[Tracking Requested - why for this release]:

In case of setting Windows text zoom to bigger but dpi scale is staying 100%.
The user intends to make only the text bigger. However, the user do not want to make everything bigger.

Despite this, all sizes are bigger in Firefox 103.0a1.
This is inconsistent with other Windows applications.

STR:

1. Set Windows text zoom to 133% but dpi scale is 100%
2. Start Firefox 102.0b6 with new profile and quit
3. Start 103.0a1 (autoland 20220609230435)

Actual results:
Toolbars(tab, nav, bookmarks, findbar) height is too big.
Window size(height/width) is also wider than the size of previous session.
So, window controls and right/bottom edge of browser are out of screen.
Web Page Content is also too big not only texts but also images.

Expected results:
Toolbars(tab, nav, bookmarks, findbar) height should be same as 102.0b6.
Window size (height/width) should retain the size of the previous session(before update).
Web Page Content should retain the size of the previous session(before update).

Regression window:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=cb7c4d0b946209a2a3b05cd5a35e31fadaf55135&tochange=6e1cbabce0af992a8ff5158959ac9dd69bea23a7

:emilio, since you are the author of the regressor, bug 1773342, could you take a look?
For more information, please visit auto_nag documentation.

Broken window size restore is something I had not realized... It seems hard to detect, just like it's hard to detect screen resolution changes that happen while Firefox is closed though, it's just an instance of that problem...

Regarding too big toolbar buttons and such, that's expected and how it has behaved on Linux for a long time. It also seems to match other browsers including Edge and Chrome unless I'm missing something... The previous behavior of the text zoom was basically accidental.

Am I missing something else? I guess I could try to avoid session restore on desktop position / dpi changes, or add some sort of migration... Barret, do you know if there's some sort of session store migration mechanism that would allow me to undo the text zoom effect as a one off?

I think this is expected unless you're seeing something else on other browsers as well? WIN32 apps usually can get around with only making text bigger but that on the web is a bit more dangerous / prone to break things.

If the user wants to make everything larger, just increase dpi right?.
In that case, the content area will be sacrificed.

I don't like that, so I only set to bigger text in Windows settings.

(In reply to Emilio Cobos Álvarez (:emilio) from comment #3)

I think this is expected unless you're seeing something else on other browsers as well?

Yes. Chrome seems to equate text and dpi scaling.

Also, users will have to reconfigure all site-specific Fullzoom levels each time they visit the website.

Well, only once right?

This is reasonably straight-forward and should allow us change the
desired behavior on platforms if / when we need.

Also, this adds tests for the feature by using the relevant float pref
(though we can't easily assert the full-zoom behavior because it changes
the size of the reftest window itself).

(In reply to Emilio Cobos Álvarez (:emilio) from comment #2)

...
Am I missing something else? I guess I could try to avoid session restore on desktop position / dpi changes, or add some sort of migration... Barret, do you know if there's some sort of session store migration mechanism that would allow me to undo the text zoom effect as a one off?

I don't know enough about session store to answer this question. (I only know about its shutdown IO behaviour).

Ah, sorry, might've misremembered... Do you know who would know?

According to the module list, perhaps :dao, :daleharvey or :farre

Backed out for causing failures on 1773633.html

• backout: https://hg.mozilla.org/integration/autoland/rev/c4803fd0cb20859f536a64ef697172d023d9285f
• push: https://treeherder.mozilla.org/jobs?repo=autoland&group_state=expanded&revision=5df8a7315ddbf4e514b424b79190f160ef6daea3
• 2nd push: https://treeherder.mozilla.org/jobs?repo=autoland&group_state=expanded&revision=348eb31da96b3473b0705e17b23538f2a40b2d96&selectedTaskRun=DRreGQS0TeSEpVPPdZMPQg.0
• failure log: https://treeherder.mozilla.org/logviewer?job_id=381090588&repo=autoland&lineNumber=13875

[task 2022-06-13T02:37:37.506Z] 02:37:37     INFO - REFTEST TEST-START | layout/reftests/bugs/1773633.html != layout/reftests/bugs/1773633.html
[task 2022-06-13T02:37:37.507Z] 02:37:37     INFO - REFTEST INFO | Pref ui.textScaleFactor not found, will be added
[task 2022-06-13T02:37:37.507Z] 02:37:37     INFO - REFTEST INFO | SET PREFERENCE pref(ui.textScaleFactor,150)
[task 2022-06-13T02:37:37.508Z] 02:37:37     INFO - REFTEST TEST-LOAD | file:///opt/worker/tasks/task_165508560044536/build/tests/reftest/tests/layout/reftests/bugs/1773633.html | 2063 / 2066 (99%)
[task 2022-06-13T02:37:37.557Z] 02:37:37    ERROR - REFTEST ERROR | file:///opt/worker/tasks/task_165508560044536/build/tests/reftest/tests/layout/reftests/bugs/1773633.html | can't drawWindow remote content
[task 2022-06-13T02:37:37.558Z] 02:37:37    ERROR - REFTEST ERROR | Screen resolution is too low - USE_WIDGET_LAYERS was disabled. Since E10s is enabled, there is no fallback rendering path!
[task 2022-06-13T02:37:37.559Z] 02:37:37    ERROR - REFTEST ERROR | If you cannot increase your screen resolution you can try reducing gecko's pixel scaling by adding something like '--setpref layout.css.devPixelsPerPx=1.0' to your './mach reftest' command (possibly as an alias in ~/.mozbuild/machrc). Note that this is inconsistent with CI testing, and may interfere with HighDPI/reftest-zoom tests.
[task 2022-06-13T02:37:37.559Z] 02:37:37     INFO - REFTEST INFO | drawWindow flags = DRAWWINDOW_DRAW_CARET | DRAWWINDOW_DRAW_VIEW; window size = 800,705; test browser size = 800,1000
[task 2022-06-13T02:37:37.562Z] 02:37:37     INFO - [Parent 2353, Main Thread] WARNING: Remote iframe not rendered: file /builds/worker/checkouts/gecko/layout/generic/nsSubDocumentFrame.cpp:1288
[task 2022-06-13T02:37:37.565Z] 02:37:37     INFO - REFTEST INFO | RESTORE PREFERENCE pref(ui.textScaleFactor, <no value set>) (clearing user pref)
[task 2022-06-13T02:37:37.580Z] 02:37:37     INFO - REFTEST TEST-PASS | layout/reftests/bugs/1773633.html != layout/reftests/bugs/1773633.html | image comparison, max difference: 255, number of differing pixels: 246000

https://hg.mozilla.org/mozilla-central/rev/6fdfef57e6d7

I had I think same issue but on Linux: https://bugzilla.mozilla.org/show_bug.cgi?id=1773823