On Sat, Dec 27, 2008 at 12:37:39AM +0100, David Allouche wrote: > > Yes, more or less. You indeed have 1Mb increase in memory usage > > for every four or five iterations. Now the magic scheme command is '(gc)' > > and you will have to redo your thirty iterations before noticing > > further increase. So maybe it is just the gc which is not well synchronized. > > You may try to see whether doing a '(gc)' also solves the problem with > > the increase in memory usage during idle time. > > Dunno if that changed in the past years, but when I looked at the > Guile glue, the "memory size" information was wrong for complex > objects such as trees and lists. > > Typically, the size reported to Guile for a tree structure was the > size of a single node. Admittedly, it is difficult to report a > significant size, because different trees that are visible as > different objects to Guile can share common subtrees. The only way to > report the correct size for such shared structures would be to build > the entire structure as Guile objects. > > This memory size information is critical to guile's garbage > collection. It lets guile know when to trigger a gc. Not reporting > correct sizes for guile-visible objects leads to systemic memory leaks > caused by insufficiently frequent garbage collection. > > That might be related to what you are observing now.
Thanks for recalling this fact which slept out of my mind. However: why do we only observe the strange behaviour for the Qt-version, and not the X11-version...? --Joris _______________________________________________ Texmacs-dev mailing list [email protected] http://lists.gnu.org/mailman/listinfo/texmacs-dev
