Make the configure script fail on command failures

[kit-ty-kate]

The added set -e is here to harden the script and make sure such failures are caught early. Otherwise the script will simply ignore failing commands and carry on.

[gerdstolpmann]

find not POSIX? What a crap, it's here: https://pubs.opengroup.org/onlinepubs/9799919799/utilities/find.html (POSIX-1.2024)

I wonder why we have to consider ancient implementations like MSYS2 that never got updated. If you want a working build system on Windows, run a Linux shell. It's all there.

[kit-ty-kate]

Arf sorry my bad, i worked from (flawed) memory and didn't bother to double-check what i was saying. Sorry again.

I removed that change, however the first commit about hardening the script against failures would be greatly appreciated. I think what happened with the failure on msys2 (from someone i was helping get things working) was that PATH was badly configured and find ended up being find.exe from Windows instead of find from msys2 and thus failed. However the lack of set -e made it fail silently and ended up not installing the META file for findlib which gave the impression of a working ocamlfind install but broke packages requiring that findlib library.

[dra27]

Out of curiosity to check what else may be going on, where did the problem originally arise? find is part of findutils on both MSYS2 and Cygwin and, more importantly, it's a dependency of Base - i.e. it should always be installed.

It sounds like the original issue was a user's PATH not being correctly set, but that may be worth checking further?

[kit-ty-kate]

It sounds like the original issue was a user's PATH not being correctly set, but that may be worth checking further?

yes, i believe the user had their MSYS2 path after C:\Windows\system32 in their PATH
Is opam not reordering PATH when choosing a POSIX layer other than the builtin cygwin? Thinking about it, it might be a worthwhile extra check to have during opam init if we don't.

However i think this PR is desirable regardless of what opam does

[kit-ty-kate]

I had to debug yet another issue where ocamlfind simply ignored a segfault and continued the installation as if nothing happened resulting in a broken installation.

This time the problem of the user was that they were using dash master as their main POSIX shell but it happens to have a serious bug which makes sh -c 'echo test >> "${1%.in}"' sh src/findlib/META.in segfault (reported upstream here).
However find -exec cmd {} ; simply ignores failures of cmd and just carries on (per its POSIX spec). I've added a commit to fail early which would've reduced the time to debug this issue by several days and would've been less frustrating for the user and maintainers of the downstream projects where the error appeared (since the ocamlfind installed in a broken state). I've detailed the choice of replacement in the commit message.

[kit-ty-kate]

Any chance this could be merged? Users keep hitting this issue

[tjammer]

To add another data point, I just ran into this issue (problem when installing topkg) in a rather minimal linux container. Apparently, Rocky Linux 8 doesn't come with findutils pre-installed. Like in the other cases, opam setup works just fine until installing topkg. And that's with a clean switch which hasn't been deleted manually. An error in a prior install step would have been greatly appreciated.

So, it's not just msys2.

[dra27]

I've cherry-picked the first commit (0211a88) into #112. It has an alternate fix in 376ce3d for the second commit in this PR, which I think is better because it completely eliminates the uses of find and sh which deals with the problems the improved error handling was meant to identify.