588256 - e10s HTTP: clean up more aggressively in DocumentChannelCleanup

Status: RESOLVED FIXED

(Core :: Networking: HTTP, defect)

Description

[Jason Duell]

Attachment: release nsHttpChannel and HttpChannelParentListener after child::OnStopRequest

Followup to 588200. 

For document-loads, release nsHttpChannel and HttpChannelParentListener after child::OnStopRequest called.

So it turns out that nsDocuments live for a while because of the history cache, and they hold onto their HttpChannelChild, so we're keeping PHttpChannel and friends around for quite a while (even after user has gone to a different page).  This patch just optimizes memory usage a bit by deleting everything (chrome, parent listener class) in chrome but the securityInfo, which is all we need to keep after OnStop.

Comment 1

[Jason Duell]

s/\(chrome,/(chrome nsHttpChannel/

Comment 2

[Honza Bambas (:mayhemer)]

Comment on attachment 466877 [details] [diff] [review]
release nsHttpChannel and HttpChannelParentListener after child::OnStopRequest

> HttpChannelParent::OnStopRequest(nsIRequest *aRequest, 

>+  // Not required to break refcount cycle, but speeds up
>+  // HttpChannelParentListener cleanup.
>+  mChannelListener = 0;

We already release it in OnStartRequest, and we actually could do that in HttpChannelParent::RecvRedirect2Result (or newly Verify), up to you.

>+  nsCOMPtr<nsIAssociatedContentSecurity>  mSecurityInfo;

This is not security info, this is associated content security, please rename the member to mAssociatedContentSecurity.  The best would be to have nsCOMPtr<nsISupports> mSecurityInfo and QI where needed, but let's leave the logic as is for now as we need just the associated content security.

r=honzab with those fixed.

Comment 3

[Josh Matthews [:jdm]]

Jason, is this patch still desirable?

Comment 4

[Jason Duell]

Attachment: v2. Fix bitrot, cleaned up a little

This just fixes bitrot and cleans up a little (there's an extra state variable we didn't need--just leverage mKeptAlive).

Comment 5

[Josh Matthews [:jdm]]

Comment on attachment 555960 [details] [diff] [review]
v2. Fix bitrot, cleaned up a little

My only beefs with this patch are cosmetic.

>+  // Normally we Send_delete in OnStopRequest: only do it with refcounting when
>+  // we need to hold onto IPDL channel for security info (in that case, IPDL
>+  // itself hold 1 reference, so we Send_delete when refCnt==1).

Normally we Send_delete in OnStopRequest, but when we need to retain the remote channel for security info IPDL itself holds 1 reference, so we Send_delete when refCnt==1.

>+    // IPDL channel still open, but only for updating security info.  Remote
>+    // necko channel will be deleted after this.

s/after this/when all references are dropped/ or s/after this/later/

>+#include "nsIAssociatedContentSecurity.h"

You should just be able to forward-declare nsIAssociatedContentSecurity and keep this header inclusion in the cpp file.

Comment 6

[Jason Duell]

Thanks--took all your comments, except I'm just using

   // IPDL channel still open, but only for updating security info.

i.e omit mention of when remote channel deleted.

Off to tryserver.

Comment 7

[Jason Duell]

http://hg.mozilla.org/integration/mozilla-inbound/rev/8215d12a0015

Comment 8

[Marco Bonardo [:mak]]

http://hg.mozilla.org/mozilla-central/rev/8215d12a0015