1487465 - LastPass causes 6s hang in content

Status: VERIFIED FIXED

(WebExtensions :: Developer Outreach, defect, P5)

Description

[:Harald Kirschner :digitarald]

STR:
- Install LastPass extension (and log in)
- Open https://dxr.mozilla.org/mozilla-central/search?q=actor%20perf
- Click on a folder (like `/ lib /` in the first result)

ER: Menu opens instantly
AR: Menu opens after 6 second hang

Profile: https://perfht.ml/2oomMEz

High level analysis:

- mouseup event triggers `L</LPSiteNotification` (which seems to be called twice)
- `getText` (which is called by `getUsername`) calls `Symbol.split` many times (see Stack Chart)
- `Symbol.split` does regexp work and substr matching

Comment 1

[Andrew Swan [:aswan]]

This is an issue in the lastpass extension.

Comment 2

[Markus Stange [:mstange]]

Harald, can you check whether this is faster in Chrome? LastPass should presumably be hitting the same code path there. If they have a faster implementation of String.prototype.split on regex, there's a Spidermonkey bug here, otherwise it's really just something for LastPass to fix.

Comment 3

[Markus Stange [:mstange]]

(In reply to Markus Stange [:mstange] from comment #2)
> If they have a faster
> implementation of String.prototype.split on regex, there's a Spidermonkey
> bug here,

Which might be bug 1437914.

Comment 4

[:Harald Kirschner :digitarald]

Attachment: trace_dxr-lastpass.json.gz

Chrome with LastPass no significant delay.

Attached a Chrome trace recorded for LastPass on DXR, using the "rendering & input latency" preset.

Comment 5

[Daniel Holbert [:dholbert]]

(In reply to Markus Stange [:mstange] from comment #3)
> might be bug 1437914.
(In reply to :Harald Kirschner :digitarald from comment #4)
> Chrome with LastPass no significant delay.

(OK -- then this isn't bug 1437914, since that one is equivalently bad between Firefox & Chrome, aside from the fact that Chrome runs the regex less often in the original testcase there.

Comment 6

[Mike Conley (:mconley) (:⚙️) (PTO: May 24 - June 2nd)]

What would be lovely here is to reproduce this, and then use the Add-on Debugging tools to find what the slow regular expression operation is.

Comment 7

[Mike Conley (:mconley) (:⚙️) (PTO: May 24 - June 2nd)]

cc'ing Matthew Miller, since Lockbox might care about this class of bug.

Comment 8

[David Durst [:ddurst]]

(In reply to Daniel Holbert [:dholbert] from comment #5)

(In reply to Markus Stange [:mstange] from comment #3)

might be bug 1437914.
(In reply to :Harald Kirschner :digitarald from comment #4)
Chrome with LastPass no significant delay.

(OK -- then this isn't bug 1437914, since that one is equivalently bad
between Firefox & Chrome, aside from the fact that Chrome runs the regex
less often in the original testcase there.

So.... did we determine that this is a bug in Spidermonkey or just something for LastPass?

Comment 9

[Markus Stange [:mstange]]

The fact that it's slow in Firefox and fast in Chrome suggest that it is a Spidermonkey bug. That's about all we know at the moment.

Comment 10

[David Durst [:ddurst]]

Based on #c9, I'm going to move this to reflect that it's not about the extensions framework; but it's unclear if this is entirely on us or something in LastPass is exacerbating something. I'll contact LastPass about this and bug 1486303.

Comment 11

[David Durst [:ddurst]]

Harald, can we get some more data on this?

• Can you still reproduce?
• Firefox version?
• OS?
• LastPass version?

Comment 12

[:Harald Kirschner :digitarald]

Now its an 800ms hang on a high-end MBP: https://perfht.ml/2SO0RDq

Nightly 67.0a1 (2019-03-08) (64-bit), OSX

LastPass 4.35.0.4

Comment 13

[Mark Nakib]

Hi Harald, I'm an engineer on LastPass at LogMeIn. I think I've fixed the issue, and I'm seeing roughly a 5x improvement in performance on that specific page. Would you please email me at [e-mail] so I can provide you a developer build to test the fix?

Comment 14

[:Harald Kirschner :digitarald]

Thank you Mark! Email sent.

Comment 15

[David Durst [:ddurst]]

Closing this per #c13.

Comment 16

[:Harald Kirschner :digitarald]

Tested Mark's build and saw the click overhead go down from 973ms to 42ms 🎉.