1492780 - 100% cpu usage after pausing debugger

Status: RESOLVED DUPLICATE of bug 1513737

(DevTools :: Debugger, defect)

Description

[yingted]

Attachment: stacks of the stuck threads

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:63.0) Gecko/20100101 Firefox/63.0
Build ID: 20180826100553

Steps to reproduce:

1. Open a page with JS, like https://material.angularjs.org/latest/demo/input
2. Pause the debugger
3. Trigger some JS


Actual results:

CPU rises to 100% on one core. Doesn't go down after unpausing or closing the tab.
Repeating this causes another thread to get stuck.

I've attached a couple of stacks. Occasionally, RInstructionResults::frame appears on top of the stack.
maybeIonFrameRecovery is also not returning at all.


Expected results:

CPU usage should stay low.

Comment 1

[Jason Orendorff [:jorendorff]]

Here's what happens for me in Firefox Nightly (66), using these steps on a mac:

1. Open https://material.angularjs.org/latest/demo/input
2. Press Cmd+Alt+S to open the Script Debugger
3. Click the pause button. (The UI is not responsive enough; I can't tell that the button knows it's been clicked on.)
4. Mouse over the web page (which is apparently enough to trigger some content JS).

Observed: The Mac beachball for about 9 seconds; then the debugger is paused on a plausible line of code. (Stepping after this is weird, probably due to crummy source maps.)

Profile says: pause points processing calling Array.prototype.shift on a big array.

    https://perfht.ml/2LgYbvU

We do have a hack that makes .shift() fastish as long as your array is small, but treating an Array as a queue is still O(n^2) in the limit.

ni?jimb who was asking about .shift() performance just yesterday. Coincidence?

Comment 2

[Jason Laster [:jlast]]

yep - we're fixing this :)

Comment 4

[yingted]

Hi, I just checked and it looks like this was fixed already here:
https://bugzilla.mozilla.org/show_bug.cgi?id=1510684
https://hg.mozilla.org/integration/mozilla-inbound/rev/39441d1aaa16