solo5-bindings-* need conf-gcc

[samoht]

No description provided.

[samoht]

Ok so ./configure.sh was broken for old gcc ... not sure if we want to fix this

(cf Solo5/solo5@18021b0#diff-0ba17b1855b0f97042c31b2b2c2fd2b075fa355a840b152f1959daf2a755266e)

[samoht]

so this PR is probably useless ... it's better to add a flag to say that's an expected error

[samoht]

New tentative with a fix for all the solo5-bindings-* packages + using the avoid-version flag.

[dra27]

There is a PR for opam-repo-ci to improve the error message, but what it's trying to tell you is that opam-version: "2.1" is incorrect - in all files where that's been changed it should be opam-version: "2.0" (there is no version 2.1 of the opam format).

The avoid-version flag is ignored by opam 2.0.x (as that's an opam 2.1 feature). If that's a problem then you either need to do a trick like the ocaml-beta package, add available: opam-version >= "2.1", or just live with it for opam 2.0.x (I expect that it's actually a case of just living with it?!)

[samoht]

Is opam 2.1 able to understand ignored flags in 2.0 files? If yes, I guess I can live with it (even if the CI will continue to fail :-/)

[dra27]

No, opam 2.1 will interpret the flag

[dra27]

In retrospect, the opam-version field is extremely unfortunately named - it is really the opam-format-version, so you're using the 2.0 specification of the format (which is what 2.1 uses). The flags field is part of opam 2.0 and the semantics are already that opam ignores flags it doesn't understand. So opam 2.1 will read and obey avoid-version, opam 2.0 will do nothing with it (it will display a warning if you turn on verbose output when updating the repo)

[samoht]

In retrospect, the opam-version field is extremely unfortunately named

I wonder who choose this name :p

[samoht]

Revdeps testing is going wild here, probably best to merge quickly..

[avsm]

opam-version is about the same as (lang 2.0) in dune files. We just need to be consistent about explaining that these represent metadata format versions and not client versions.

[hannesm]

imho this PR is wrong. solo5* works fine with clang. some old version do not work with gcc 10+ (see #19655 (comment))

[samoht]

Good point, I've reverted the conf-gcc part here: #19766

[dra27]

In retrospect, the opam-version field is extremely unfortunately named

I wonder who choose this name :p

One who didn’t have the benefit of hindsight!! 🙇