Do not swallow the configure exit code in configure_wrapper

[vbraun]

Followup for #39363: do not ignore the ./configure exit code

[github-actions]

Documentation preview for this PR (built with commit 1f1fa4f; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[dimpase]

@vbraun - you have lost somewhere in the release process 72876f0 from #40765

configure_wrapper should already be

#! /bin/sh
cp conftest.py bak_conftest.py
./real_configure $@
realconf_result=$?
mv bak_conftest.py conftest.py
exit $realconf_result

[dimpase]

Perhaps your fix here is better than 72876f0 - but you should not just silently drop commits while making a beta!!!

[user202729]

some notes.

• looks like there was no good reason for not using atexit trap back up/restore conftest.py #39363 (comment) , so it's fine to use one. I believe it even makes sense to restore the conftest.py on ctrl-c.
• bash sometimes has an implicit exec at the last command https://unix.stackexchange.com/questions/279024/reason-for-exec-in-wrapper-scripts#comment1435139_279030, but I believe it shouldn't do so if there's a atexit.
• https://unix.stackexchange.com/a/286383/ mention that if the subprocess returns exit code X > 0, the shell will return 128+X, although I don't think this is a large issue, certainly it's smaller than always returning 0 or whatever.
• according to https://stackoverflow.com/a/14812383/5267751 , in bash then just trap '...' EXIT suffice, but in other shells it might not. So the current implementation looks like the safest.

[user202729]

@vbraun - you have lost somewhere in the release process 72876f0 from #40765

it was dropped in #40877 .

As for comment on which fix is better, I'd say this one, because it's preferable to restore the backup on ctrl-C interruption.

[dimpase]

Well, @vbraun - I congratulate you on giving everyone extra work to do with your meaningless reverts

[dimpase]

the most satisfying solution is of course to slowly cook in boiling oil the guy who refused to even consider fixing pytest to avoid this nonsense...

(his name means "maker of pans", so that would be particularly fitting)

[dimpase]

I complain in particular as you reverted what I did in #40765 to fix @mezzarobba 's forkbomb problem reported there.

Marc, does this PR fix work for you too?

[dimpase]

Unfortunately this does not quite cut it. Indeed, interrupting ./configure with Ctlr-C does not get conftest.py, but I still get conftest.py deleted if ./bootstrap && ./configure is triggered by make.

The latter happens after one, say, updates an spkg. and then hits make.
Seen on macOS with zsh.

And also on Linux with bash.