opam2 built-in os detection sets os to macos instead of darwin

[avsm]

opam1 set the os variable in constraints to darwin (this is used in various opam packages for osx-specific config)

opam2 sets it to macos, which in turn causes build failures due to the package repository testing for os = "darwin".

We should reconcile this to be consistent between opam1 and 2

[marco-m]

+1 for darwin, which is arguably the name to use (output of uname) and, if I understand correctly, would reduce the overall number of changes.

[dbuenzli]

This was discussed in #3058 I think macos is the right name especially since it is used with sw_vers -productVersion for os-version, which wouldn't make any sense for darwin.

You could argue to have darwin for os and uname -r for os-version but that's really not what you want from an end-user perspective: you want to constrain things by macOS versions.

[avsm]

I'm ok with using macos in opam2, as long as the repo rewriter logic also maps the variable name back to darwin for opam1 compatibility

[AltGr]

The rewriting should have accounted for that change, it seems like a bug.

[AltGr]

Ok, the rewriting was done properly for depexts, but not in other places (available, and filters). Fixing that.

[AltGr]

Now fixed in the 2.0.0 branch of the repo. Note that we don't handle the cases where the os variable would have been used as a string (e.g. make "%{os}%.target". But thankfully nobody is doing that !)