Re: dicache valgrind log entries

On Fri, May 30, 2014 at 10:28:05AM -0400, Jorge Arellano Cid wrote:
> On Fri, May 30, 2014 at 05:19:10AM +0000, eocene wrote:
> > 
> > Jeremy's logs have been showing lots of dicache
> > entries in recent days.
> 
>   Thanks. I'm working in the dicache now and there plenty
> of things to iron out...

  Well, it turned out to be much less than it seemed. Just
a simple bug in the new code I was working on.

  On current tree, valgrind reports "definitely lost"
DICacheEntry blocks created by Dicache_entry_new(), but in
fact, they're not lost.

  e.g. running:

  $valgrind --tool=memcheck --leak-check=yes --log-file=valg3678.out \
    ./dillo http://lwn.net/

  and exiting dillo after the page is fully loaded, reports
lost blocks from Dicache_entry_new() (one per image).

  Now, if you run the same command, wait until the page is fully loaded,
then switch to bookmarks and exit, there's no lost block reported.

  The  point  is:  bookmarks has no images, and once it loads the
dicache  cleans  previous  images (from lwn.net); hence valgrind,
sees no leaks.

  There're ways to fine tune valgrind with specific parameters for
leak detection, but I don't know them in the necessary depth to 
chose the correct ones.


-- 
  Cheers
  Jorge.-