1177293 - Costs still wrong with flattened recursion toggled on

Status: RESOLVED FIXED

(DevTools :: Performance Tools (Profiler/Timeline), defect)

Description

[Shu-yu Guo [:shu]]

We overcount frames when flattening recursion.

Comment 1

[Shu-yu Guo [:shu]]

Attachment: Don't overcount samples when flattening recursion.

Comment 2

[Jordan Santell [:jsantell] [@jsantell]]

To deal with recursion, this would work, but wouldn't affect the use-case of any kind of repetitive calls in the stack, like A()->B()->C()->B()->D() -- B would be counted twice. I think I have a way to fix both adjacent repeated calls, and otherwise.

Comment 3

[Shu-yu Guo [:shu]]

Oops, I forgot to land this.

(In reply to Jordan Santell [:jsantell] [@jsantell] from comment #2)
> To deal with recursion, this would work, but wouldn't affect the use-case of
> any kind of repetitive calls in the stack, like A()->B()->C()->B()->D() -- B
> would be counted twice. I think I have a way to fix both adjacent repeated
> calls, and otherwise.

Those two B frames in your example should be counted twice. Remember that the sample count is tied to each row in the call tree, not to some function. In your example, B would have two different rows, and should be counted twice. 

The bug here was that recursion flattening flattens multiple rows into a single row, and we were overcounting for the single flattened row.

Comment 4

[Pulsebot]

https://hg.mozilla.org/integration/fx-team/rev/c8066db6f2e9

Comment 5

[Jordan Santell [:jsantell] [@jsantell]]

tests added to bug 1167021

Comment 6

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/c8066db6f2e9