Fixes crash of mshtml_winetest:htmldoc (introduced by 716613c). Please do not ask me why.
The test started crashing with commit 716613cddc "[CMAKE] Use new policies for CMP0005 and CMP0018 (#1481)", which moved this definition from the command line to urlmon_main.h. This got later removed by commit 6c3c2e33f6 "[URLMON] Sync with Wine Staging 4.18. CORE-16441"
Our FT_Bitmap_Convert function had a hack to make the
bitmap image compatible to ReactOS. I think the hack on
FT_Bitmap_Convert should be separated from our font
engine.
JIRA issue: CORE-16047
- Add FT_Bitmap_Convert_ReactOS_Hack function, that is
based on FT_Bitmap_Convert, and split the hack of
FT_Bitmap_Convert.
- Use FT_Bitmap_Convert_ReactOS_Hack in
IntGetBitmapGlyphWithCache function instead of
FT_Bitmap_Convert.
- Modify ftfd.spec to add
FT_Bitmap_Convert_ReactOS_Hack.
The VolumePath buffer returned by IOCTL_MOUNTMGR_QUERY_DOS_VOLUME_PATH
contains one string stored as a multi-NUL-terminated string, whose
total length is given by its `MultiSzLength` member.
The DosName UNICODE_STRING just returns the (single) string as a normal
NUL-terminated string. So, we need to remove the two NUL-terminators
from the `MultiSzLength` count to retrieve the correct length.
- When trying to find a device, don't fail with STATUS_NOT_FOUND if no
associated symbolic links are present. Instead, that test was wrong,
and should go out of the search loop to do the drive-letter processing.
Addendum to commit f9f5a78715.
- In addition, when using the associated-device method, fix the list
used to retrieve the corresponding device.
- In the TryWithVolumeName: block, reset DeviceString to NULL after
freeing, so that we can correctly fail with STATUS_NOT_FOUND if no
suitable device was found, before initializing the output data.
- Use FIELD_OFFSET to correct structure and member instead of hardcoding
sizeof-s of fields until the member of interest.
- Fix a bug in MountMgrQueryDosVolumePath() where the FIELD_OFFSET used
in the entry structure size validation was incorrect.
FIELD_OFFSET(MOUNTMGR_TARGET_NAME, DeviceNameLength) is == 0 since
DeviceNameLength is the first member of the MOUNTMGR_TARGET_NAME
structure. The intended member was DeviceName.
Addendum to commit f9f5a78715.
- Limit the HarddiskVolume loop to 32 trials (instead of an unknown
number of them). Ideally the list of volumes has to be queried from
the volume manager, instead of hardcoding them (TODO).
- Verify that the buffer returned by IoVolumeDeviceToDosName() is
indeed a unicode multi-string (ends with two NULs), and, in the
case it specifies a volume GUID, that it's a valid one.
- Use %lu instead of %lx for printf-ing last-errors;
- Don't call GetLastError() within ok, or after a previous invocation,
when the aim is to retrieve the last-error set by a tested API.
Instead, store its value in a local variable, that is then used in
the ok() tests.
- Use win_skip() only for functionality that _may_ not be implemented
in Windows/ReactOS yet, but not for unexpected situations leading to
tests being skipped.
- Add comments; improve buffer size specifications.
- Reformat the file license header.
Follow the standard way.
JIRA issue: CORE-19469
- Use STDMETHODIMP and override keyword
instead of virtual HRESULT STDMETHODCALLTYPE.
- Delete CTaskBand::ProcessMessage and
CTaskBand::ContainsWindow methods.
I think they are useless.
Support Unicode on Start Button
context menu for international text
support.
JIRA issue: N/A
- Support Unicode in
CStartMenuBtnCtxMenu::InvokeCommand
method by using
CMINVOKECOMMANDINFOEX structure.
Supporting raster fonts. Using
num_faces is the generic way.
JIRA issue: CORE-16165
- Use FT_FaceRec.num_faces instead
of TT_Face.ttc_header.count in
counting the font faces in
IntGdiLoadFontsFromMemory function.
This avoids a linker error due to a duplicated symbol in ntdll and the statically linked chkstk library. This happens when the linker first resolves _chkstk from the CRT or the chkstk library (which also pulls in _alloca_probe(_16)) and then tries to resolve _alloca_probe(_16) from ntdll.
This setting can be overwritten by defining the CMake variable HOST_BUILD_TYPE.
The target build type is passed to the host build as TARGET_BUILD_TYPE CMake variable.
This improves time to build cab+iso:
MSVC x86: 37s -> 28s
GCC x86: 32s -> 16s
Follow-up of 42d5dfd3de.
- Revert "shell32: Fix ShellExecute for non-filespec paths." 0bad544aab.
- Apply "shell32: Look for the file name without extension also for the path search case." 38b6640be9.
- Clear leading/trailing whitespaces (an improvement by Doug Lyons).
- Update the comment for xlpFile buffer definition, to match the current code behaviour.
This fixes some failures for the following tests:
- shell32_apitest:ShellExecCmdLine: 12 failures less,
- shell32_apitest:ShellExecuteEx: 5 failures less,
- shell32_winetest:shlexec: crash fixed, now 13 failures as before,
- wshom_winetest:wshom: crash fixed, now 2 failures less too (0 failures).
I've also tested: the problem which was intended to be fixed by the guilty commit (CORE-19953) still remains fixed after this change.
CORE-19964
Useful for debugging.
Motivation: With SMP on x64 I found a number of instances where critical sections would be left abandoned, causing lockups. From what I can tell it was exceptions inside rpcrt4, which leave the process in a blocked state. Might or might not be related to x64 / SMP.
For real value, you still need to put checks at certain places manually, but this is not super straight forward, because there can be false positives, e.g. when a process is terminated due to an exception, where the abandoned lock is acceptable, and we have this during testing. It's difficult to 100% distinguish this from silent and very bad lock leaks.
Problematic code:
__try
{
SomeFunction(); // throws an exception with a CS held, e.g. heap code
}
__except(1)
{
DPRINT1("Oops. let's just pretend it's all ok!\n");
}
