1774271 - Pasted objects in pdf.js editor are sometimes empty or show rendering artifacts

Status: VERIFIED FIXED

(Core :: Widget: Gtk, defect, P1)

Description

[Marco Castelluccio [:marco]]

STR:

1. Enable PDF.js editing by setting pdfjs.annotationEditorEnabled to true
2. Draw something somewhere
3. Select what you drew
4. Press CTRL+C
5. Press CTRL+V

When pasting, sometimes the pasted content is empty or has artifacts.
It is easily reproducible if you press CTRL+V, then CTRL+Z to cancel the paste, then CTRL+V again in a loop.

Comment 1

[Lee Salzman [:lsalzman]]

Marco, can you please detail why you think this is a regression and what you think caused it?

Comment 2

[Marco Castelluccio [:marco]]

The bug isn't reproducible if you set gfx.canvas.accelerated to False, so it must be a regression of the GPU-accelerated canvas implementation.

Comment 3

[Lee Salzman [:lsalzman]]

Can you please attach your about:support information?

Comment 4

[Marco Castelluccio [:marco]]

https://bugzilla.mozilla.org/attachment.cgi?id=9281355

Comment 5

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1770088

Comment 6

[Emilio Cobos Álvarez [:emilio]]

I can repro this on X11+EGL and Wayland, but not on X11+GLX nor macOS (with accelerated canvas on). Same with bug 1774214, actually.

Comment 7

[Darkspirit]

Is this a duplicate of bug 1773755? (Caused by bug 1735929 and fixed by bug 1773968?)

Comment 8

[Emilio Cobos Álvarez [:emilio]]

I could repro bug 1774214 with a rather old build (2022-01-01) but I didn't try this one.

Comment 9

[Emilio Cobos Álvarez [:emilio]]

Attachment: Test-case that consistently produces artifacts for me.

If I change the thickness/width/height variables to be whole pixels it doesn't.

Comment 10

[Emilio Cobos Álvarez [:emilio]]

Attachment: Screencast of Emilio's nightly

Comment 11

[Emilio Cobos Álvarez [:emilio]]

I can't repro with my test-case nor the original bug if I set widget.dmabuf-webgl.enabled=false (and restart). Can you confirm Marco?

Comment 12

[Marco Castelluccio [:marco]]

Yes, same here!

Comment 13

[Emilio Cobos Álvarez [:emilio]]

Martin does this ring any bell? It seems something going very wrong with shared DMABuf surfaces.

Comment 14

[Emilio Cobos Álvarez [:emilio]]

Attachment: DMABuf log when adding a canvas in the test-case

Comment 15

[Emilio Cobos Álvarez [:emilio]]

Ah, so at first I thought this was about the fractional stuff, but no, it's about the difference between width and height. Maybe a stride issue in the buffer or something? If in comment 9 I set width = 100 and height = 100 then there are no glitches.

Comment 16

[Emilio Cobos Álvarez [:emilio]]

So with mismatched width/height I get:

Unsupported modifier, resource creation failed.
XXX: resource creation failed

In the terminal, and glitches. With matched size I don't get that.

Comment 17

[Emilio Cobos Álvarez [:emilio]]

Attachment: Simpler test-case.

Comment 18

[Martin Stránský [:stransky] (ni? me)]

I can reproduce that on Intel only, radeon and NVIDIA works ok.

On Fedora 36 I don't see

Unsupported modifier, resource creation failed.
XXX: resource creation failed

Comment 19

[Martin Stránský [:stransky] (ni? me)]

I suspect this is a bug in EGL_MESA_image_dma_buf_export where stride is not exported correctly or we don't import it correctly.
I think EGL_MESA_image_dma_buf_export is a bit fresh and not well tested (see Bug 1774075).

Comment 20

[Martin Stránský [:stransky] (ni? me)]

We can extend https://phabricator.services.mozilla.com/D149238 to disable EGL_MESA_image_dma_buf_export on Intel too.

Comment 21

[Martin Stránský [:stransky] (ni? me)]

Hm, I can find anything obviously wrong here - exported dmabuf params are the same on amd and Intel for with/height 100/200, stride is 512 in both cases. DRM format looks also ok - ABGR8888.

Comment 22

[Martin Stránský [:stransky] (ni? me)]

Attachment: Bug 1774271 [Linux] Don't use EGL_MESA_image_dma_buf_export in Mesa/Intel due to https://gitlab.freedesktop.org/mesa/mesa/-/issues/6688 r?jgilbert

Depends on https://phabricator.services.mozilla.com/D149238

Depends on D149238

Comment 23

[Daniel Holbert [:dholbert]]

(In reply to Darkspirit from comment #7)

Is this a duplicate of bug 1773755?

