444034 - jprof fails to sort symbols correctly

Status: RESOLVED FIXED

(Core :: General, defect)

Description

[Joshua Cranmer [:jcranmer]]

While trying to do some profiling of code, I found that a startling 97.7% of time was spent in the function name "Flist." A little bit of sleuthing showed that to be the last symbol (it's not even a function) defined in the executable. Given that the profile report shows no calls in JS despite knowledge that heavy JS usage is present (as well as other relevant JS called), as well as similar statistics showing no XPCOM or NSPR or NSS code, all of which are dynamically linked.

Looking at the code tells me that it fails to resolve these, especially with this dynamic run: b7f196b7 8be585c. The former is the address, the latter the highest known address.

Comment 1

[Joshua Cranmer [:jcranmer]]

More fun to the puzzle:
The symbols are loaded (and so are non-function string names, but that's an easy if statement), but the array is not sorted properly:

it goes b15a5000...b7fe9654, 08048114..08be585c

Comment 2

[Joshua Cranmer [:jcranmer]]

And, in short, the failing code was the sort code. This code works:

static int symbolOrder(void const* a, void const* b)
{
  Symbol const* ap = (Symbol const *)a;
  Symbol const* bp = (Symbol const *)b;
  return ap->address == bp->address ? 0 :
    (ap->address > bp->address ? 1 : -1);
}

Comment 3

[Joshua Cranmer [:jcranmer]]

Changing the name to be representative of the error which causes this as opposed to the error which is the effect.

Comment 4

[Joshua Cranmer [:jcranmer]]

Attachment: fix the problem

A little tardier than I said, but here's a patch nonetheless.

Comment 5

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 340023 [details] [diff] [review]
fix the problem

r=dbaron

Comment 6

[Joshua Cranmer [:jcranmer]]

Checked in as changeset 3bfd66c81a78.