Le 20/02/2017 à 18:25, Jean-Marc Lasgouttes a écrit :
Le 10/02/2017 à 17:58, Guillaume Munch a écrit :
There's more data, but I am not sure how to interpret the results that
massif-visualizer shows.

If you send the file or make it available, I can take a look.

Here it is. But if it is like callgrind, it might lack the symbols.

Interestingly, the massif output that you sent also points to
MathMacro::clone(). Besides the obvious getTextLayout leak that needs to
be plugged, would you have an idea on why it is specifically macros that
appear here (with an increasing memory use, like the other)? I do not
see other math objects doing that.

Note that there are other sources that point to MathAtom::MathAtom, as
the drop in percentage shows.


Interesting, massif-visualizer shows the results differently. With it
one sees that MathAtom quickly reaches its final space usage and even
decreases at some point. It does not look similar to the slow creep of
the memory leak. Again, I am not sure that I know how to interpret the
results.

Guillaume


Reply via email to