934770 - divide large text runs into smaller runs to avoid chrome hangs

Status: RESOLVED INCOMPLETE

(Core :: Layout: Text and Fonts, defect, P3)

Description

[John Daggett (:jtd)]

Tinderbox logfile with text perf logging enabled:

(presshell) 122409000 begin load [https://tbpl.mozilla.org/php/getParsedLog.php?id=30024572&tree=Mozilla-Central&full=1]
(textperf-reflow) 0x122409000 time-ms:    165.29 reflow: 1 content-textruns: 7 chrome-textruns: 0 chars: 218490 max-textrun-len: 218334 word-cache-lookups: 18254 word-cache-hit-ratio: 0.833 word-cache-space: 0 word-cache-long: 1448 pref-fallbacks: 0 system-fallbacks: 0
(textperf-reflow) 0x122409000 time-ms:    478.03 reflow: 2 content-textruns: 190 chrome-textruns: 0 chars: 547790 max-textrun-len: 539400 word-cache-lookups: 48272 word-cache-hit-ratio: 0.945 word-cache-space: 0 word-cache-long: 3735 pref-fallbacks: 0 system-fallbacks: 0
(textperf-reflow) 0x122409000 time-ms:   1003.27 reflow: 5 content-textruns: 3 chrome-textruns: 0 chars: 1004331 max-textrun-len: 1004058 word-cache-lookups: 87048 word-cache-hit-ratio: 0.920 word-cache-space: 0 word-cache-long: 7676 pref-fallbacks: 0 system-fallbacks: 0
(textperf-reflow) 0x122409000 time-ms:   3063.15 reflow: 8 content-textruns: 4157 chrome-textruns: 0 chars: 3476958 max-textrun-len: 708855 word-cache-lookups: 319828 word-cache-hit-ratio: 0.927 word-cache-space: 1724 word-cache-long: 22847 pref-fallbacks: 3448 system-fallbacks: 0
(textperf-reflow) 0x122409000 time-ms:  33517.60 reflow: 10 content-textruns: 74 chrome-textruns: 0 chars: 32171697 max-textrun-len: 20274371 word-cache-lookups: 2974573 word-cache-hit-ratio: 0.936 word-cache-space: 25 word-cache-long: 204409 pref-fallbacks: 50 system-fallbacks: 7
(presshell) 122409000 end load time-ms:  39899.49 [https://tbpl.mozilla.org/php/getParsedLog.php?id=30024572&tree=Mozilla-Central&full=1]
(textperf-loaddone) 0x122409000 reflow: 12 content-textruns: 4431 chrome-textruns: 0 chars: 37419266 max-textrun-len: 20274371 word-cache-lookups: 3447975 word-cache-hit-ratio: 0.934 word-cache-space: 1749 word-cache-long: 240115 pref-fallbacks: 3498 system-fallbacks: 7
(textperf) 0x122409000 reflow: 12 content-textruns: 4431 chrome-textruns: 0 chars: 37419266 max-textrun-len: 20274371 word-cache-lookups: 3447975 word-cache-hit-ratio: 0.934 word-cache-space: 1749 word-cache-long: 240115 pref-fallbacks: 3498 system-fallbacks: 7

The key item here is that one of the reflows is shaping a textrun 20,274,371 bytes long!  This will definitely cause a chrome hang.  It turns out pre elements and raw text files are both handled this way.

We need to slice up text runs when they get too large and we need to do it in a way that can be cached efficiently to avoid reconstructing the same textrun multiple times.

Comment 1

[Boris Zbarsky [:bzbarsky]]

This sounds a lot like bug 386122.  I guess there we made scrolling faster but the initial pageload was still allocating the giant textrun?

Comment 2

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Yes. I don't know why we're creating a giant textrun here, unless the log contains one incredibly long line somewhere.

Comment 3

[John Daggett (:jtd)]

Simple testcase for generating large text elements:

generate 1000 character block
http://people.mozilla.org/~jdaggett/tests/generatetext.html?size=1000

generate 10mb character block
http://people.mozilla.org/~jdaggett/tests/generatetext.html?size=10mb

Comment 5

[Chris AtLee [:catlee]]

The log from bug 1228469 has one line that is 20k long. 1% of the lines are > 4k long.

Comment 7

[Peter Magyari (Desktop QA)]

Hi John,

Sorry for the old poke, I'm aware that this is an old report, but I'd like to close it if the issue no longer occurs.

Have you by any chance seen this issue on the latest Firefox versions?

Thank you!

Comment 8

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

Redirect a needinfo that is pending on an inactive user to the triage owner.
:jfkthame, since the bug has high severity and recent activity, could you have a look please?

For more information, please visit auto_nag documentation.

Comment 9

[Jonathan Kew [:jfkthame]]

I think we should close this as incomplete at this point. The specific log files referenced here and in bug 1228469, for example, are long gone, so direct comparisons to current behavior are difficult.

It's definitely still true that loading a huge log file can have noticeable delays; e.g. https://share.firefox.dev/3MoAQGD is a profile of loading https://firefoxci.taskcluster-artifacts.net/EOpL-WzTRC2Duw9EZJ2vqQ/0/public/logs/live_backing.log in Nightly on macOS, and shows a 5.5-second reflow. However, e10s means that this doesn't hang the entire browser UI (just the tab).

If we want to look into this further I think it would be best to open a new bug with suitable test files (as attachments, so they don't go away!) and profiles etc.