fix: use last focused/resized view to determine shape of BqplotImageView

[iisakkirotko]

Pull Request Template

Description

Instead of always using the lowest cid view to determine the aspect ratio / shape of a Bqplot image viewer, use the last focused / resized one. focused_at describes the time that the window of that view was last focused, while resized_at describes when each viewer was last resized, so this way distortion from resizing based on a view in a popout window will be fixed once the original window is brought into focus.

Needs glue-viz/bqplot-image-gl#112 to work. Let me know if you need this to use the old behaviour as a fallback with previous versions of bqplot-image-gl.

TODO:

• Bump bqplot-image-gl in the dependencies once feat: resize & focus tracking in view_data bqplot-image-gl#112 is released.

[maartenbreddels]

Let me know if you need this to use the old behaviour as a fallback with previous versions of bqplot-image-gl.

I think it's fine like this, if we also bump

glue-jupyter/setup.cfg

Line 26 in 1876875

bqplot-image-gl>=1.5.0

Otherwise some people might get the old behaviour.

[maartenbreddels]

We can use version 1.6.0 now (https://pypi.org/project/bqplot-image-gl/1.6.0/)

[dhomeier]

Could you rebase to try and get the tests running?

[iisakkirotko]

@dhomeier done!

[dhomeier]

@maartenbreddels: visual failure is related to astropy/astropy#17683 / widgetti/solara#990, right?

[maartenbreddels]

Indeed

[dhomeier]

Hmm, updating to solara 1.44 seems to have broken more than it fixed – any ideas what those pytest-playwright failures are?

[dhomeier]

The last run prior to upgrading solara already was at pytest-playwright==0.6.2 so thought it was already pinned somehow. But I think that already raised on an earlier point in collection, so perhaps does not tell us much.

[maartenbreddels]

Downgrading pytest-playwright helped, but now I hit widgetti/solara#606 - which I can now reproduce (so working on that now).

[dhomeier]

Log looks like two comparisons failed, but I only see one in

Changed background colour probably would not be critical, but any idea where it comes from?

[dhomeier]

Something therein is pinning jupyter-client 7.x, jupyterlab 3.x and notebook 6.x now, among others.
But for visualtest only we probably can/have to live with that for now.

[dhomeier]

Approving as is, but if you want to pull the CI fixes out to further investigate the frame colour issue @maartenbreddels, fine with me, too.

[maartenbreddels]

I've digged into why this is happening, and I think this is not a related change, and we should be able to merge this.

It seems that the output we see now is correct, since I see that bqplot-image-gl gets fed the border with values 0.2 0.2 0.2 1. which is an opaque dark grey. However, I don't understand why the current reference image is white. This might be a change in glue (maybe @astrofrog can confirm).

[codecov]

Codecov Report

Attention: Patch coverage is 33.33333% with 2 lines in your changes missing coverage. Please review.

Project coverage is 86.02%. Comparing base (52a3c29) to head (2e5310b).
Report is 7 commits behind head on main.

Files with missing lines	Patch %	Lines
glue_jupyter/bqplot/image/viewer.py	33.33%	2 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #475   +/-   ##
=======================================
  Coverage   86.01%   86.02%           
=======================================
  Files          91       91           
  Lines        5271     5272    +1     
=======================================
+ Hits         4534     4535    +1     
  Misses        737      737           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[maartenbreddels]

@dhomeier if you agree, please rebase and merge, not squash.

[dhomeier]

Can't see the plots now, since the Image Comparison page redirects to a nonexistent server, but sounds OK, thanks!