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

[Nix]

On 29 Dec 2006, Aidan Kehoe spake thusly:
>  Ar an naoú lá is fiche de mí na Nollaig, scríobh Nix: 
>
>  > > Part of implementing the portable dumper involved removing support for
>  > > purespace, and this leak never stops garbage collection--it seems to be
>  > > independent of the XEmacs memory accounting.
>  > 
>  > Perhaps it's a different leak. The bug I just found (and posted a patch
>  > for) involves GC becoming less and less frequent (on the average) as
>  > soon as memory allocation exceeds the threshold where gc-cons-percentage
>  > would start to matter, or ~20Mb, whichever is larger.
>
> Hmm. How does the discrepancy between what M-: (garbage-collect) RET returns
> and the virtual memory allocated according to the OS, look on your build?
-------------- next part --------------
> I?m building with your patch and will experiment further tomorrow, but the
> two did diverge notably for me. 

They've started doing so for me, under --enable-debug at least. (I've
not checked --disable-debug since applying that patch.)


However, I would expect such divergence to be a consequence of the
increasingly rare GCs triggered by the bug fixed by that patch as well:
after all, GC does still run, sometimes, and if you check
object-memory-usage after GC has just run, it will report a fairly low
figure: if you run it when GC hasn't run for ages, it'll report a large
one...

The figure that really shows this bug is the return value of
(consing-since-gc): it climbs to ridiculous values after this bug kicks
in.

>  > I've not been running with this patch for long, but preliminary checks
>  > indicate that leak rates are down to those I'd associate with 21.4.x.
>
> Good!

I woz wrong. One leak dead, more to find: the search continues. (Bah.)