subversion FTBFS on alpha

[Peter Samuelson]

Hi Luk,

Steve Langasek tells me you have been delegated to help figure out the
subversion FTBFS on alpha and arm.  (It also FTBFSes on s390 and hppa,
but s390 is clearly a kaffe bug, and anything weird on hppa we like to
blame on the kernel.)  I'll be eternally grateful if you can, in fact,
get this nailed down.  My own efforts have been hampered by lack of
access to an alpha running sid.  (I have tried to use fakechroot +
fakeroot + debootstrap on a sarge alpha, but so far I haven't been able
to get it to work.)

So here's a few notes about what is known so far ---

- I added some targets to debian/rules in 1.3.2-1, so that once you
  have tried a build, retry with 'debian/rules check-swig-rb'.  This
  ought to save a lot of time.

- The ruby bindings testsuite failure looks identical to one we had on
  powerpc a few months ago, which we solved by requesting a binNMU of
  ruby1.8 on that platform.  It turns out that it mattered which
  version of gcc was used to compile ruby.

- Troy Heber verified today that subversion 1.3.2-1 builds
  correctly with ruby1.8 1.8.4-1 but not with 1.8.4-2.  I suspect,
  again, a compiler-dependent issue - but this time in the opposite
  direction (gcc 4.0.2 works, 4.0.3 doesn't).  Troy is away from his
  alpha for the weekend.  AFAIK, nobody has yet tried building ruby
  with gcc 4.1.

- There was a rumor some time ago that some ruby problems could be
  fixed by building it with gcc -fno-strict-aliasing.  There are a lot
  of gcc warnings to that effect, and that's the sort of warning one
  ought not to ignore in any case.  However, when I tracked down the
  powerpc instance of this problem, -fno-strict-aliasing did _not_
  help.

Good luck,
Peter
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.alioth.debian.org/pipermail/pkg-subversion-maintainers/attachments/20060602/8470c1f8/attachment.pgp