`make sharedinstall` does not return failure if install commands fail

[mgorny]

Bug report

If make sharedinstall fails to install some Python extensions, the make target wrongly succeeds. For example, I'm seeing:

/usr/bin/install -c -m 755 Modules/array.cpython-312-x86_64-linux-gnu.so /usr/lib/python3.12/lib-dynload/array.cpython-312-x86_64-linux-gnu.so
/usr/bin/install: cannot create regular file '/var/tmp/portage/dev-lang/python-3.12.0_alpha3/image/usr/lib/python3.12/lib-dynload/array.cpython-312-x86_64-linux-gnu.so': No such file or directory
/usr/bin/install -c -m 755 Modules/_asyncio.cpython-312-x86_64-linux-gnu.so /usr/lib/python3.12/lib-dynload/_asyncio.cpython-312-x86_64-linux-gnu.so
/usr/bin/install: cannot create regular file '/var/tmp/portage/dev-lang/python-3.12.0_alpha3/image/usr/lib/python3.12/lib-dynload/_asyncio.cpython-312-x86_64-linux-gnu.so': No such file or directory
/usr/bin/install -c -m 755 Modules/_bisect.cpython-312-x86_64-linux-gnu.so /usr/lib/python3.12/lib-dynload/_bisect.cpython-312-x86_64-linux-gnu.so
[...]

Nevertheless, make install returns successfully in this case. This causes major problems for automated builds since they end up with broken Python installs when the make target should have failed.

I need to investigate why it's failing but that's a separate issue.

Complete build log (1.2M): dev-lang:python-3.12.0_alpha3:20221207-142002.log

The problem seems to be that the sharedinstall targets runs a single shell command and make doesn't check the exit status until its very end.

I suspect the same problem may apply to other install rules.

Your environment

• CPython versions tested on: 3.12.0a3
• Operating system and architecture: Gentoo Linux/amd64

Linked PRs

• gh-100220: Fix error handling in make rules #100328

[zware]

GH-100328 merged, thanks for the report and patch!

I think we'll pass on backporting this; there haven't been reported problems in this vein before (though the recipes haven't changed much, so the underlying issue does exist in older branches), and there could be some lesser-used recipes that shouldn't use -e that we missed.

[mgorny]

Yeah, I suppose the risk is relatively small, as it affects only a few less frequently used rules.