Le jeudi 30 juillet 2020 à 08:23 +0000, [email protected] a écrit :
> On Wed, Jul 29, 2020 at 03:36:12PM +0000, Debian FTP Masters wrote:
> >    * lib{blas,lapack}.so.3 are again dynamically linked against 
> > libopenblas.so.0.
> >      This restores the pre-multi-flavour behaviour.
> >      Incidentally, this brings back the possibility of checking at runtime 
> > for
> >      the presence of openblas_get_config symbol. (Closes: #960728)
> 
> I reviewed the changes, and actually it introduced a regression bug
> where the choice of update-alternatives gets rather complicated.
> 
> e.g. if I set libblas.so.3 <--provides-- libopenblas0-serial, and
> libopenblas.so.0 <--provides libopenblas0-pthread, then since
> openblas-serial/libblas.so.3 NEEDS libopenblas.so.0, and has no
> RPATH attribute, ld.so will resolve libblas.so.3's dependency to the
> global /usr/lib/<triplet>/libopenblas.so.0, which points to
> openblas-pthread/libopenblas.so.0 . This breaks the users assumption
> on the "serial" threading of libblas.so.3.
> 
> Severity: important
> Fixed in git repo:
> https://salsa.debian.org/science-team/openblas/-/commit/5bd3216d3d78c1315627f0e117f2d20c8962eef6

Thanks for spotting and fixing this.

Two additional questions (I did not yet try to compile your changes):
- I guess that the override_dh_shlibdeps that I added in debian/rules
is no longer needed?
- Does lintian complain about the rpath? Setting rpath is generally
forbidden, but allowed in the case of a private library, so we should
be fine:
https://lintian.debian.org/tags/binary-or-shlib-defines-rpath.html

-- 
⢀⣴⠾⠻⢶⣦⠀  Sébastien Villemot
⣾⠁⢠⠒⠀⣿⡁  Debian Developer
⢿⡄⠘⠷⠚⠋⠀  https://sebastien.villemot.name
⠈⠳⣄⠀⠀⠀⠀  https://www.debian.org

Attachment: signature.asc
Description: This is a digitally signed message part

-- 
debian-science-maintainers mailing list
[email protected]
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/debian-science-maintainers

Reply via email to