Bug summary

I stumbled upon this issue when I tested the codes in#30305.

basically, I am trying to create a canvas buffer with a prescribed pixel size. The formula I understood thatbuffer_rgba() output should follow is the figure's size in inch, multiplied by the dpi setting.

However, upon scanning a range of desired sizes, I found that at a small subset of sizes, thebuffer_rgba() output buffer size failed to match - and it is always 1 pixel larger than desired.

The below script tests this for output pixel size between 1 and 1000, and it discovered 21 outliers, see below

my 11yr old took a quick look and found some vague patterns - 29 * 2^N is always in this list, other than that, the gap between the adjacent outlier frequently show 25 for some reason - I do not believe these are related to numerical precision, but perhaps it may be related to rounding algorithms?

  i      w   h[29, [28, 28]][58, [57, 57]][116, [115, 115]][205, [204, 204]][207, [206, 206]][232, [231, 231]][410, [409, 409]][414, [413, 413]][439, [438, 438]][464, [463, 463]][489, [488, 488]][803, [802, 802]][820, [819, 819]][828, [827, 827]][845, [844, 844]][853, [852, 852]][878, [877, 877]][903, [902, 902]][928, [927, 927]][953, [952, 952]][978, [977, 977]]

Code for reproduction

importmatplotlib.pyplotaspltforiinrange(1,1000):fig=plt.figure(figsize=(i*0.01,i*0.01),dpi=100)plt.draw()fig.canvas.draw()w,h=fig.canvas.get_width_height()plt.close(fig)ifnot (i==wandi==h):print([i, [w,h]])

Actual outcome

at some figure sizes, the output buffer has an extra pixel, and this is not predictable

Expected outcome

expect the output buffer size should match figure size in inch * dpi

Additional information

No response

Operating system

Ubuntu

Matplotlib Version

3.10.3

Matplotlib Backend

No response

Python version

3.10

Jupyter version

No response

Installation

pip