[Debichem-devel] Please review my changes to elpa package

Michael Banck mbanck at debian.org
Wed Jun 17 22:22:33 UTC 2015


On Mon, Jun 15, 2015 at 04:03:57PM +0200, Graham Inggs wrote:
> I see there was a build failure on s390x because of my new symbols file.

armel and mips also failed due to timeouts in the testsuite.

The testsuite does take rather long without output, I have to admit.

> Is there any rush to fix that in a 2013.11.008-4 upload, or can we
> fix that in 2015.05.001-1?

I don't mind either way, as you prefer.
 
> >The main question is whether to enable the ELPA kernels on base builds -
> >i.e. will a SSE ELPA kernel built on my notebook run on my other
> >noteboook with a different amd64 CPU?
> 
> I have access to various Opterons and Xeons from about 8 years old
> to current.
> We can test builds on the Ubuntu PPA builders.

Probably enough to build on a new machine and test on an older one.

Note that you might have to explicitly enable the SSE kernel for that,
e.g. by rebuilding the latest subversion debian/ for cp2k (I backported
the changes to support newer ELPA there) with it, and set ELPA_KERNEL
(or what it was) in the input file.
 
Or maybe there is an easier way.

> >In order to get it to build I had to rename BLACS_LIBS to
> >SCALAPACK_LDFLAGS (besides the SONAME change from 0 to 3).
> I fixed that by modifying blacs-lib.patch slightly:
> 
> --- a/configure.ac
> +++ b/configure.ac
> @@ -365,7 +365,7 @@
>    done
> 
>    if test x"${blacs_in_scalapack}" = x"no"; then
> -    LIBS="${old_LIBS}"
> +    LIBS="${old_LIBS} $BLACS_LIBS"
> 
>      dnl Test for stand-alone blacs
> AC_SEARCH_LIBS([bi_f77_init],[mpiblacsF77init],[],[],[-lmpiblacs])

ok.
 
> I'll check whether your method still builds on Ubuntu, where they
> build everything with -Wl,--as-needed.
> Maybe we should just build with that in Debian too?
 
Actually, that would probably be a good idea.

> Upstream have fixed the issue with fetching the .asc now, so we can
> enable signed tarballs.
> 
> Can I start committing changes for 2015.05.001-1?

Sure!


Michael



More information about the Debichem-devel mailing list