On Wed, May 22, 2013 at 2:03 PM, Alan W. Irwin irwin@beluga.phys.uvic.ca wrote:
Generally I have found in still fairly limited testing that MSYS bash.exe redirection is fine on wine-1.5.30 (which I have now been able to build in 32-bit mode on Debian wheezy x86_64 thanks to the 32-bit library symlink trick discussed in my previous recent thread). However, I have just discovered a puzzling case where MSYS bash.exe redirection now fails for ctest.exe compared to wine-1.5.19.
To reproduce this regression download the Windows binary version of cmake-1.8.10.2 (which includes cmake.exe and ctest.exe) and install it. Also download and install MinGW/MSYS. In the past, I have found results depend very much on how you get into the MSYS bash.exe environment. In my case I execute
wineconsole --backend=curses MinGW_4.7.0/msys/1.0/bin/bash.exe
to do that, and then from that interactive environment execute the commands below.
Here is how I try the --version option for cmake and ctest with redirection under bash.exe.
bash.exe-3.1$ cmake --version >& cmake_version.out bash.exe-3.1$ ctest --version >& ctest_version.out
For Wine-1.5.19
bash.exe-3.1$ head *.out ==> cmake_version.out <== cmake version 2.8.10.2
==> ctest_version.out <== ctest version 2.8.10.2
For Wine-1.5.30
bash.exe-3.1$ head *.out ==> cmake_version.out <== cmake version 2.8.10.2
==> ctest_version.out <==
N.B. note the incorrect empty ctest_version.out file for Wine-1.5.30.
For both Wine-1.5.19 and Wine-1.5.30 if I don't redirect the output I simply get the following results bash.exe-3.1$ cmake --version cmake version 2.8.10.2 bash.exe-3.1$ ctest --version ctest version 2.8.10.2
So somehow bash.exe redirection for ctest (but not cmake and a small number of other commands I have tried) generates an incorrect empty file for Wine-1.5.30. I never have had trouble with bash.exe redirection for any command (including ctest) on either Linux bash or MSYS bash.exe on Wine-1.5.19. So it is likely there is some minor (but probably standard) variation in how ctest handles its stdout that is triggering this Wine-1.5.30 regression compared to Wine-1.5.19. So there might be other commands as well where bash.exe redirection doesn't work for Wine-1.5.30.
If it is already known that MSYS bash.exe redirection is sometimes problematic with Wine-1.5.30 compared to earlier versions, is there some patch I could try to see if it fixes this specific issue?
My primary motivation is developing my own free software. Therefore, I only have limited time to help figure out wine bugs although I am willing to report the ones I discover here (and on the bug tracker once discussed here to reduce bug tracker noise).
That said, it does appear that git-bisect would be an ideal way to find which commit between Wine-1.5.19 and Wine-1.5.30 caused the bash.exe redirection issue with ctest.exe. Unfortunately, I have zero experience with git-bisect. However, if somebody could give me some step-by-step instructions I could follow by rote that would be great. Ideally, those instructions would be a simple template bash script I could run unattended overnight from Linux that used git-bisect to detect the first commit between Wine-1.5.19 and Wine-1.5.30 where
ctest --version >& ctest_version.out
produces an empty file under MSYS bash.exe.
I can fill in the parts of that template script concerning the wine build for each git commit that is being tested since I have such builds (including necessary *.so links for later wine versions) largely automated now. But I am not sure about how you get into MSYS bash.exe directly (my normal method using wineconsole as above appears to be interactive rather than something you could run unattended) from a Linux bash script so I would need advice about that.
Alan
This is better suited for the forum, but: http://wiki.winehq.org/RegressionTesting#head-a7150fa43baeaab304403f27a93064...
-- -Austin