2050692 - last character in channel names on Slack cut off

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Attachment: screenshot: characters cut off

Firefox Nightly 154.0a1 on Windows 11

Often the last character in channel names on Slack is cut off.

Comment 1

[Glenn Watson [:gw]]

Could you attach your about:support (or comment here what display scale ratio you're running at)?

Comment 2

[Glenn Watson [:gw]]

I can't repro locally on Linux at a variety of resolutions and device scales yet - is there anything about your setup or repro steps that might be worth trying here?

Comment 3

[Hiroyuki Ikezoe (:hiro)]

I can see it. I have to join the mozilla-org channel. FWIW, my DPR is 125%.

Anyway, that's odd, the clipped #mozilla-org is no clipped by any scroll container. And yeah, I think we can see #mozilla-ceo is also cut off in comment 0. Oops, #moco is also cut off.

Oops ^ above is wrong.

<span dir="auto" class="p-channel_sidebar__name" data-qa="channel_sidebar_name_mozilla-org" data-sk="tooltip_parent"><span>mozilla-org</span></span>

The .p-channel_sidebar__name has overflow: hidden style.

Comment 4

[Hiroyuki Ikezoe (:hiro)]

FWIW, the styles are:

p-channel_sidebar__name {
  color: var(--_sidebar-row-color,inherit);
  font-weight: var(--_sidebar-row-text-weight,inherit);
  opacity: var(--_sidebar-row-text-opacity,1);
  text-overflow: ellipsis;
  white-space: nowrap;
  overflow: hidden;
}

Comment 5

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Windows Device Pixel Ratio is 1.25

Comment 6

[Hiroyuki Ikezoe (:hiro)]

Okay,

.c-scrollbar__hider {
 transform: translateZ(0);
}

This transform style is a trigger. If you uncheck the style on DevTools. the cut-off is gone.

Comment 7

[Hiroyuki Ikezoe (:hiro)]

Though mozregression didn't reliable work but I did kida git bisect. Now I am ~100% sure this is regressed by bug 2004666.

Comment 8

[Glenn Watson [:gw]]

I'm still trying to repro this (setting up a comparable windows env), in the meantime it's possible that this might be fixed by https://phabricator.services.mozilla.com/D309279 if anyone who can repro it has time to apply that and check.

Comment 9

[Hiroyuki Ikezoe (:hiro)]

No, it's not. I applied D309279 and D309280.

Comment 10

[Glenn Watson [:gw]]

Minimized wrench repro reftest:

root:
  items:
    - type: stacking-context
      bounds: [100.5, 80, 300, 100]
      transform: identity()
      is-2d: true
      items:
        - type: text
          text: "XXXXXXXX"
          origin: [0, 30]
          size: 32
          color: [0, 0, 0, 1]
          font: "Ahem.ttf"

root:
  items:
    - type: text
      text: "XXXXXXXX"
      origin: [100.5, 110]
      size: 32
      color: [0, 0, 0, 1]
      font: "Ahem.ttf"

Comment 11

[Glenn Watson [:gw]]

Attachment: Bug 2050692 - Leave offset-only reference frames sub-pixel when snapping device-mode text.

A device-mode text run snapped its full reference-frame origin to the device
grid. An offset-only reference frame - one that merely positions content (an
nsIFrame layout offset or an identity transform such as translateZ(0)) - then
shifted the text ~1px off where the same content renders unframed, clipping the
last character of Slack channel names on Win11 at fractional device scales.

Mark such frames with an is_offset_only flag on ReferenceFrameKind::Transform,
set by the embedder at the single site that synthesizes them (the offset-only
reference frame in wr_dp_push_stacking_context, mirrored by wrench). Snap their
text like unframed content so the static origin stays sub-pixel; frames that
genuinely scale or rotate content keep the full-origin snap.

This also fixes css-transforms/backface-visibility-hidden-animated-002.html
(text inside a will-change:transform element), so drop its now-passing
expectation.

Comment 12

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/56bc004050fd
https://hg.mozilla.org/integration/autoland/rev/30d67c948213
Leave offset-only reference frames sub-pixel when snapping device-mode text. r=gfx-reviewers,lsalzman

Comment 13

[Iulian Moraru]

https://hg.mozilla.org/mozilla-central/rev/30d67c948213

Comment 14

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Behavior did not change with the latest Nightly.

Comment 15

[Daniel Holbert [:dholbert]]

Reclassifying given the regressor and assignee.

Comment 16

[Glenn Watson [:gw]]

Attachment: Bug 2050692 - Render text runs in device space and snap clips to the device grid

Slack channel-name characters were clipped at fractional device scales. Render
text runs at their exact device positions and keep a device-space text run's
clips unsnapped, so a fractional clip edge no longer rounds inward and shaves
the last glyph; a snapping primitive instead snaps every clip in its chain to
the device grid. Also rounds out the bounding rect of non-snapped primitives and
drops the now-dead is_offset_only reference-frame flag.

Updates the affected wrench reftest baselines and adds a unit test asserting the
device-text clip snap policy (the rendered difference is a sub-pixel clip shift
that headless software rasterization cannot surface, so it can't be reftested).

Comment 17

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/0b40ca4b71f3
https://hg.mozilla.org/integration/autoland/rev/eb835be37e1f
Render text runs in device space and snap clips to the device grid r=gfx-reviewers,lsalzman

Comment 18

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/c522ec359f7f
https://hg.mozilla.org/integration/autoland/rev/6bfe3eff1a9c
Revert "Bug 2050692 - Render text runs in device space and snap clips to the device grid r=gfx-reviewers,lsalzman" for causing wr failures at attributes-for-embedded-content-and-images/input-image-inline-alt.html

Comment 19

[rperta]

Backed out for causing wr failures at attributes-for-embedded-content-and-images/input-image-inline-alt.html
Backout link
Push with failures
Failure log(s)
and also this reftest

Comment 20

[Glenn Watson [:gw]]

Fixed the fuzzy annotation for this test

Comment 21

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/98b6b93145bf
https://hg.mozilla.org/integration/autoland/rev/79a3125a8baa
Render text runs in device space and snap clips to the device grid r=gfx-reviewers,lsalzman

Comment 22

[Serban Stanca [:SerbanS]]

https://hg.mozilla.org/mozilla-central/rev/79a3125a8baa