FIX: round instead of truncate Agg canvas size by tacaswell · Pull Request #8265 · matplotlib/matplotlib · GitHub

As suggested by @njsmith in #8253 round, rather than truncate the size in pixels of the Agg canvas. This fails two tests for me locally

test_boxarrow[0-boxarrow_test_image-png]
test_bbox_inches_tight_suptile_legend[0-bbox_inches_tight_suptile_legend-png]

iirc the second one is also unstable on travis. Do not have time to investigate further tonight, but regenerating two images seems like a small price to pay for making it more reliable to get renders of a fixed pixel size.

This conversion to python int is not done in the other two backends, so presumably it is not needed here. (I see that numpy.int64 and int instances both hash to the corresponding python int.)

I get one failure locally now:

which is due to some text shifting vertically, not really sure what is going on here. The logic inside of bbox_tight is a bit confusing and hacks at the inneards of the transform stack in interesting ways.

FWIW, this doesn't fix the floating point round-off issues in the animation framework that #8253 resolves. I will try to make a simple test script to attach to that issue.

@ngoldbaum, is it possible that using rounding consistently--here and in #8253, in place of int with the nextafter adjustment--would solve the problem?

I don't know - all I tried was running my test yt script with this PR applied in my working copy. It failed, but it's possible there needs to be other code changes in the animation module.

Thanks for the PR. This hasn't seen action for a long time so marking stale. Note that PRs sometimes fall through the cracks, so if this is important, please ping for more reviews.

This avoids marking the figure as stale un-necessarily. Also remove a local variable which is immediately replaced below.

The change to rounding made these images one pixel wider

When rasterizing the figure, the allowed sizes are discrete due to a finite dpi. When getting the renderer (at which point we have committed to rasterizing the figure at this dpi) feedback the actual size.

self.window.ClientToWindowSize(wx.Size(\nRuntimeError: wrapped C/C++ object of type FigureFrameWx has been deleted

This still needs some work and I think tweak around the short-circuit on the resize is not working quite right either.

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.

It seems past time for something like this to get in; let's keep it alive.

Looking at the results here, we have:

Contains text:
- FAILED lib/matplotlib/tests/test_arrow_patches.py::test_boxarrow[png] - matplotlib.testing.exceptions.ImageComparisonFailure: Image sizes do not match expected size: (833, 266, 3) actual size (833, 267, 3)
- FAILED lib/mpl_toolkits/axes_grid1/tests/test_axes_grid1.py::test_image_grid_single_bottom[png] - matplotlib.testing.exceptions.ImageComparisonFailure: Image sizes do not match expected size: (166, 288, 3) actual size (167, 289, 3)
No text in image:
- FAILED lib/mpl_toolkits/axes_grid1/tests/test_axes_grid1.py::test_image_grid[png] - matplotlib.testing.exceptions.ImageComparisonFailure: Image sizes do not match expected size: (332, 332, 3) actual size (333, 333, 3)
Not actually images:
- FAILED lib/matplotlib/tests/test_offsetbox.py::test_annotationbbox_extents[window_extent] - AssertionError:
- FAILED lib/matplotlib/tests/test_offsetbox.py::test_annotationbbox_extents[tightbbox] - AssertionError:

So there's a minor case for putting these into the text overhaul, but it doesn't seem super important, and could be done by itself.

Can we fully trust this with the way we handle HiDPI (where figure.dpi changes based on screen)?

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.

#32038 has now fixed the bug with the canvas height due to a floating-point issue without switching to rounding "true" fractional pixels (e.g., 0.75 of a pixel). I think that rounding such "true" fractional pixels may have some undesired effects when rendering artists that are supposed to run up to the edge of the canvas. In the absence of an explicit request to change how "true" fractional pixels" are handled, I suggest closing this PR.

tacaswell added this to the 2.0.1 (next bug fix release) milestone Mar 11, 2017

efiring reviewed Mar 11, 2017

View reviewed changes

tacaswell modified the milestones: 2.1 (next point release), 2.0.1 (next bug fix release) Mar 11, 2017

dstansby added the backend: agg label Mar 13, 2017

efiring mentioned this pull request Mar 26, 2017

Handle floating point round-off error when converting to pixels for h264 animations #8253

Merged

tacaswell modified the milestones: 2.1 (next point release), 2.2 (next next feature release) Aug 29, 2017

timhoffm mentioned this pull request Oct 3, 2019

Reliably set the output image size #15363

Closed

tacaswell mentioned this pull request Jun 5, 2020

FIX: support Qt 5.15 #17565

Merged

jklymak marked this pull request as draft September 10, 2020 16:25

tacaswell mentioned this pull request Dec 29, 2020

Line2D can snap incorrectly and request snap(x)(y) parameters #19054

Closed

tacaswell mentioned this pull request Dec 7, 2021

[Bug]: bbox floating point size mismatch #21879

Closed

tacaswell force-pushed the fix_stable_pixel_size_rounding branch from 0324a6b to 08a91cf Compare December 7, 2021 22:09

tacaswell modified the milestones: needs sorting, v3.6.0 Dec 7, 2021

tacaswell force-pushed the fix_stable_pixel_size_rounding branch 2 times, most recently from 153b637 to 98acc4f Compare December 7, 2021 22:48

