>>> Source compiled.
>>> Test phase [not enabled]: dev-ml/parmap-1.0_rc6
>>> Install parmap-1.0_rc6 into /var/tmp/portage/dev-ml/parmap-1.0_rc6/image/ category dev-ml
make -j4 OCAMLLIBDIR=lib64/ocaml MANDIR=/var/tmp/portage/dev-ml/parmap-1.0_rc6/image//usr/share/man/man3o DESTDIR=/var/tmp/portage/dev-ml/parmap-1.0_rc6/image//usr install
ocamlfind ocamlopt -shared -o parmap.cmxs bytearray_stubs.o setcore_stubs.o bytearray.cmx parmap_utils.cmx parmap.cmx
ocamlfind remove parmap
* ACCESS DENIED: unlink: /usr/lib64/ocaml/stublibs/dllparmap_stubs.so.owner
ocamlfind: /usr/lib64/ocaml/stublibs/dllparmap_stubs.so.owner: Permission denied
Makefile:34: recipe for target 'install' failed
make: *** [install] Error 2
please attach the full build log; nothing in the makefile shows that 'install' should call 'uninstall' first :/ (which is what is happening here)
Created attachment 406102 [details]
I have also tried rebuilding the few ocaml libs and programs I have with USE=-ocamlopt, but it still fails the same way.
hmmm, can you try with /bin/sh as bash ?
Good catch, I completely missed configure almost-failing-but-not-quite.
I tried fixing the obvious == into =, but it still fails elsewhere. Rerunning eautoreconf fails due to missing macros (notably AC_PROG_OCAML).
I guess the best fix is to call econf with CONFIG_SHELL=/bin/bash
(In reply to Rémi Cardona from comment #5)
> I tried fixing the obvious == into =, but it still fails elsewhere.
> Rerunning eautoreconf fails due to missing macros (notably AC_PROG_OCAML).
this is dev-ml/ocaml-autoconf
> I guess the best fix is to call econf with CONFIG_SHELL=/bin/bash
yep could be the easiest thing to do, but if you/we have an upstreamable fix it's much better :)
Ah! Thanks for the pointers, I'll take another swing at it then!
Created attachment 406624 [details, diff]
Here's the upstream patch. The "==" was fairly obvious, but it took me a bit longer to figure out that "&>" isn't kosher in POSIX. This seems to work ok with both bash and dash.
Created attachment 406626 [details, diff]
And here's the ebuild patch. I'm not 100% confident about the autotools eclass part, I'm not quite sure about the WANT_* variables. Maybe you'll have a better clue.
great, thanks; could you please send a pull request at https://github.com/rdicosmo/parmap ?
for WANT_* variables, they default to 'latest' which is the best choice; those variables are there when packages that require older versions and for which it is not easy to port them
Damn! I thought I could weasel out of going to upstream with the patch! :p
+ 14 Jul 2015; Alexis Ballier <email@example.com> parmap-1.0_rc6.ebuild,
+ Fix bashisms in configure, bug #553596 by Rémi Cardona.
thanks a lot!