Problem with make on windows?

[mdavidsaver]

GHA builds of epics-base revisions which previously built successfully, have started failing with errors which might be related to make --output-sync. I have seen a couple of variations:

  make: *** invalid output sync mutex: --.  Stop.

   make: *** invalid output sync mutex: �����.  Stop.

  make: *** invalid output sync mutex: �*V8��.  Stop.

This looks like corruption. Could this be a gnumake bug?

 $ make --version
GNU Make 4.4.1
Built for x86_64-w64-mingw32
...

This is somewhat different than a successful build from a couple of weeks ago.

$ make --version
GNU Make 4.4.1
Built for Windows32
...

[mdavidsaver]

make from chocolatey was last updated in May 2023.

The GHA windows-2019 image was updated last week.

[mdavidsaver]

Switching to the windows-2022 image has the same failure. But shows the same "Built for" as older successful builds.

$ make --version
GNU Make 4.4.1
Built for x86_64-w64-mingw32
...

Appveyor builds continue to (slowly) succeed, and show:

$ make --version
GNU Make 4.4.1
Built for Windows32
...

I guess that difference does not mean anything.

[mdavidsaver]

... Could this be a gnumake bug?

Reported as https://savannah.gnu.org/bugs/index.php?64806

[mdavidsaver]

I have gone ahead of committed 899b183 as a workaround.

[mdavidsaver]

After applying 899b183, I had to make another epics-base/pvDataCPP@5387fac to fix what was either a make or perl issue where print.obj somehow became PRINT.obj sometimes.

I am now wondering if our windows builds are actually using make+perl from the GHA image instead of chocolatey as intended. This might explain why an update of the GHA image could have such effects.

[mdavidsaver]

@pheest points out that strawberryperl 5.38.0.1 includes "make.exe" (previous releases only installed "gmake.exe"). This version appeared on October 2, 2023. I guess this, not the GHA image update, was the initiating event.

Found exec make at 'C:\\Strawberry\\c\\bin\\make.EXE' 
$ make --version
GNU Make 4.4.1
Built for x86_64-w64-mingw32

The current PATH manipulations put this "make.exe" first.

ci-scripts/cue.py

Lines 690 to 696 in 899b183

	if os.path.exists(r'C:\Strawberry\perl\bin'):
	# Put strawberry perl in front of the PATH (so that Git Perl is further behind)
	logger.debug('Adding Strawberry Perl in front of the PATH')
	os.environ['PATH'] = os.pathsep.join([r'C:\Strawberry\c\bin',
	r'C:\Strawberry\perl\site\bin',
	r'C:\Strawberry\perl\bin',
	os.environ['PATH']])

[mdavidsaver]

fyi. StrawberryPerl/Perl-Dist-Strawberry#148

[mdavidsaver]

fyi. the shifting sands of GHA have shifted back. The current windows image reverts strawberryperl back to 5.32.1.