[Sbcl-devel] Regressions since 1.0.13 [was Re: upcoming sbcl-1.0.14 release ]

[Richard M K. <kr...@pr...>]

William Harold Newman writes:
> Known bugs willing, I expect to release sbcl-1.0.14 around the 26th.
> Until that time, please be conservative in commits. (And at that time,
> I will try to make sure to tag it.)

Changes during the last month seems to have introduced new chrashes
during the tests on some platforms I test on, which others can likely
investigate more effectively than I can.  You can see the crashes in the
full logs of the tests in the buildbot at

http://sbcl.static.net/buildbot/

Here's what I know:

* On Linux/ppc and Darwin/ppc since at least 1.0.13.20, there have been
  problems in builds with :GENCGC, leading to crashes in GC.  The
  precise symptoms have varied from commit to commit, but the failures
  have usually been reproducible while running core.test.sh on both
  Linux/ppc and Darwin/ppc.  Some versions have not built at all on
  Linux/ppc.

  Earlier this month, I think I convinced myself that the crashing can
  be prevented by reverting the introduction of CALL-HOOKS in
  src/code/gc.lisp in 1.0.13.20; but I don't know where the real problem
  is.

  (Anecdotally, I have not seen these failures on builds using
  CheneyGC.)

* Somewhere in the last month (I don't yet know when), room.test.sh has
  begun to fail on NetBSD/x86.  Again, this can be seen in the buildbot
  logs.  If it helps, I can provide a login account on a NetBSD/x86
  host.

--
RmK