Are ghc, gnat, mono and ocaml actually available for ppc64le? That's what these macros indicate, and we shouldn't add new arches for pieces that dont (yet) exist.

Created attachment 886397 [details]
redhat-rpm-config.add_ppc64le_for_ocaml.patch

as suggested by Panu reduce the patch to the ocaml macro.

(In reply to Michel Normand from comment #2)
> Created attachment 886397 [details]
> redhat-rpm-config.add_ppc64le_for_ocaml.patch
> 
> as suggested by Panu reduce the patch to the ocaml macro.

In general terms (not related to ppc64le) I wish that
RPM would *not* have an ocaml_arches macro at all.

The macro usually has the wrong contents, but more seriously
it is incorrect: There are two separate sets of OCaml native
arches, those supporting compilation and those supporting
native dynamic linking.  We could define the correct macro(s)
in the OCaml package.

These macros are not my invention, its nothing but ridiculous (not to mention cumbersome to all involved parties) that I'm the "maintainer" of these macros when I haven't got a clue. Gnat? Is that a sound caused by swallowing with a dry throat? Although I suppose it means something different here...

The reason these macros are in redhat-rpm-config is a compromise because they need to be present on the buildsystem at .src.rpm creation time (ie before buildrequires have been evaluated) and you dont want to pull in every possible language stack into the buildsys "just because".

Now, perl already has its own "perl-srpm-macros" package which redhat-rpm-config requires for the above reason, this arrangement transfers the language stack control and knowledge out of my lap to where it belongs. So if you Richard want a similar arrangement for OCaml, I'm game. Just be careful not to add excess dependencies to the ocaml-srpm-macros package because that'll get pulled in on a LOT of systems.

Hmm .. OK.   I didn't realize there was this requirement to have
the macros available before ocaml.rpm was installed.  I will see if
it's easy to build a subpackage with no dependencies and open a
new bug about it if I get anywhere.

This should be fixed in ocaml-4.01.0-14.fc21.  However I'll
leave this open because of the dependent bug 1087794, and
because it would be good if someone could check that it
actually works.

the changes made with rhbz #1087794
removed ocaml_arches macro and
added ocaml_native_compiler and ocaml_natdynlink macros

old ocaml_arches had two archis (alpha, ia64) that are not present in new macros.
meaning these two (and may be others) do not have ocaml native compiler support.

At least need to update two packages below that used ocaml_arches:
ocaml-easy-format
hevea

I did a test on ppc64le for this two packages removing in spec the line
ExclusiveArch:  %{ocaml_arches}
but I want confirmation that this is the right thing to be done.

(In reply to Michel Normand from comment #7)
> the changes made with rhbz #1087794
> removed ocaml_arches macro and
> added ocaml_native_compiler and ocaml_natdynlink macros
> 
> old ocaml_arches had two archis (alpha, ia64) that are not present in new
> macros.
> meaning these two (and may be others) do not have ocaml native compiler
> support.

At least IA64 does have a working code generator (although I think
upstream dropped it).  In any case both architectures are obsolete.

> At least need to update two packages below that used ocaml_arches:
> ocaml-easy-format
> hevea

Yup, any use of %{ocaml_arches} is wrong.  I think I got most of
the packages, but I didn't see these two.

> I did a test on ppc64le for this two packages removing in spec the line
> ExclusiveArch:  %{ocaml_arches}
> but I want confirmation that this is the right thing to be done.

Yup, it's correct to remove that line from any spec file that
you see.

Since I don't think you're a Proven Packager, I can do these two
packages if you want.

(In reply to Richard W.M. Jones from comment #8)
> (In reply to Michel Normand from comment #7)
> > the changes made with rhbz #1087794
> > removed ocaml_arches macro and
> > added ocaml_native_compiler and ocaml_natdynlink macros
> > 
> > old ocaml_arches had two archis (alpha, ia64) that are not present in new
> > macros.
> > meaning these two (and may be others) do not have ocaml native compiler
> > support.
> 
> At least IA64 does have a working code generator (although I think
> upstream dropped it).  In any case both architectures are obsolete.

I checked this out further.  The "retired camels" branch:

https://github.com/retired-camels/ocaml

has support for Alpha & IA64 (and others).  However we don't
ship the retired camels backends in Fedora.  So it is correct
that these two architectures do *not* appear in the new list.

FWIW I've no intention of shipping the obsolete backends from
retired camels.  Someone else will need to come along and do that
work if they care about it.

(In reply to Richard W.M. Jones from comment #8)
> 
> Since I don't think you're a Proven Packager, I can do these two
> packages if you want.

I would appreciate, thanks.

We already have releases 14-17 built for ppc64le. I'm closing this to clean the tracker bug. Feel free to reopen if you wanted to close this later. Thanks for your help.