293942 - Memory Leak introduced May 10th.

Status: RESOLVED INVALID

(Core Graveyard :: GFX: Win32, defect)

Description

[Brian Polidoro]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8b2) Gecko/20050509 Firefox/1.0+
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8b2) Gecko/20050510 Firefox/1.0+

I noticed that memory usage is elevated in recent nightlies.  In previous
nightlies I would be in the 30-40 mbs range.  Now after doing my usual morning
browsing where I browse some in 12 tabs, memory usage is as high as 60mbs.  The
leaks seems to occur the more you browse in multiple tabs.  I went back and
found the May 10th build was the first to exhibit this behavior.  

Reproducible: Always

Steps to Reproduce:
1. Open multiple tabs and browse.  Note memory usage growing.
2.
3.

Actual Results:  
Memory usage is significantly above what it was previously.


I can't determine which platforms other than windows are affected by this.  The
checkins for that day from Peter(6)'s build thread were:

   1. bug 248827 [Core]-Support HTTP/1.1 408 response code [was: 408 request
timeout on a used, persistent, keep-alive connection is mistakenly used as the
response on a subsequent request] [All]
   2. bug 273200 [Firefox]-Ctrl+W in Find bar breaks middle-clicking links, find
toolbar, and back button [All]
   3. bug 292051 [Core]-Shorten Image DDB's life [Win]
   4. bug 293135 [Core: History:Session]-regression: form content is lost when
pressing back (bfacache on&off)[Win]
   5. bug 293161 [Toolkit]--jsconsole command line flag is broken on the trunk [All]
   6. bug 293225 [Core]-Default height of CANVAS should be 150 [Win]
   7. bug 293306 [Core]-canvas leaks lots of memory [Win]
   8. bug 293407 [Core]-Canvas frame uses wrong frame type [All]
   9. bug 293543 [Core]-RenderOffscreen creates an overlarge drawing surface [All]

I'm filing this against Gfx:win32 because my best guess is that this is from bug
292051.

Comment 1

[Worcester12345]

Should add mlk key word.

Comment 2

[ArronM (:paper)]

I'd rule out the canvas ones, #6, #7, #8

I'm leaning towards #3, which is my patch.  If it is, it may not be a leak per
se, just that image bits are being transfered from video memory to main memory.
 In other words, we always used that much memory, but the number was just hidden
before.

However, I don't have knowledge of the other bugs listed, so I can't rule them out.

Comment 3

[Brian Polidoro]

Marking Invalid per comment #2.  It seems to fit the bill as memory usage was
higher but didn't seem to continue to grow.