[Bug]: memleak is failing, but not really? #21646
Description
Bug summary
For 3.5.0 checklist, I ran the memleak script, and it failed in an unusual way. Some time after the warmup, there is a single large jump in memory usage.
A bisect points to d11022e, but that seems strange. Reverting that drops the leak from 2K to 155.2168 bytes per loop. If this is something we can't fix, we should work around it in the test somehow so it doesn't count.
Code for reproduction
python tools/memleak.py agg 1000 agg.pdfActual outcome
There is a random large jump in memory after about 100 steps. Then it says we 'leaked' a couple KB per loop.
Expected outcome
No leaks
Additional information
No response
Operating system
No response
Matplotlib Version
3.5.0
Matplotlib Backend
No response
Python version
3.7.6
Jupyter version
No response
Installation
from source (.tar.gz)