scm_register_collectable_memory() ?

[Han-Wen]

Hi,

Question: why does GUILE do a GC when a lot of memory malloced memory
is used?  I understand that doing GC frees up some of that memory, but
every piece of collectable malloc memory is already connected with a
cell, and we do GC for those regularly (i.e. when we run out of
cells.)

For example, the doc snarfer, which is quite heavy with malloced stuff
(strings), does many times more often malloc triggered GCs, and most
the of the yields are low. Why don't we measure yields, and adjust the
triggers, similar to how cells are done?

Precise numbers:

58 gcs, of which 47 are induced by scm_register_collectable_memory().
Average yield (i.e (mallocedbefore - mallocedafter) / mallocedbefore )
of 14 %. This is for the snarf-check-and-output-texi script, run on
the GUILE sources.




-- 

Han-Wen Nienhuys | address@hidden | http://www.cs.uu.nl/~hanwen