1945382 - Testcase generating N "nested" stylesheets is 117x slower in Nightly for N=10000.

Status: RESOLVED DUPLICATE of bug 1931789

(Core :: CSS Parsing and Computation, enhancement)

Description

[Mayank Bansal]

Attachment: nested stylesheet_no BLOB.HTML

Two testcases. 10K input value.

No-Blob Testcase:

• Nightly: https://share.firefox.dev/4aPuFsT (53s)
• Chrome: https://share.firefox.dev/4jJ9sVu (450ms)

Blob Testcase:

• Nightly: https://share.firefox.dev/3EfUuWM (85s)
• Chrome: https://share.firefox.dev/4aZstiN (13s)

Chrome also throws an error on loading the testcases
Uncaught ReferenceError: attachShadow is not defined at <anonymous>:1:1

Testcase created by chatgpt based on my prompts. Definitely not realistic. As usual, feel free to WONTFIX/INVALID.
Also, I am aware that the bug title is clickbait :P

Comment 1

[Mayank Bansal]

Attachment: nested stylesheet - blob.HTML

Comment 2

[Mayank Bansal]

(In reply to Mayank Bansal from comment #0)

Created attachment 9463371 [details]
nested stylesheet_no BLOB.HTML

Two testcases.

Open the "no-blob" testcase
Input 10k and press enter

Nightly: https://share.firefox.dev/4aPuFsT (53s)
Chrome: https://share.firefox.dev/4jJ9sVu (450ms)

Chrome also throws an error on loading the testcase
Uncaught ReferenceError: attachShadow is not defined at <anonymous>:1:1

Testcase created by chatgpt based on my prompts. Definitely not realistic. As usual, feel free to WONTFIX/INVALID.

ni? for this.

Comment 3

[Emilio Cobos Álvarez [:emilio]]

I don't get the "nested" bit, this is just appending N stylesheets...

But, this is basically bug 1931789, aiui.

Comment 4

[Mayank Bansal]

Profile with latest Nightly containing the fix from bug 1931789

no-blob testcase

• https://share.firefox.dev/3Qxyk5h (170ms)
• We improved by 311x and are now 2.6x faster than Chrome!

"nested stylesheet" testcase run locally:

• https://share.firefox.dev/41u2RXR (9.3s)
• We are now 9x faster than before and 1.4x faster than Chrome.
• But now there is that weird blue (URI?) stuff in the content-process thread and some time in parent-process. Worth having a look?

"nested stylesheet" testcase run from BMO:

• https://share.firefox.dev/3XctAFW (43s)
• Weird URI stuf in parent-process and in content-process. Worth having a look?

Comment 5

[Emilio Cobos Álvarez [:emilio]]

But now there is that weird blue (URI?) stuff in the content-process thread and some time in parent-process. Worth having a look?

The URI freeing seems not too problematic... The seems the same as the slow IO operations on Windows from the other bug (ugh). On my phone now but if you can't file it please ni? me and I can.

Weird URI stuf in parent-process and in content-process. Worth having a look?

You mean the CSPReportSender stuff? That's because BMO has CSP reports enabled which logs a lot of messages to the console. If that's what you mean I don't think it's worth digging too much into it.