Lift the restriction on installing 5.3.0+BER on macos+arm64

[yallop]

@kayceesrk reports that MetaOCaml 5.3.0+BER can be installed successfully on macos+arm64, so the current constraint is apparently unnecessary.

[dra27]

LGTM! This style of constraint was added in #17716 for all versions of OCaml which didn't support Apple silicon (i.e. everything before 4.12.0 apart from 4.10.1).

FWIW the same can be done for 4.14.1+BER:

diff --git a/packages/ocaml-variants/ocaml-variants.4.14.1+BER/opam b/packages/ocaml-variants/ocaml-variants.4.14.1+BER/opam
--- a/packages/ocaml-variants/ocaml-variants.4.14.1+BER/opam
+++ b/packages/ocaml-variants/ocaml-variants.4.14.1+BER/opam
@@ -77,4 +77,4 @@ depopts: [
   "ocaml-option-nnp"
   "ocaml-option-nnpchecker"
 ]
-available: !(os = "macos" & arch = "arm64") & os != "win32"
+available: os != "win32"

[dra27]

Orthogonal to this PR, but this:

Suggested change

	available: os != "win32" & arch != "arm32" & arch != "x86_32"
	conflicts: [
	"ocaml-option-32bit"
	"ocaml-option-bytecode-only"
	]
	available: os != "win32"

and removing them from the list above is more correct.

[yallop]

The arm32- and x86_32- variants are failing in CI. I wonder if it's due to this change.

[yallop]

FWIW the same can be done for 4.14.1+BER:

Thanks. I've made that change in 20f0163

[dra27]

I've pushed two additional commits which will hopefully improve the CI. As it's written at the moment, the 4.14.1 and 5.3.0 packages don't support bytecode-only builds (I'm not sure if MetaOCaml actually supports that itself?), so I added the conflict to the 4.14.1 version as well (that happened to be being caught by CI).

The second commit brings a few of the changes which are in ocaml-compiler.5.3.0 into the BER version - hopefully the 32-bit buillds will be correctly rejected this time.

[dra27]

The lint error is incorrect (as for compiler packages in general). The failures on Arch I think can be ignored, although there's trouble on the horizon! I think that's GCC 15. For the 4.14.1 version, it's missing ocaml/ocaml#12577 (from 4.14.2). For the 5.3.0 version, I'm wondering if the BER patches have accidentally backed out ocaml/ocaml#12509 and ocaml/ocaml#11764 (which are the "trunk" parts of the 4.14.2 PR), but I haven't looked in detail.

[shonfeder]

The outstanding CI errors:

archlinux-ocaml-4.14
ocaml-variants.4.14.1+BER (failed: too many arguments to function ‘(value ()(void))(caml_prim_table.contents + (sizetype)((long unsigned int)*pc * 8))’; expected 0, have 1)
archlinux-ocaml-5.3
ocaml-variants.4.14.1+BER (failed: too many arguments to function ‘(value ()(void))(caml_prim_table.contents + (sizetype)((long unsigned int)*pc * 8))’; expected 0, have 1)

IIUC, @dra27 you think we should hold off until these are fixed? (The other option being we mark the package unavailable or in conflict?).

[yallop]

For the 5.3.0 version, I'm wondering if the BER patches have accidentally backed out ocaml/ocaml#12509 and ocaml/ocaml#11764 (which are the "trunk" parts of the 4.14.2 PR), but I haven't looked in detail.

At a glance, it doesn't look that way: the commit metaocaml/ber-metaocaml@d6b868c (which seems to be the reworking of those PRs in 5.3.0) is on the ber-n153 branch, which is the source for the 5.3.0+BER variant.