Two small changes to make by default, OOTB, work better with msys2.

[xeiavicaflashrepository]

msys2 is a rather popular semi-fork of cygwin to bring linux software to microsoft windows.

I made two tiny additions to make it work OTTB better with windows without having to patch it ourselves.

First one is if $HOSTTYPE is detected as msys2, do something differently from cygwin because of how symlinks work in msys2 compared to cygwin. Second one is to add a shebang to the start of src/cmd/INIT/mamprobe.sh because windows cannot load the file without sed-ing the shebang into it. This shouldn't make a difference because it's already a shell script to begin with.

I would like to see better type and HOSTTYPE detection with msys2. Let me know your thoughts and how I can change the patch to be better fitting for ksh93 and msys2.

[xeiavicaflashrepository]

Also, I should add, there already exists someone with problems with msys2. This should help fix things.
#801

[jeremyd2019]

Are there other places that know HOSTTYPE of cygwin.* that need to also check for msys.*?

[jeremyd2019]

I'm concerned this may be the wrong thing, ie for Cygwin, which does the opposite as MSYS2 (it bind mounts /usr/bin to point to /bin). Could we check HOSTTYPE at this point in the script?

Suggested change

	# Fix for msys2 because /bin is the same thing as /usr/bin.
	if test /bin -ef /usr/bin; then
	DEFPATH="/usr/bin:$DEFPATH"
	fi
	# Fix for msys2 because /bin is the same thing as /usr/bin.
	case $HOSTTYPE in
	msys.*)
	if test /bin -ef /usr/bin; then
	DEFPATH="/usr/bin:$DEFPATH"
	fi
	esac

[xeiavicaflashrepository]

I want to add, that the files src/lib/libast/path/pathposix.c and src/lib/libast/path/pathnative.c use a function that has been deprecated and removed a long time ago. GNU ldd reports back a linking error. I fixed by commenting it out which I'll remove in my PR.

[jeremyd2019]

If you want to convert paths, you would use cygwin_conv_path

[xeiavicaflashrepository]

@jeremyd2019
Get this, there was nothing wrong with the compiler script nor ksh93. It was windows itself that was the problem (What else is new). There is a for loop that starts with for i in ksh and what this would do is set the default shell to ksh. Windows didn't like that and it is why it resulted in a fork bomb of sorts. I'll close my other issue since I know it's not the script to blame.

I simply moved sh to the front, and it works now. As for the two files I modified, it would definitely break something if I removed them. They were merely deprecated but not removed. That's why it worked IF we set it to cygwin, because the if macro that pulls in sys/cygwin.h is ONLY if __CYGWIN__ is detected. I added in an || statement to include __MSYS__.

[jeremyd2019]

On Fri, 14 Mar 2025, xeiavicaflashrepository wrote: I simply moved sh to the front, and it works now. As for the two files I modified, it would definitely break something if I removed them. They were merely deprecated but not removed. That's why it worked IF we set it to cygwin, because the if macro that pulls in sys/cygwin.h is ONLY if __CYGWIN__ is detected. I added in an || statement to include __MSYS__.

MSYS2 gcc still sets __CYGWIN__ in addition to __MSYS__ so there shouldn't have been any problem.

[jeremyd2019]

There is no cygwin_conv_to in Cygwin codebase anymore. I still think you want to use cygwin_conv_path, but apparently this code isn't used because it should fail to link. BTW, don't know what you meant by GNU ldd. Did you mean ld? Or did you mean Cygwin's ldd?

[jeremyd2019]

No. That code would fail on modern Cygwin too. Should be changed to use cygwin_conv_path instead.

[jeremyd2019]

I was thinking, if there's interest here, I can put together a PR adding GHA testing on Cygwin. I don't understand why this code wasn't breaking the build on Cygwin/MSYS2 already. I am considering if I need to debug the "botch" check, because it definitely comes out false on MSYS2 at least.

[jeremyd2019]

Suggested change

	cygwin_conv_path(CCP_POSIX_TO_WIN_W | CCP_ABSOLUTE, path, tmp, 0);
	cygwin_conv_path(CCP_POSIX_TO_WIN_A | CCP_ABSOLUTE, path, tmp, sizeof(tmp));

Similar for the other calls to cygwin_conv_path.

No idea where conv_path_list_buf_size (or conv_last_list_buf_size) come from. But if you need to convert a path list, you can use cygwin_conv_path_list

[jeremyd2019]

so it looks like these old deprecated functions were 32-bit only, and removed in cygwin/cygwin@3e917da

[jeremyd2019]

This code does not seem to be enabled on (current) Cygwin because the test in src/lib/libast/features/omitted tries to use functions prefixed with _ and Cygwin does not define them. If I define them all to the non-underscore variant, the test produces:

#define _win32_botch_copy       1
#define _win32_botch_rename     1
#define _win32_botch_execve     1
#define _win32_botch    1

I added include of <stdio.h> and:

#ifdef __CYGWIN__
#define _open open
#define _fstat fstat
#define _fstat fstat
#define _read read
#define _write write
#define _close close
#define _mkdir mkdir
#define _chdir chdir
#define _access access
#define _chmod chmod
#define _getpagesize getpagesize
#define _link link
#define _pathconf pathconf
#define _rename rename
#define _stat stat
#define _truncate truncate
#define _unlink unlink
#define _utime utime
#endif

But if this code is enabled, it is almost certainly going to be really broken.

[jeremyd2019]

I think maybe it's better that those "botch" patches are not enabled, Cygwin does those behaviors for a reason