[Bug 1020088] Review Request: blosc - A high performance compressor optimized for binary data

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1020088



--- Comment #7 from Zbigniew Jędrzejewski-Szmek <zbyszek at in.waw.pl> ---
(In reply to Thibault North from comment #5)
> > BTW, the proper way to refer to %{docdir}/%{name} is through %{_pkgdocdir}.
> 
> Yes, but it looks like %{_pkgdocdir} still points to
> %{_docdir}/%{name}-%{version} even after %global _docdir_fmt %{name}.
> Or I am doing something wrong here, I don't know ?
Are you building this under F20/F21 or F19? You get an unversioned
%_pkgdocdir only in F >= 20.

> > But I think that a different solution is actually better:
> > c) simply install a compiled version of 'bench'.
> > 
> > I think this is better because as a user, I don't want to have to find out
> > how to compile the .c file to run the benchmarks, I would prefer to be able
> > to invoke it directly. I have now run bench myself, and I think it would be
> > worthwhile to package, because the results are quite interesting, and
> > relevant to how one would use blosc.
> 
> This makes sense. Moreover, the -O2 flag will match the actual blosc library
> from the package.
> On the other hand, the optimization brought by -O3 as well as SSE are lost
> with this packaged blosc binary, right? (doesn't that somehow defeat the
> purpose of blosc ?)
How the binary itself is compiled probably doesn't matter so much, compared to
how the library is compiled.

> 
> > There's a problem that making the bench binary and the associated
> > plot-times.py script part of either of the two binary packages is
> > problematic. If it is moved into the main package, it would start requiring
> > python, and x86_64 versions would nod be co-installable. If is is installed
> > as part of the -devel package, again, -devel would require python, and also
> > not be coinstallable. I think that adding a -bench (or -test) package is the
> > best option, with
> > /usr/bin/blosc-bench and /usr/bin/blosc-plot-times.
> 
> Good idea. I tried to implement it in the new spec, which is probably not
> perfect. The blosc-plot-times is a link pointing to the actual python file
> in %doc. I guess this one should be in %{_datadir} instead ?
Maybe just install the script in %{_bindir}, just removing the extension?

> I still had to use %exclude to avoid duplicates. Using %doc bench, for some
> reason, also adds *.rst.
> 
> > So, whatever you decide wrt. the %files problem, please post a new .spec. As
> > a reviewer, I don't think I should impose my view here, and you should pick
> > whatever you think best from the maintainer point of view.
> 
> New spec online:
> http://tnorth.fedorapeople.org/rev/blosc.spec
> http://tnorth.fedorapeople.org/rev/blosc-1.2.3-4.fc19.src.rpm
> 
> (Is it ok to overwrite this SPEC file everytime? Old ones are still
> available in the SRPMS packages at
> http://tnorth.fedorapeople.org/rev/blosc-1.2.3-?.fc19.src.rpm)
Yeah, I think that's common practice.

-- 
You are receiving this mail because:
You are on the CC list for the bug.