Brian McFee

This comment summarizes some thoughts on the display module design after I sank way more time and thought into it yesterday than I really should have. TLDR: I'm becoming convinced...

... nevermind, weakrefdictionary isn't going to work on axes objects.

> Thanks for this wildly in-depth sleuthing. How bad will the memory issues get, really? Ehhh, most of the time it would be fine, but in some cases it could...

Back on this for a bit today, came across this issue, which if resolved, would almost cover half of what we need here: https://github.com/matplotlib/matplotlib/issues/19479 (making cycler's accessible / sharable across...

OOOoookay, I think I have a clumsy prototype with weakkeydict that actually works. It turns out that it does actually work fine with axes objects directly, but the problem is...

This modernization PR is getting unstuck now. I'm doing a bit of cleanup along the way and regenerating regression test targets where needed (eg because font rendering changed slightly). I...

> the new version isn't actually covering the full y range, is it? MIDI would be 0->127, unless I'm misunderstanding. Yeah, good catch. This is from https://github.com/craffel/mir_eval/blob/6b82ddef3675ba7613aa454f74bc182ee3ade4a3/mir_eval/display.py#L713 Removing that line...

> Let's revisit this after I rewrite the styling code to use fresh cyclers and we can remove the hijacking kludge. Got this part done, and it's definitely much cleaner:...

Alright, I'm sticking a fork in this. :sweat_smile: All deprecated functionality is purged. Property cyclers are used everywhere, limiting and scaling behaviors are all compliant with matplotlib idioms. Generated figures...

> it leaves some funny whitespace only on the top in some cases Heh. The "some cases" here is doing a lot of work :grin: What's actually happening is that...