On 17 Dec 2006, nix(a)esperi.org.uk verbalised:
> I just upgraded to latest CVS (2006-12-16, 21:20) to try to dig
> further into this KKCC-related (?) memory leak. (I'll know for
> sure if it's KKCC-related in an hour or so.)
It isn't KKCC. I've built an Emacs without newgc and --without-kkcc.
Yet half an hour later, after reading 500-odd posts on the
spamassassin mailing list in Gnus:
Half of this appears unused. I'm going to get some stats out
of malloc() next, to see if arena fragmentation is at fault.
The reported memory doesn't seem unreasonable; Gnus conses like it's
going out of style. (Yeah, I know you'd like to go back to 2-word
conses and save 15% of that, but that's not in the cards.)
Is this an Xft build? It's hard to get Xt/Xlib stats, as you know,
but Xft does provide some hooks for its memory usage.
If you have the chatter from Xft turned on, you'll notice that it
instantiates the same font many times. I *think* that Xft itself
caches those properly, but there's XEmacs superstructure too, which is
not recorded. This should be "small", but I don't know. Also, I
don't know where Xft stores its actual font data, which can get fairly
large for Asian fonts at least, I suppose.
XEmacs-Beta mailing list