@jklymak Correct, the only way we know how big the text is going to be is to draw it (exact same problem we have with constrained layout / tight layout). The tests all have explicit .draw() calls in them.

I am a bit concerned that this PR is going to rapidly spiral out of scope.

I don't think we want to start adding a second (third, depending on if you count tight_layout and constrained_layout is different things) thing that will internally force draws as that can easily start to add multiples to the draw time.

On the other hand, we can just skip this logic when the figure isn't drawn yet, but that has the issue that saving twice will give different results and saving from the GUI will "work", but saving from a script won't.

Unless there is a third option here (:crossed_fingers: ) I don't think this will be able to go in. Despite it making things better in some ways, it will make things worse in other ways.

I think we could possibly get a provisional extent somehow which would be an overestimate. Might come in handy in other situations as well. Something like text_get_approx_extent that assume the text is font size high and len(str) * font size wide.

Something like text_get_approx_extent that assume the text is font size high and len(str) * font size wide.

That is true, @RiazCharania @Jordan711 I suggest having a look at how we do this in the autoticker / autolabeler where we use an estimate of the tick label size to pick the number of ticks.

This may grow from "bug fix" to "new feature".

I would actually prefer avoiding overestimates, see #16153, #16171 for issues with that.
Normally text rendering is cached (per-string) so this shouldn't be too costly, although I'm not sure we also cache the extents (but I guess we could too).

quick question, running this code seems to work

import matplotlib.pyplot as plt
import matplotlib as mpl

fig, ax = plt.subplots()
plt.plot([0, 1], label="foo")
plt.text(0, 1, "Sample")
plt.text(1, 0, "Text")
plt.legend() # Create the legend

fig.savefig("legend.png")

resulting in

is there a draw in there that im missing?

I would actually prefer avoiding overestimates, see #16153, #16171 for issues with that.

I agree with that, but in this case you are just trying to not put something near the text so an overestimate is probably OK....

@RiazCharania

Hmmm, interesting. Can you insert a print statement into the place where you get the extents and see what they are set to? Sometimes matplotlib is mysterious when things are drawn and its possible the legend is being updated after the text is drawn?

running
import matplotlib.pyplot as plt
import matplotlib as mpl

fig, ax = plt.subplots()
plt.plot([0, 1], label="foo")
plt.text(0, 1, "Sample")
plt.text(1, 0, "Text")
print('pre plt.legend()')
plt.legend() # Create the legend
print('post plt.legend()')

fig.savefig("legend.png")

with print(texts) line immediately after texts = ... in legend.py

Actually (with approx. 10s thought...) I think we should just make Text.get_window_extent perform whatever drawing is necessary (creating a renderer if necessary), and then make sure the result is cached so we only do it once per string/renderer.

yeah, but based on the above it seems to work anyway?

Since this Pull Request has not been updated in 60 days, it has been marked "inactive." This does not mean that it will be closed, though it may be moved to a "Draft" state. This helps maintainers prioritize their reviewing efforts. You can pick the PR back up anytime - please ping us if you need a review or guidance to move the PR forward! If you do not plan on continuing the work, please let us know so that we can either find someone to take the PR over, or close it.

This was done at #27469.