nsMetaCharsetObserver takes 2.2 secs out of 19 secs of startup

VERIFIED FIXED in M15

Status

()

defect
P3
normal
VERIFIED FIXED
20 years ago
20 years ago

People

(Reporter: dp, Assigned: ftang)

Tracking

({perf})

Trunk
x86
Windows NT
Points:
---

Firefox Tracking Flags

(Not tracked)

Details

(Whiteboard: we have a fix)

Reporter

Description

20 years ago
Function					%	#call	millisecs
---------------------------------------------------------------------------
nsMetaCharsetObserver::nsMetaCharsetObserver	100.00	1	2239.61
Reporter

Updated

20 years ago
Blocks: 7251
Reporter

Comment 1

20 years ago
Most of time time seems to be two dll loads : chardet and uconv. So this may be 
ok.
Reporter

Comment 2

20 years ago
If I remove dll loading time, here is the pure function timing:
nsMetaCharsetObserver::nsMetaCharsetObserver	100.00	1	1642.74

The timing is dominated by this call:
nsCOMPtr<nsICharsetAlias>::nsCOMPtr<nsICharsetAlias>	100.00	1	1642.73

So what are we doing with CharsetAlias that is costing us 1.6 secs that isn't 
dll load timeing.
Assignee

Updated

20 years ago
Status: NEW → ASSIGNED
Keywords: beta1, perf
Target Milestone: M15
Assignee

Comment 3

20 years ago
It seems that we should lazy load the nsICharsetAlias . Let me work on a patch 
and send to you. 
Add perf for performacne reason. Add beta1 since this is to improve startup 
performance.
Assignee

Comment 4

20 years ago
I put the patch in 
http://warp/u/ftang/tmp/metalazy.txt

Please try it. The patch delay the load of the nsICharsetAlias. This won't save 
the total time needed, but it will delay the loading till when we really need 
it.
Assignee

Comment 5

20 years ago
reassign back to dp- dp, after you try, reassign it back to me. Thanks.
Assignee: ftang → dp
Status: ASSIGNED → NEW

Comment 6

20 years ago
Putting on PDT+ radar for beta1.  
Whiteboard: [PDT+]
Reporter

Comment 7

20 years ago
Looks like this doesnt defer the timing far enough to see a performance win for 
startup. So I dont think we should do this for beta1.

Voting for PDT- unless ftang has a way to moving this off past startup.
Assignee: dp → ftang
Keywords: beta1

Comment 8

20 years ago
Clearing PDT+ to get re-eval to minus per test and suggestion of DP
Whiteboard: [PDT+]
Reporter

Comment 9

20 years ago
Ok frank and I found a way to eliminate the 1.6 secs by delaying the loading
of the charset aliases properties files far enough.

Putting this back into PDT+
Keywords: beta1
Whiteboard: we have a fix
Assignee

Comment 10

20 years ago
fix and check in 
Status: NEW → RESOLVED
Closed: 20 years ago
Resolution: --- → FIXED

Updated

20 years ago
Status: RESOLVED → VERIFIED

Comment 11

20 years ago
verified.

Updated

19 years ago
No longer blocks: 7251
You need to log in before you can comment on or make changes to this bug.