Bug#721537: perl: FTBFS on hppa: lib/locale test fails

[Niko Tyni]

On Sun, Sep 01, 2013 at 05:17:50PM -0400, John David Anglin wrote:
> On 1-Sep-13, at 5:02 PM, Dominic Hargreaves wrote:
> >On Sun, Sep 01, 2013 at 02:37:23PM -0400, John David Anglin wrote:

> >>lib/locale ....................................................
> >>Can't load '../l
> >>ib/auto/re/re.so' for module re: ../lib/auto/re/re.so: failed to
> >>map segment fro
> >>m shared object: Cannot allocate memory at ../lib/XSLoader.pm
> >>line 68.
> >>at ../lib/re.pm line 85.

> It's a rp3440 with 8 GB memory and 4 processors.

Wow, that's quite beefy for an hppa machine.

> If this is a mmap issue, there are some restrictions on the placement of
> memory shared between modules due to aliasing issues.  This occurs when
> software tries to specify the mapping location.

> I can poke more at this if I knew where to look.

I think H. Merijn Brand still builds and tests upstream Perl on HP-UX /
PA-RISC, but I doubt anybody has tested it on Linux. Perhaps something
in hints/hpux.sh is relevant? (seems doubtful though.)

I guess you're best off reporting this upstream. 

See if you can reproduce it with the pure upstream tarball or git
checkout [1], with a configuration as simple as you can make it,
ideally something like
 ./Configure -des -Dusedevel

Possibly relevant options could be -Dusethreads and -Duse64bitint.

If you have the time, upstream would surely appreciate bisecting the
upstream commit that broke it.  See pod/perlgit.pod in the perl source
tree.

Alternatively, I guess I can try and do the above if you can give me
access to an hppa host where this is reproducible and which doesn't
take ages to build Perl.  Let me know in a private email if you'd like
to set that up.

[1] git://perl5.git.perl.org/perl.git

Cheers,
-- 
Niko Tyni   ntyni at debian.org