19170 - Need to yield to necko thread more often

Status: VERIFIED FIXED

(Core :: Networking, defect, P3)

Description

[Simon Fraser [no longer active]]

There are performance problems now (e.g. image rendering 18738) which result from
us not yielding to the necko thread often enough. We need to yield every time
through the main event loop.

Comment 1

[Simon Fraser [no longer active]]

Patch for this (line numbers omitted, because I have other changes)

  		DoIdle(*anEvent);
  		if (mRunning)
  			Repeater::DoIdlers(*anEvent);
+
+ 		// yield to other threads
+ 		PR_Sleep(PR_INTERVAL_NO_WAIT);

Comment 2

[Simon Fraser [no longer active]]

Fix checked in.

Comment 3

[Simon Fraser [no longer active]]

This change should make rendering of large image snappier. It should also help
other stuff that runs on threads, like IMAP.

Comment 4

[(not reading, please use [e-mail] instead)]

this smells like something that may be hurting unix too.

adding mscott, pavlov and alecf to the cc list.

anyone have ideas how to do this on unix, too?

the fix was to

Comment 5

[(not reading, please use [e-mail] instead)]

add me to the cc list too.

Comment 6

[(not reading, please use [e-mail] instead)]

ok, I'm trying a similar fix for unix now...

cross your fingers, unix lovers!

Index: nsWidget.cpp
===================================================================
RCS file: /cvsroot/mozilla/widget/src/gtk/nsWidget.cpp,v
retrieving revision 1.168
diff -p -r1.168 nsWidget.cpp
*** nsWidget.cpp        1999/11/16 01:10:46     1.168
--- nsWidget.cpp        1999/11/18 06:22:22
*************** NS_IMETHODIMP nsWidget::DispatchEvent(ns
*** 1324,1329 ****
--- 1324,1332 ----
    }
    NS_RELEASE(aEvent->widget);

+   // yield to other threads
+   PR_Sleep(PR_INTERVAL_NO_WAIT);
+
    return NS_OK;
  }

Comment 7

[leger]

Setting QA contact for verification.

Comment 8

[leger]

Bulk move of all Necko (to be deleted component) bugs to new Networking

component.

Comment 9

[rubydoo123]

setting milestone and marking verified