Now the new sysreg2 just needs to be set up on the Buildslave and we might finally be able to remove all those "Skip Test xyz, because ROS sucks" hacks.
svn path=/trunk/; revision=40157
- Use NULL as the lpVerb to get the default action ("open" is not necessarily the default)
- Allow passing parameters besides just the file name
svn path=/trunk/; revision=40148
rosautotest
- Rewrite rosautotest in C++
Should increase maintainability and expandability, since most of the functionality is encapsulated in classes and there exist some abstract classes for further enhancements (i.e. new test types).
Furthermore, due to the usage of STL strings, we don't need x lines anymore just for building a string out of several small parts.
- The new codebase made it fairly easy to implement a Crash Recovery feature based on a journal.
If you start rosautotest with the /r option under ReactOS, it will keep a journal about the tests to run and the tests already ran. In case of a crash, it can just continue with the next test in the list then.
- Add some reasonable timeouts to avoid certain hangs in case a test crashes
sysreg2
- Make the necessary changes to sysreg2 to restart the VM in case of such a crash in 3rd stage, but set a maximum number of allowed crashes as well.
Christoph, please test and review that on the Buildslave :-)
- Prepend all sysreg messages with [SYSREG] through a new function SysregPrintf, so the BuildBot aggregator script of testman can distinguish between debug output and sysreg messages.
- Put all header includes into the central header file "sysreg.h"
- Remove unnecessary libs from the Makefile
testman
- Change the testman Web Interface to show such crashes as CRASH in the Compare and Detail views.
svn path=/trunk/; revision=40147
There is a strict difference between a BRUSH and an (E)BRUSHOBJ. Please don't mix this up by giving them all names like BrushObj. Please use (proper!) hungarian notation.
Now the willing reader might actually get a clue how this stuff is supposed to work and maybe also realize how broken it is.
svn path=/trunk/; revision=40143
Make pdcattr alsways point to a DC_ATTR, either the user mode struct or the local part. Get rid of all the If (!pdcattr) pdcattr = &dc->Dc_Attr; That are not required anymore.
svn path=/trunk/; revision=40115
- My branch needs to be synced with trunk so I'm not using it right now (plus these changes aren't likely to break anything)
svn path=/trunk/; revision=40107
- Rework 8bit DIB alphablend to do alpha operations in source's palette (32bpp) to obtain much better output quality with lowest possible quality loss.
See issue #4291 for more details.
svn path=/trunk/; revision=40103
- Although RECT and RECTL are defined equal, the compiler treats them as incompatible. MS has created a lot of definitions with RECT and a lot with RECTL. So far we had to typecast them. Now with adding some defines into a win32k header file, we can finally treat them as equal inside win32k and get rid of a lot of type casts. Also use RECTL in favour of RECT internally, as this is the type that MS uses for the DDI and seems to be the more appropriate type.
- We had a lot of "const PRECT" stuff inside win32k. Note: there's difference between "const RECT *" (what you currently want) and "CONST PRECT" (which you proabably don't want). Despite the fact that CONST is not an official modifier (please stick to const) the main difference is that the former describes a pointer to a constant structure, while the latter describes a constant pointer to a modifyable structure.
- In an attempt to clean up the overpolluted IntGdi namespace, "objectify" the rect functions, moving them into their own namespace RECTL_Xxx.
svn path=/trunk/; revision=40100
- Fix overflow bug in FREE command that caused values >= 4GB to wrap around.
- A little simplification of ConvertULargeInteger function
svn path=/trunk/; revision=40094
- Make prompt code $T and %TIME% have the hours space-padded, not zero-padded.
- Allow delayed expansion in the parameter of IF ERRORLEVEL.
svn path=/trunk/; revision=40084
- Fixed initialization for Console apps with Unicode entry point. Now stuff like "cout" in <iostream> doesn't lead to instant crashes.
- Unified the name of the entry point for ANSI and Unicode. The difference is just in the library now ("mingw_main" vs. "mingw_wmain")
RBuild was changed appropriately to support this
- No more ReactOS-specific changes inside the initialization code
svn path=/trunk/; revision=40077
