Fix MPR#7852 by reverting #1737

[trefis]

Fixes MPR#7852 by reverting GPR #1737 .

It's currently impossible to test trunk on janestreet's codebase (and probably on a good chunk of opam) because of this bug, so let's do the quick fix and we can have a look later at how to arrange things.

Ping @Drup , @hhugo .

[Drup]

This makes me sad. The fault lies in the implementation in Env, not in your patch. Let's revert until we have a better solution.

[gasche]

Good to go if the CI passes.

In my experience it's tricky to remember to re-merge these things once they are reverted. Maybe one way to avoid this would be to open a new PR with the (about-to-be-reverted code), the WIP status, and a pointer to the issues that were created.

[trefis]

I had forgotten to git add the new test file 😆
Running the CI again.

[damiendoligez]

@trefis will you open a new PR as suggested by @gasche? Or should I do it?

[trefis]

Meh. I'm not sure it's worth it honestly.
There were regression tests for #1737, which I left in (the output is just not what we'd want them to be). Apart from that I'm not sure an open GPR with no content buried 5 pages away on the GPR list is going to be much of a reminder.

Also, AFAICT, it'd be a non trivial amount of work to get that thing to work properly with the current architecture. (There might be some horrible hacks we can implement to get it working without too much work, but I really don't want to go there)

[gasche]

Could someone produce a clear explanation, somewhere, of what the problem is? @trefis and @Drup appear to know what the problem is, but there are no explanations in MPR#7852 for others to understand.

[trefis]

I'll write something up.

[damiendoligez]

ping @trefis

[trefis]

I updated MPR#7852.