1539366 - Typing in Phabricator textarea on long pages is super slow

Status: RESOLVED FIXED

(Core :: Web Painting, defect, P2)

Description

[Nicolas Chevobbe [:nchevobbe]]

Steps to reproduce

1. Go to https://phabricator.services.mozilla.com/D24930
2. Scroll all the way to the bottom to access the text box
3. Focus the input and type a letter

Results

It takes more than 10s for the typed letter to appear in the text box.
Here's a profile of typing a single letter in there: http://bit.ly/2HTsn0p

---

Filing on Graphics as it looks like this is the culprit here.

Comment 1

[Jonathan Kew [:jfkthame]]

Seems to be more of a DisplayList-building issue around painting backgrounds, with a huge amount of time going into nsBlockInFlowLineIterator::nsBlockInFlowLineIterator. Moving to Web Painting.

(FWIW, just the initial rendering and scrolling to the bottom of that page takes "forever" (almost) here, this isn't specifically about typing.)

Comment 2

[Matt Woodrow (:mattwoodrow)]

Attachment: Bug 1539366 - Use SetCoord to restore mPIStartBorderData's coord so that it also set mValid. r?mats

Without this the restore doesn't have any effect as the next frame sees that coord has mIsValid=false

Comment 3

[Matt Woodrow (:mattwoodrow)]

Attachment: Bug 1539366 - Use a temporary object to compute the mPIStartBorderData coords so that we don't clobber the regular state. r?mats

Depends on D26648

Comment 4

[Matt Woodrow (:mattwoodrow)]

Attachment: Bug 1539366 - Avoid calling AreOnSameLine once we know we've already changed lines since it can be expensive to compute. r?mats

Depends on D26649

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/e8b188fd600d
Use SetCoord to restore mPIStartBorderData's coord so that it also set mValid. r=mats
https://hg.mozilla.org/integration/autoland/rev/d32769ee5991
Use a temporary object to compute the mPIStartBorderData coords so that we don't clobber the regular state. r=mats
https://hg.mozilla.org/integration/autoland/rev/9d54e34f7f66
Avoid calling AreOnSameLine once we know we've already changed lines since it can be expensive to compute. r=mats

Comment 6

[Dorel Luca [:dluca]]

https://hg.mozilla.org/mozilla-central/rev/e8b188fd600d
https://hg.mozilla.org/mozilla-central/rev/d32769ee5991
https://hg.mozilla.org/mozilla-central/rev/9d54e34f7f66

Comment 7

[Julien Cristau [:jcristau]]

Is this worth an uplift to beta (and safe enough)?

Comment 8

[Matt Woodrow (:mattwoodrow)]

Comment on attachment 9056790 [details]
Bug 1539366 - Avoid calling AreOnSameLine once we know we've already changed lines since it can be expensive to compute. r?mats

Beta/Release Uplift Approval Request

• User impact if declined: Slow display list building on some inline heavy pages, include Phabricator.
• Is this code covered by automated tests?: Yes
• Has the fix been verified in Nightly?: No
• Needs manual test from QE?: No
• If yes, steps to reproduce:
• List of other uplifts needed: None
• Risk to taking this patch: Low
• Why is the change risky/not risky? (and alternatives if risky): Fairly simple performance optimization, just fixes cases where the existing cache wasn't being hit.
• String changes made/needed: None

Comment 9

[Julien Cristau [:jcristau]]

Comment on attachment 9056788 [details]
Bug 1539366 - Use SetCoord to restore mPIStartBorderData's coord so that it also set mValid. r?mats

painting perf fixes, approved for 68.0b8

Comment 10

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

https://hg.mozilla.org/releases/mozilla-beta/rev/d2388f832a7d
https://hg.mozilla.org/releases/mozilla-beta/rev/43f54f4b1e2a
https://hg.mozilla.org/releases/mozilla-beta/rev/17b33cebec20