+mirage-platform.3.3.0

[avsm]

supports forthcoming dune support as part of mirage/mirage#969:

• Add support for OCaml 4.08 (Add support for OCaml 4.08 mirage/mirage-platform#206 by @anmonteiro)
• libxenasmrun.a is symlinked in the ocaml directory in order to be able to use ocamlopt's -runtime-variant option.
• Expose flags through files as well as pkg-config (Add support for OCaml 4.06 and 4.07 mirage/mirage-platform#205 by @TheLortex). Instead of pkg-config, one can use the following files to get compilation flags:
  • mirage-xen-ocaml/libs
  • mirage-xen-ocaml/cflags
  • mirage-xen-posix/minios-cflags
  • mirage-xen-posix/minios-libs
  • mirage-xen-posix/posix-cflags
  • mirage-xen-posix/posix-libs
    With dune this allows us to write %{lib:mirage-xen-posix:posix-libs} to get the flags instead of having a script invoking pkg-config mirage-xen-posix --libs.

[camelus]

☀️ All lint checks passed 260293e

• These packages passed lint tests: mirage-xen-ocaml.3.3.0, mirage-xen-posix.3.3.0

---

☀️ Installability check (10501 → 10503)

• new installable packages (2): mirage-xen-ocaml.3.3.0 mirage-xen-posix.3.3.0

[mseri]

The failure with the system switch is expected?

[avsm]

Hm, no - will respin release

[avsm]

Hm, after some discussion with @TheLortex, I think we need to restrict this package to non-system installations. That won't disrupt existing users of mirage-platform, but will unblock the dune release of Mirage. @samoht @hannesm are you ok with this? The reason is that we need to symlink into the ocaml libdir in order to support -output-complete-obj for the mirage runtime variants, and you cant do that on a system switch.

[hannesm]

@avsm is there at some point in the future an idea how to support system switches? or will the -output-complete-obj approach only ever work with non-system switches?

is a system switch detectable from within an opam file - and is it possible to print a message "you're using a system switch that is currently not supported"? (or inform the user "in order to use a system switch, you'll have to manually ln -s YYY /usr/local/lib/ocaml/runtime/ZZZ?)

I'm slightly concerned to introduce behaviour that only works on non-system opam switches, and would be in favour to figure out how system switches can be supported.

[TheLortex]

It could be supported by patching OCaml so that the -runtime-variant option is more flexible: if we're able to provide a path to libasmrun.a then everything could work fine even on system switches.

[hannesm]

IIUC, runtime-variant is located in the default search path (Asmlink.runtime_lib uses Load_path.find, which uses the path initialized by Compmisc.init_path, using Clflags.std_include_dir), which is Config.standard_library, and can be set by the environment variable OCAMLLIB. Maybe not doing a symlink, but instead setting this environment variable when compiling/linking is sufficient (sorry, haven't tested this, only tried to read the code in the OCaml compiler).

[TheLortex]

It looks like OCAMLLIB allows only one single path, so that means we have to import the whole /usr/lib/ocaml/ content inside a directory that opam can write into (like .../ocaml-system/lib/ocaml).

[hannesm]

@TheLortex true. a plan!?:

• (a) figuring out in the opam file whether we're a system switch or not (and print a message which asks the user do the symlink manually) - once this is in place, we can merge this PR imho
• (b) ask OCaml distribution whether we can have a more mature -runtime-variant option (likely in 4.09 then!?)

[hannesm]

I suggest a conflict of mirage-platform=3.3.0 with ocaml-system would be feasible to move forward. WDYT?

[TheLortex]

Wouldn't an ocaml-system conflict cause trouble with CI ?
I'll make a patch to print a warning message if it's not possible to do the symlink.

[hannesm]

why should it cause trouble with CI? it will lead to not being installable if a system switch is used, which is the currently proposed support level (warning messages have the downside that people will ignore them). there is a variety of CI systems around, some use system switches, but most use opam switches, so they should be fine :)

[TheLortex]

Okay that's fine with me !

[avsm]

ive pushed the ocaml-system conflict

[kit-ty-kate]

Thanks

[hannesm]

@avsm fine, the conflict should be synchronised at upstream

[hannesm]

we discussed this a bit further in marrakesh, and it turns out we can use -I for including paths, and don't need any symlink. imho we should revise this 3.3.0 release to deal with that (and then remove the system compiler conflict, and remove the symlink). see also ocaml/ocaml#7926 ocaml/ocaml#2309

edit: see #13679 for same thing on freestanding.