88075 - Cancelling iBench test mid-way leaks nsSocketTransports

Status: RESOLVED WORKSFORME

(Core :: Networking, defect, P3)

Description

[Simon Fraser [no longer active]]

If I cancel the iBench test in mid-flow, it sometimes (~50% of the time) leaves 
an nsSocketTransport around, which means that the mSelectFDSetCount never drops 
back to its base level (1 if using pollable events).

This causes the Mac build to think that there is still network activity going on, 
and thus permanently hog the CPU.

Comment 1

[Simon Fraser [no longer active]]

Attachment: nsSocketTransport log

Comment 2

[Doug Turner (:dougt)]

simon, are you still seeing this?  Rick checked in a fix to the URILoader which
prevented OnStopRequest from firing on a cancelled request.

Comment 3

[rpotts (gone)]

No.. my patch was just the opposite.  The URILoader now *fires* OnStopRequest 
even if the load was cancelled...

(i'm sure this is what you meant - doug, but I just wanted to be clear :-)

Comment 4

[Doug Turner (:dougt)]

right, right.  sorry about that..  I am having my house painted right now, and
the fumes man, the fumes are killing me.

Comment 5

[Darin Fisher]

sfraser: do you know if this is still a problem?

Comment 6

[Simon Fraser [no longer active]]

I have no idea.

Comment 7

[Darin Fisher]

-> future

Comment 8

[Frankie]

No new reports of this behavior in the past 18 months.