I'll leave that up to stransky/other developers working directly on this -- but FWIW, here a few relevant notes:

• I can reproduce both bugs in Ubuntu 22.04 on my ThinkPad (using emilio's "Simpler test-case" attachment for this bug here)
• I also hit what seems to be this same sort of corruption on the web UI for my Synology NAS device; see screenshot in attachment 9281718 [details] (posted in bug 1773755 comment 14).
• I used mozregression to get a regression range for this bug here (with that testcase from emilio), and it matches bug 1773755's regression range (in bug 1773755 comment 1) -- both of these bugs seem to be regressions from bug 1735929:
  https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=145141e21a29ad02ea84392f9a5e51803e68a3b7&tochange=f9c796ae5117065116947d61b8fe2bcae1f6b78c
  (--> updating regressed-by field to replace bug 1770088; this started happening more recently than when that patch landed, so that's not proximally the thing-that-regressed-this, even though this is about a configuration that was enabled in that bug.)

Comment 24

[Daniel Holbert [:dholbert]]

Also FWIW, I just tested a local debug+opt build with https://phabricator.services.mozilla.com/D149238 and https://phabricator.services.mozilla.com/D149608 (this bug's associated patch stack).

I can confirm that with that patch stack applied, I get the expected rendering on emilio's testcase here, on bug 1773755 (windy.com), bug 1774214 (web.whatsapp.com), and on my Synology NAS device's web UI.

Comment 27

[Chris Siebenmann]

Recent Firefox Nightly builds running on Linux X11 on Intel graphics (Fedora 35 64-bit, non-compositing window manager, i7-8700K onboard graphics) unpredictably render some Grafana dashboard panels in a completely garbled way. Bisecting with mozregression lands me on https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=145141e21a29ad02ea84392f9a5e51803e68a3b7&tochange=f9c796ae5117065116947d61b8fe2bcae1f6b78c and as a result this bug. A screenshot that looks like my experience is https://twitter.com/moderat10n/status/1538923071939063808. Not all panels on a Grafana dashboard are mis-rendered but usually some are, frequently graph panels.

Comment 28

[Martin Stránský [:stransky] (ni? me)]

Attachment: Bug 1774271 [Linux] Use DMABUF_SURFACE_EXPORT feature to control dmabuf surface export from EGLImage and disable that on Mesa/Intel and Mesa/AMD r?jgilbert

Comment 29

[Francois Guerraz]

Also affects datadog dashboards on intel linux.
See this public dashboard for example: https://viarezo.fr/en/public_dashboard

Comment 30

[Donal Meehan [:dmeehan]]

:stansky just a reminder that Monday, June 27th is merge day for 103 to beta. If this will land before then?

Comment 31

[Martin Stránský [:stransky] (ni? me)]

We need review first. [:jgilbert]?

Comment 32

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/6dce488087f1
[Linux] Use DMABUF_SURFACE_EXPORT feature to control dmabuf surface export from EGLImage and disable that on Mesa/Intel and Mesa/AMD r=jgilbert

Comment 33

[Iulian Moraru]

https://hg.mozilla.org/mozilla-central/rev/6dce488087f1

Comment 34

[Martin Stránský [:stransky] (ni? me)]

Comment on attachment 9282358 [details]
Bug 1774271 [Linux] Use DMABUF_SURFACE_EXPORT feature to control dmabuf surface export from EGLImage and disable that on Mesa/Intel and Mesa/AMD r?jgilbert

Beta/Release Uplift Approval Request

• User impact if declined: Visual corruption when WebGL is used on Intel/AMD devices.
• Is this code covered by automated tests?: No
• Has the fix been verified in Nightly?: No
• Needs manual test from QE?: Yes
• If yes, steps to reproduce: Run testcase https://bugzilla.mozilla.org/attachment.cgi?id=9281575 in any Intel hardware and make sure the image is not corrupted.
• List of other uplifts needed: None
• Risk to taking this patch: Low
• Why is the change risky/not risky? (and alternatives if risky): Fallback to old and tested WebGL/dmabuf surface creation on Intel/AMD.
• String changes made/needed:
• Is Android affected?: No

Comment 35

[Donal Meehan [:dmeehan]]

Comment on attachment 9282358 [details]
Bug 1774271 [Linux] Use DMABUF_SURFACE_EXPORT feature to control dmabuf surface export from EGLImage and disable that on Mesa/Intel and Mesa/AMD r?jgilbert

Approved for 103 beta 2, thanks.

Comment 36

[Donal Meehan [:dmeehan]]

https://hg.mozilla.org/releases/mozilla-beta/rev/4474f009aefd

Comment 37

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1735929

Comment 38

[Monica Chiorean]

I tried to reproduce the issue on Ubuntu20.4 (wayland) having GPU: Intel(R) HD Graphics 530 and using build without the fix 103.0a1 (20220614213729) and the steps from description, but all I got was a bold drawing on both 'Simpler test-case' attached and on other pdf. Are there any other steps that I can try?
If not can you please confirm issue is not reproducing on latest Beta (https://archive.mozilla.org/pub/firefox/candidates/103.0b2-candidates/)? Thank you so much.

Comment 39

[Marco Castelluccio [:marco]]

I can't reproduce in Beta.

Comment 40

[Monica Chiorean]

Marked as verified based on comment #39.