Re: Help-wanted tasks for musl

[Gregor Richards <gr@purdue.edu>]

Ruby and pari fail for uninteresting reasons I haven't diagnosed yet. 
i.e., it's something about the whole stack, not musl in particular. SDL 
works now, and should be reported as such when the new pkgsrc results 
come out. The Fortran stuff is just because pkgsrc doesn't know how to 
build GCC for musl; I just added the ability for Snowflake to build 
gfortran, so in the future that might work, but it's not ready in the 
current pkgsrc run.

With valediction,
  - Gregor Richards

On 08/19/2012 05:46 PM, idunham@lavabit.com wrote:
>>> Analysis of Gregor's pkgsrc failure results
>>> Gregor Richards has run the whole NetBSD pkgsrc build (over 10k
> packages) against musl and posted reports to the mailing list and wiki.
> Some analysis on the most frequent causes of failure could be extremely
> helpful to improving compatibility. It would also be nice to identify
> major dependency failures that can be fixed (either in the upstream
> package if it's buggy, or in musl if it's due to missing features) so
> that the packages which depend on them can be tested too. I'm looking
> to get results in the form of "we should fix X and Y and Z and then
> lots more packages will work".
>> ISTR that Gregor has something that gives him scores for how important
> different packages are.
>
> This is mvd.txt in the results tarball. (see musl.codu.org)
> Higher scores (at the top) mean more valuable/more packages stopped here.
> And looking at what it says, I see these scores:
> Ruby193-base: 904
> pari: 674
> qt3-libs: 394
> qt4-libs: 358
> ....
> libf2c: 282
> ....
> g95: 152
> ....
> SDL: 135
>
>> I look at this once in a while already, but now that I've done a little
> repartitioning, should be able to do a little more...
>> Just for a brief overview of a few things:
>> -SDL: patches haven't been merged
>> -There are at least a dozen packages that should be building, per my own
> tests without pkgsrc, but aren't.
>> -e2fsprogs provides libcomerr, so I suspect blocks zephyr, which blocks
> libpurple/pidgin/...
>> -avahi is semi-important
>> -ruby has been one of the higher-priority ones to fix
>> -R needs g77 or gfortran a/k/a g95 (I have g77, but that's not in
> pkgsrc...).
>> Also needs lapack & blas which need the same.
>> Octave needs all of the above, plus more...
>> I suspect that applying musl patches to gcc-core, and untarring gfortran
> on that, would be adequate-that's what I did for g77.
>> -qt3 & qt4 libs won't build OOB.
>
>
>
>