Does this fix the leak, or just provide more info? In other words, should I try looking into this?

Just more information. I was checking that the memory increase is uniform (which it is 😞 )

I don't think it is all in one place either. Removing all of the plotting calls, and just making, saving, and closing the figures/axes results in a 3k/loop leak.

I am leaning towards this not blocking the release.

I suppose the real question is how much worse 1.5 is than 1.4.3... Even then, maybe not blockable -- some magical fix could always go into 1.5.1

running in in 1.4.3 now, but it looks just as bad (from eye balling the prints)

py3.5, mpl 1.4.3: Average memory consumed per loop: 45.2139k bytes
py2.7, mpl 1.4.3: Average memory consumed per loop: 0.1642k bytes

Maybe that explains why Python 3 always feels sluggish 👿

Python 3 startup time is also much slower than python 2.

Somehow, we need to make checking for memory leakage standard, not necessarily on every commit, but at least something that can be triggered weekly. In my opinion, memory leaks are such fundamental errors that they should have top priority for attention. Easy for me to say, though; tracking them down is not something for which I have much competence or time.

I find the python version dependence strange, I thought the gc got better in 3...

Probably matters by your definition of "better". Better in keeping memory
under control while not causing much overhead, or better in keeping memory
usage to its smallest amounts?

On Thu, Oct 29, 2015 at 4:40 PM, Thomas A Caswell [email protected]
wrote:

In unit/memleak_hawaii3.py
#5353 (comment):

end = val
-print('Average memory consumed per loop: %1.4fk bytes\n' % ((end - start) / float(indEnd - indStart)))
+print('Average memory consumed per loop: %1.4fk bytes\n' %

•  ((end - start) / float(indEnd - indStart)))
  

  +fig, ax = plt.subplots()
  +ax2 = ax.twinx()
  +ax.plot(mem_size, 'r')
  +ax.set_ylabel('memory size', color='r')
  +ax2.plot(coll_count, 'k')
  +ax2.set_ylabel('collect count', color='k')
  +fig.savefig('report')

I find the python version dependence strange, I thought the gc got
better in 3...

—
Reply to this email directly or view it on GitHub
https://github.com/matplotlib/matplotlib/pull/5353/files#r43441186.

There's a similar line in https://github.com/tacaswell/matplotlib/blob/1.5.0_final_work/INSTALL#L187

Good catch, thanks