Josh Pieper <jpieper@andrew.cmu.edu> and I independently figured out
the major cause of the memory leak: the open_tmp_file callback.
Someone will have to write "I will not design interfaces without
pools" on the blackboard a hundred times. I committed his patch in
r7381 after verifying that it passes basic_tests and commit_tests over
ra_dav.
With that fixed, an import of the Mozilla source over ra_dav grows
very very slowly, probably owing to the valid_targets hash. We could
reduce the size of that hash by not adding pathnames to it twice, or
by checking whether a path already exists in it before making a copy
of the path in the hash's pool. Or we could do nothing, since we're
talking about a difference of a couple of megabytes for a gigantic
import.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 11 08:10:03 2003