1545034 - Significant increase of loading times on CodiMD caused with bug 1504947

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Dennis Schubert [:denschub]]

The latest patch (8eaa269b362f) landed in bug 1504947 8 days ago apparently causes some significant increases of loading time on CodiMD.

If you try to load this pad in a Nightly after landing that patch, you'll notice a significant delay (multiple seconds) between the header showing up and the pad actually loading. Before landing that patch, the loading starts almost instantaneously.

Here are profiles from loading said pad to compare:

• Last good revision (e9408a26af674fbf535d853ede4ab629ac652c05): https://profiler.firefox.com/public/e088529ac0ccce71b5b0d30c62a567ba1f7363c3
• First bad revision (8eaa269b362fd429c04779eaff8e2e7cd0075ad7): https://profiler.firefox.com/public/81accdbe2227eeb0d81dbf86b0c93769c32c6641

Comment 1

[Dennis Schubert [:denschub]]

Jeff, as discussed on IRC, I'd appreciate if you could take a look at this. :)

Comment 2

[Jan de Mooij [:jandem]]

[Tracking Requested - why for this release]:
Perf regression.

Comment 3

[Jeff Walden [:Waldo]]

Unsurprisingly inverting that first-bad profile shows up TokenStreamSpecific::columnAt as the culprit. The profile identifies the script that was being parsed at the time, but loading the URL manually hits a 404 -- maybe it was a temporary one-off thing. Still looking to see if I can pull it up to identify the characteristics of the source text that mean the current column-caching approach doesn't work enough...

Comment 4

[Dennis Schubert [:denschub]]

maybe it was a temporary one-off thing

No. Loading https://demo.codimd.org/z4lH4LmqQIqTqr9Vg4b0-Q still results in the same behavior. The hashes in the JS URLs did change (it's now https://demo.codimd.org/build/index-pack.cfd5a8ec26563ff8d00c.js), but that's probably just because they deployed a new version.

Comment 5

[Jeff Walden [:Waldo]]

Sorry, I meant that URL might have been a one-off URL only accessible on a per-visitor basis. Thanks for the new packed URL, it's exactly what I was looking for.

Comment 6

[Jeff Walden [:Waldo]]

Attachment: Problematic script wrt columns, in case the site changes further

Comment 7

[Jeff Walden [:Waldo]]

Fixed by the reversion in bug 1549758.

Comment 8

[Jeff Walden [:Waldo]]

(And incidentally, bug 1551916's patchwork handle parsing the problematic script here just fine, without observable delay, when we get the column number semantics changed back to code points again.)