Package failures on macppc-7.99.4

["John D. Baker" <jdbaker%mylinuxisp.com@localhost>]

On Wed, 28 Jan 2015, Greg Troxel wrote:

> I am not enthused about committing a MAKE_JOBS_SAFE=no for those
> packages when they seem to build fine most everywhere else in parallel.
> Most MAKE_JOBS_SAFE failures in the past have been provokable somewhat
> repeatedly (as in if 10 people run the build 10 times each, there will
> be some failures).  I looked at iso-codes and it's straight automake,
> which is usually ok.  But that's not a proof, just trying to estimate
> the odds.
> 
> Unfortunately I'm not clear on how to debug this.  It would seem there
> should be some way to get make to log command start/finish with times
> and then you could go over the log and see if you can find the bug.
> 
> I don't see how compiler options would change iso-codes, but as jperkin
> says it would be good to know if this is reproducible with default
> options.
> 
> I don't mean to be giving you a hard time; I believe it's failing for
> you :-)   But it's tricky to figure out the right thing to do.

I'm beginning to think I'm just seeing the general instability of the
macppc port that I'd encountered in the past.  On the upside it takes
longer for it to manifest than before (purely subjective).  The two
packages mentioned in this thread represented novel failure modes I'd
not seen before and what seemed a solution.  Now I'm not so sure.  I'd
definitely need to build them several times over to say for sure.
Whatever the case, it's probably specific to macppc (other powerpc ports?).

The "libvpx" failure was completely reproducible, see pkg/49600.

I've finally starting seeing more familiar failure modes.  While compiling
"print/ghostscript-agpl", the compiler would die with "internal compiler
error: Segmentation fault".  Restarting the build, it would successfully
compile the file on which it previously failed, build a few more and then
fail in the same way again later.  (Once I got internal compiler error:
Illegal instruction, but it's mostly Segmentation fault.)

Eventually, it starts segfaulting on every subsequent source file.
Setting MAKE_JOBS=1 has no effect.  The only recovery I've found so far
is to reboot the machine and resume the build.  I can then build more
packages.  For gargantuan packages like firefox, it may take many reboot
cycles for it to complete (at least it did last time I managed to compile
firefox on macppc).

The other persistent problem on macppc has been "lang/perl5" and random
packages that require perl to build.  "panic: memory wrap" has been the
bane of trying to build perl itself and any package that uses perl to
build.  See pkg/49598.  Either my un-hacking it or dumb luck let me
build it with no interruptions--I'm not sure which.  So far, packages
which use perl to build have not failed due to "panic: memory wrap".

But this has drifted--perhaps more relevant to "port-macppc" than
"pkgsrc-users".  Cc:-ing there as well just in case.

-- 
|/"\ John D. Baker, KN5UKS               NetBSD     Darwin/MacOS X
|\ / jdbaker[snail]mylinuxisp[flyspeck]com    OpenBSD            FreeBSD
| X  No HTML/proprietary data in email.   BSD just sits there and works!
|/ \ GPGkeyID:  D703 4A7E 479F 63F8 D3F4  BD99 9572 8F23 E4AD 1645