TangChiaHsin mentioned this pull request Apr 9, 2022

misoperation #22812

Closed

QuLogic modified the milestones: v3.6.0, v3.7.0 Jul 5, 2022

github-actions Bot added the status: needs rebase label Oct 18, 2022

FIX: round instead of truncate Agg canvas size

19630b2

tacaswell added 5 commits December 4, 2022 20:36

MNT: if the figure size is not changed, bail early

01e261e

This avoids marking the figure as stale un-necessarily. Also remove a local variable which is immediately replaced below.

API: when forwarding figure resize, round not truncate pixel numbers

85867e0

TST: regenerate a test image

80a7a02

The change to rounding made these images one pixel wider

API: feedback to figure size due to pixel counts

701ae74

When rasterizing the figure, the allowed sizes are discrete due to a finite dpi. When getting the renderer (at which point we have committed to rasterizing the figure at this dpi) feedback the actual size.

API: add API change note

7acb303

tacaswell force-pushed the fix_stable_pixel_size_rounding branch from 98acc4f to 7acb303 Compare December 5, 2022 01:41

github-actions Bot removed the status: needs rebase label Dec 5, 2022

github-actions Bot added the status: needs rebase label Dec 8, 2022

tacaswell modified the milestones: v3.7.0, future releases Dec 9, 2022

github-actions Bot added the status: inactive Marked by the “Stale” Github Action label Apr 9, 2023

tacaswell mentioned this pull request Apr 11, 2023

ylabel clipped with Qt5Agg backend when saving figure to png with both bbox_inches='tight' and pad_inches=0 #8473

Open

efiring modified the milestones: future releases, v3.8.0 Jun 5, 2023

ksunden modified the milestones: v3.8.0, future releases Aug 8, 2023

tacaswell mentioned this pull request Dec 8, 2023

Fixing Issue #21879: Changed int to round to help with floating point issue #27473

Open

2 of 5 tasks

QuLogic linked an issue Jul 14, 2025 that may be closed by this pull request

Reliably set the output image size #15363

Closed

github-actions Bot removed the status: inactive Marked by the “Stale” Github Action label Jul 19, 2025

QuLogic reviewed Jul 24, 2025

View reviewed changes

github-actions Bot added the status: inactive Marked by the “Stale” Github Action label Sep 24, 2025

QuLogic mentioned this pull request Jul 13, 2026

Fix occasional misalignment in reported mouse position (also fixes a bug with canvas height) #32038

Merged

1 of 2 tasks

                       # in get_diff_image
-                      _, _, w, h = self.figure.bbox.bounds
+                      _, _, w, h = np.round(self.figure.bbox.bounds).astype(int)
                       w, h = int(w), int(h)

+                      # we know we are using Agg, thus are tied to discrete sizes
+                      # set by the dpi.  Feed this back so that the transforms are
+                      # mapped to the available pixels
+                      self.figure.set_size_inches(w / dpi, h / dpi)

FIX: round instead of truncate Agg canvas size by tacaswell · Pull Request #8265 · matplotlib/matplotlib · GitHub

Repository navigation

FIX: round instead of truncate Agg canvas size - #8265

FIX: round instead of truncate Agg canvas size#8265tacaswell wants to merge 6 commits intomatplotlib:mainmatplotlib/matplotlib:mainfrom tacaswell:fix_stable_pixel_size_roundingtacaswell/matplotlib:fix_stable_pixel_size_roundingCopy head branch name to clipboard

Conversation

tacaswell commented Mar 11, 2017

Uh oh!

efiring Mar 11, 2017

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Mar 11, 2017

Uh oh!

ngoldbaum commented Mar 12, 2017

Uh oh!

efiring commented Mar 15, 2017

Uh oh!

ngoldbaum commented Mar 15, 2017

Uh oh!

jklymak commented Jul 25, 2019

Uh oh!

tacaswell commented Dec 5, 2022

Uh oh!

github-actions Bot commented Apr 9, 2023

Uh oh!

efiring commented Jun 5, 2023

Uh oh!

QuLogic commented Jul 18, 2025

Uh oh!

QuLogic Jul 24, 2025

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Sep 24, 2025

Uh oh!

ayshih commented Jul 17, 2026

Uh oh!

This branch has not been deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants

FIX: round instead of truncate Agg canvas size#8265
tacaswell wants to merge 6 commits into
matplotlib:mainfrom
tacaswell:fix_stable_pixel_size_rounding