Since 2006-10-28, someone has broken `Face-frob-property'

[Stephen J. Turnbull]

Nix writes:

 > ... hm. I just remembered Ulrich Drepper and Wolfram Gloger mentioning
 > (I think on libc-alpha a while back but it may have been in some paper
 > somewhere) that the glibc malloc hooks are pretty much completely broken
 > and don't work properly,

If they're partially broken they might still work properly? ;-)

 > especially in glibc 2.4+: as far as I can tell
 > we don't rely on those hooks except when terminating (to a very limited
 > degree) and as a debugging hack with no effect on memory-allocation
 > semantics, right?

I believe that is true.  There are some very old comments about magic
free hooks, but IIRC the code that used them is also inside the
comment....

There was a long-standing bug in "large string allocation", where the
cons counter didn't get bumped properly for large strings.  It was
possible to do something like `(while t (make-string 1000000 ?A))' and
crash XEmacs in short order.  Martin Buchholz submitted a patch for
that, which I applied based on my limited understanding that it was
correct and the strength of Martin's reputation for finickiness.

 > (if so, damn, I'll have to actually do some *work* to diagnose this
 > rather than leaping directly to the solution by magic. :) )

Looking at the large strings first is insider information, not magic,
but it might save you some work.

However, the bug was not that they never get collected---if something
else triggered a GC they got collected.  I wouldn't bet on it being
related, but you have to start somewhere....