unclean to lock a DC in a USER function, 2nd its a bug to keep the lock while sending a message.
svn path=/trunk/; revision=50959
- Prevent code duplication by using vfatAttachFCBToFileObject() for volume opening, instead of rewritting the whole function.
- Properly check requested disposition when opening a volume.
This fixes bug #5839. Trying to exec \\.\C: in explorer run dialog now ends with an error, as it does on Windows.
svn path=/trunk/; revision=50958
Patch by Thomas Faber:
When the Edit receives WM_KILLFOCUS, it will notify its parent (the ListView)
of losing focus, which in turn will send WM_CLOSE to destroy the edit control.
This will cause the edit to receive WM_DESTROY and free the EDITSTATE.
When control returns to the WM_KILLFOCUS handler, this would call
EDIT_UnlockBuffer on the now invalid EDITSTATE.
Fix this by checking the validity of the EDITSTATE before calling EDIT_UnlockBuffer.
Fixes explorer crash, when cancelling file renaming.
See issue #5895 for more details.
svn path=/trunk/; revision=50956
- Update path.c code to Wine 1.3.14. Fixes a buffer overwrite happening in RtlDosSearchPath_U() when invoked by "shell32_winetest.exe shlexec".
- RtlDosPathNameToNtPathName_U remains unsynced.
- Author names added to the header of the file.
See issue #5964 for more details.
svn path=/trunk/; revision=50954
- Test behaviour of SetDIBits for 1bpp bitmaps.
- Add small test to GetPixel just to verify that SetDIBits doesn't say BS.
svn path=/trunk/; revision=50950
- in 1bpp bitmaps, 0 means white. Take that into in SetDIBits
- fix a fixme in XLATEOBJ implementation
- remove useless field from ROS_DCINFO
Fixes fox audio player GUI :-) Enjoy!
svn path=/trunk/; revision=50949
- Raster operations in user mode are on higher bytes, whereas they are on lower bytes for drivers. Try to clarify this situation.
- Add sanity check about what was said previously.
- Implement masking in EngBitBlt
- Rewrite NtGdiMaskBlt accordingly
- Realize the palette when selecting it into a device DC.
- When applying raster operation, do so only on 24 bits, we don't support alpha channel in win32k
This fixes VLC pink icons, Timo's MaskBlt tests and probably a lot of other things.
[SHELL32]
- Use correct (?) raster operations for drawing sjortcuts.
Also note that now NtGdiMaskBlt locks the device contexts : this should avoid some race conditions, such as icons drawn on top of windows and the like.
Win32k sucks less.
So does reactos.
Dedicated to Timo. Sleep well, mate.
svn path=/trunk/; revision=50941
Fix calculating of ScanLines and source point in NtGdiGetDIBitsInternal for top-down bitmaps to more closely mimic win xp behavior. (Timo - high five for your test app)
See issue #5524 for more details.
svn path=/trunk/; revision=50935
- Clarify the meaning and the use of 0, 1 and 2 for user window regions, wine never did. Noticing patches not corresponding in correctness and understanding, leaving potential crashes in the DCE code, which depend on these types.
- Patch based on Rafal Harabien patch, fixes wine msg SetWindowRgn test, see also bug 5959.
svn path=/trunk/; revision=50933
