Repository navigation
Improve initial macosx device scale - #18274
dopplershift wants to merge 2 commits into
Conversation
Sorry, something went wrong.
This reverts commit db6fb4d. This should now be fixed with the previous commit.
|
Is this robust across mixed dpi devices? I seem to remember some issue with folks who had retina and non retina setups |
Sorry, something went wrong.
|
Well, the goal here is to be safe across those, that's the reason for the whole device_scale song and dance. Whether we're actually doing that robustly 🤷♂️ . The Apple Docs lead me to believe we're not exactly doing things the way they want, but I'm not signing up to rewrite all of that--some brief testing with some more APIs has led me to conclude I don't understand this well enough. This PR only changes things to use an alternate API initially to try to get the scale factor from the Window class rather than requiring a full graphics context. The rest of the device scale handling is unchanged from before. The change now means it gives me the proper value that we eventually want, at least on my Retina system, of 2.0 at a time in the init sequence that allows us to avoid ever making a figure with a lower DPI. This makes the test that was constantly failing on my machine now pass. It also is passing Travis, at least this time. |
Sorry, something went wrong.
|
To confirm, what DPI is the figure saved as now? |
Sorry, something went wrong.
|
Before: 100 DPI initially, 200 DPI after display |
Sorry, something went wrong.
|
Has the macosx backend been wrong all this time? Figures should be saved at the logical DPI, so 100, not 200. |
Sorry, something went wrong.
|
It's possible I introduced the shift while fixing issues. Let me see if I can carve out some time how this should work (Apple's view of things aside). |
Sorry, something went wrong.
This was not originally implemented in matplotlib#19126, but causes some inconsistencies with other backends. This also sets the initial scale as implemented in matplotlib#18274.
This was not originally implemented in matplotlib#19126, but causes some inconsistencies with other backends. This also sets the initial scale as implemented in matplotlib#18274.
This was not originally implemented in matplotlib#19126, but causes some inconsistencies with other backends. This also sets the initial scale as implemented in matplotlib#18274.
This was not originally implemented in matplotlib#19126, but causes some inconsistencies with other backends. This also sets the initial scale as implemented in matplotlib#18274.
PR Summary
Fixes #18213. The problem is that the initial device scale in the macosx backend was set to 1.0--it would then update on the first display of a figure. This calls an API on window creation to update the device scale. This fixes the failing test for me locally.
PR Checklist