There is a permission test that fails because MSYS uses noacl in fstab by default.

$ ctest -R 'kwsys.testSystemTools' -V
...
test 5
    Start 5: kwsys.testSystemTools

5: Test command: /f/Projects/build-cmake-msys/Source/kwsys/cmsysTestsCxx "testSystemTools"
5: Test timeout computed to be: 1500
5: SetPermissions failed to set permissions (2) for: /f/Projects/build-cmake-msys/Source/kwsys/testSystemToolsNewDir/testNewFile.txt: actual = 32768; expected = 511
5: Failed to write file: file_b
5: Failed to write file: file_b
1/1 Test #5: kwsys.testSystemTools ............***Failed    0.07 sec

I tried to change fullMask to S_IWRITE, or to 0444, but it still didn't pass. Ideas are welcome (this should not hold back the patch anyway, because it probably fails with the existing package as well).

Thanks!

This link may be the answer to the question why ctest does not work http://cygwin.com/ml/cygwin/2013-11/msg00018.html. The symlink shortcut to the /bin /lib /share directories does not work properly in the cchache. You do not need it in cygwin because there is a virtual directory /usr mounted in binary mode. So we need a better version of gcc with the fortran -static option working. And parsing in grep sed. We also have an invalid c89 mode in gcc. You can try to compile the yasm package.

The following tests FAILED:
5 - kwsys.testSystemTools (Failed)
26 - FindPackageTest (Failed)
82 - Module.CheckIPOSupported-C (Failed)
83 - Module.CheckIPOSupported-CXX (Failed)
130 - CPackComponents (Failed)
136 - CPackComponentsPrefix (Failed)
137 - X11 (Failed)
145 - ExternalProject (Failed)
203 - CTest.UpdateSVN (Failed)
270 - Fortran (Failed)
272 - FortranC.Flags (Failed)
353 - RunCMake.Make (Failed)
370 - RunCMake.FileAPI (Failed)
374 - RunCMake.GenerateExportHeader (Failed)
381 - RunCMake.GenEx-TARGET_FILE (Failed)
415 - RunCMake.MaxRecursionDepth (Failed)
442 - RunCMake.file (Failed)
443 - RunCMake.file-CHMOD (Failed)
444 - RunCMake.find_file (Failed)
445 - RunCMake.find_library (Failed)
446 - RunCMake.find_package (Failed)
447 - RunCMake.find_path (Failed)
463 - RunCMake.project_injected (Failed)
480 - RunCMake.configure_file (Failed)
485 - RunCMake.SymlinkTrees (Failed)
508 - RunCMake.CommandLine (Failed)
510 - RunCMake.install (Failed)
515 - RunCMake.ExternalProject (Failed)
519 - RunCMake.CPackSymlinks (Failed)
523 - RunCMake.ClangTidy (Failed)
524 - RunCMake.IncludeWhatYouUse (Failed)
526 - RunCMake.Cppcheck (Failed)
527 - RunCMake.MultiLint (Failed)
533 - RunCMake.CPack_TGZ (Failed)
553 - CMake.GetFilenameComponentRealpath (Failed)
557 - CMake.ImplicitLinkInfo (Failed)
558 - CMake.ModuleNotices (Failed)
Errors while running CTest
make: *** [Makefile:148: test] Error 8

I now found that chmod doesn't work when umask is set. That's very strange...

It's the same on Cygwin when noacl is set.

Example:

#include <unistd.h>
#include <sys/stat.h>
#include <stdlib.h>
#include <fcntl.h>
#include <iostream>

const char fileName[] = "foo";
const int mask = S_IWRITE;

void get()
{
  struct stat st;
  stat(fileName, &st);
  std::cout << (st.st_mode) << std::endl;
}

void set(int mode) {
  chmod(fileName, mode);
}

int main() {
  std::cout << std::oct;
  close(open(fileName, O_WRONLY | O_CREAT));
  get();
  umask(mask);
  set(0);
  get();
  set(mask);
  get();
}

Output:

100644
100444
100444

Without the umask call (or without noacl), the third line is 100644.

Reported to cygwin: https://cygwin.com/pipermail/cygwin/2021-April/248208.html

I found the root caused. Reported here: https://cygwin.com/pipermail/cygwin/2021-April/248241.html

The problem is that when umask is set, stat applies it on its result. So it's not chmod that is broken, but stat. Worked around it in the patch (moved the umask call to where it's needed).

Do you prefer to rename the patch file back to 0001-Support-MSYS.patch, so you can review the differences?

Looks okish now. Some tests fail, but they fail also on cygwin. See my last comment on https://gitlab.kitware.com/cmake/cmake/-/merge_requests/5969.

I've built brotli and clang with this and all seems fine.

Thanks!

That was a long journey \o/

Thanks for the careful review.

It hasn't been that careful ;) as my cmake knowledge is limited, but you seem to know your way around and I agree with the patch goal.