Normalize gridspec ratios to lists in the setter. #17291
Conversation
|
.. though I guess the test failure is real? |
|
at some point I figured out how to look at the failing images on CI, but I cannot figure it out again :-( |
|
Artifacts are here |
lib/matplotlib/gridspec.py
Outdated
| @@ -49,9 +49,9 @@ def __init__(self, nrows, ncols, height_ratios=None, width_ratios=None): | |||
|
|
|||
| def __repr__(self): | |||
| height_arg = (', height_ratios=%r' % (self._row_height_ratios,) | |||
| if self._row_height_ratios is not None else '') | |||
| if len({*self._row_height_ratios}) != 1 else '') | |||
There was a problem hiding this comment.
Slightly prefer
| if len({*self._row_height_ratios}) != 1 else '') | |
| if len(set(self._row_height_ratios)) != 1 else '') |
for better readability (also below).
There was a problem hiding this comment.
I think I already won this battle: there's multiple occurrences of {*foo} in the codebase already and none of set(*foo) :-)
There was a problem hiding this comment.
Um, set(*foo) is not valid. It's set(foo) because set() takes a single iterable as argument.
That's part of why I prefer it here: "make as set from this list" is just a single concept, whereas "unpack this iterable and create a set from its elements" is two concepts. (I don't have that much brain capacity https://youtu.be/UANN2Eu6ZnM 😄).
Yes, you've already sneaked 5 times {*foo} into the code base. - I hope I didn't approve these 😝.
But for real, {*a, *b} makes sense for multiple iterables, but for a single iterable set(a) is more readable than {*a}.
There was a problem hiding this comment.
sure, changed to set().
|
Actually that looks like a real failure, see #17304. |
|
Currently blocked by #17304. |
fb8ddd1 to
583340e
Compare
|
hopefully working now |
|
Looks like another failure in test_pickle, but that one already has a nonzero tolerance for certain architectures which suggests fp issues so I don't feel bad putting in the same tolerance for x86_64. |
|
And another test is failing only on CI with a tiny delta (I can't even repro it locally), so I just upped the tol. The baseline and results are basically impossible to differentiate visually, |
This avoids having to check for `ratios == None` every time someone actually accesses the ratios. I just deleted the never used and private `vstackeq` and `hstackeq` instead of changing their signatures to make height_ratios/width_ratios non-optional, given that they should be simple enough to bring back if needed.
This avoids having to check for
ratios == Noneevery time someoneactually accesses the ratios.
I just deleted the never used and private
vstackeqandhstackeqinstead of changing their signatures to make height_ratios/width_ratios
non-optional, given that they should be simple enough to bring back if
needed.
PR Summary
PR Checklist