34283 - Website causes double showing of Mozilla

Status: VERIFIED FIXED

(Core :: Layout, defect, P3)

Description

[Henrik Gemal]

If you go to:
http://cgl.microsoft.com/clipgallerylive/cgl30/eula.asp?nInterface=0

you will be able to see the double showing of Mozilla

Comment 1

[Kevin Puetz]

page appears fine to me, and your description is pretty vague. marking
worksforme, if you can still see this problem in a recent nightly, please add a
better description of it.

Comment 2

[jay garcia]

Win98SE - Build 40215 - First access of URL, page begins to load and a brief 
second instance of the browser will appear in the lower right quadrant of the 
screen. Only shows for a split second (flash).

Comment 3

[Kevin Puetz]

thank you for a better description, reopening and changing OS flags to what you
say you are on. Hopefully someone who has a win98 box (I don't) will be able to
figure this out for you.

Comment 4

[troy]

I can't reproduce this using viewer. The description "second instance of the 
browser will appear in the lower right quadrant of the screen" sounds like a 
mozilla.exe problem so reassigning to XP Apps

Comment 5

[Paul MacQuiddy]

john, you love these types of bugs...

Comment 6

[John Morrison]

The 'flash' of the browser into the frames of the frameset is, essentially, 
until the document loads into the frame, the contents of the frame were picking 
up random buffer contents.

However, I think this was fixed with the docShell code that landed last night
(but the network is too flaky for me right now to be certain. BBL).

Comment 7

[verah (gone)]

Reassigning as per Don

Comment 8

[John Morrison]

This worksforme mac/linux/win95 2000051108 builds. The chrome is not
painted into the lower frame in the pause before the content loads into 
the frame.

Comment 9

[Henrik Gemal]

Verified with build 2000051120

Comment 10

[Henrik Gemal]

Here again in build 2000052420. To reproduce:
- go to http://web.mail.dk/
- let the page load finished
- place your cursor in Location field of the browser and press Enter.

It only happens after pressing Enter in the Location field. Not by pressing 
Reload. Reload seems to clear the page to white. Pressing Enter doesn't seem to 
have the same effect.

Comment 11

[karnaze (gone)]

Triage duty calls. Reassinging to av.

Comment 12

[av (gone)]

I see this on NT. Looks like during reflow frame paints itself with some not 
discarded garbage. Nisheeth, does it lookk like reflow problem to you?

Comment 13

[Nisheeth Ranjan]

Kevin, this seems like a rendering issue.  When we load urls by clicking on a 
link, things work fine.  When we load a url by hitting enter in the url bar, a 
piece of the mozilla UI appear briefly on the page before the new page loads.
It seems like the root view is not getting cleared for the latter case.

Comment 14

[Kevin McCluskey (gone)]

The problem is that the entire content area is invalidated before a document is 
loaded. The paint message comes through and the frames are asked to paint, but 
since there isn't a document there isn't a frame to paint the background of the 
document. The view manager assumes everything has been painted and copies the 
back buffer forward. Since the background hasn't been rendered the previous 
content of the back buffer is copied forward. The solution is prevent the 
docshell or and layout from invalidating until there is a frame which can render 
into the area.

Comment 15

[Kevin McCluskey (gone)]

Nominating for nsbeta3. This bug shows up on a large number of pages. 

Comment 16

[Kevin McCluskey (gone)]

Adding polish keyword

Comment 17

[Kevin McCluskey (gone)]

Marking nsbeta3-

Comment 18

[Kevin McCluskey (gone)]

The double rendering can be fixed by having the HTMLFrameInnerFrame paint the 
damaged area in all cases instead of painting only when a sub-shell has not been 
created. In addition, the FillRect call should be changed to pass the damaged 
rect instead of the entire bounds for the HTMLFrameInnerFrame.

in nsFrameFrame.cpp change the Paint method from:

NS_IMETHODIMP
nsHTMLFrameInnerFrame::Paint(nsIPresContext*      aPresContext,
                             nsIRenderingContext& aRenderingContext,
                             const nsRect&        aDirtyRect,
                             nsFramePaintLayer    aWhichLayer)
{
  //printf("inner paint %X (%d,%d,%d,%d) \n", this, aDirtyRect.x, aDirtyRect.y, 
aDirtyRect.width, aDirtyRect.height);
  // if there is not web shell paint based on our background color, 
  // otherwise let the web shell paint the sub document 
  if (!mSubShell) {
    const nsStyleColor* color =
      (const nsStyleColor*)mStyleContext->GetStyleData(eStyleStruct_Color);
    aRenderingContext.SetColor(color->mBackgroundColor);
    aRenderingContext.FillRect(mRect);
  }
  return NS_OK;
}

to:

NS_IMETHODIMP
nsHTMLFrameInnerFrame::Paint(nsIPresContext*      aPresContext,
                             nsIRenderingContext& aRenderingContext,
                             const nsRect&        aDirtyRect,
                             nsFramePaintLayer    aWhichLayer)
{
  //printf("inner paint %X (%d,%d,%d,%d) \n", this, aDirtyRect.x, aDirtyRect.y, 
aDirtyRect.width, aDirtyRect.height);
  
    const nsStyleColor* color =
      (const nsStyleColor*)mStyleContext->GetStyleData(eStyleStruct_Color);
    aRenderingContext.SetColor(color->mBackgroundColor);
    aRenderingContext.FillRect(aDirtyRect,);
 
  return NS_OK;
}

This fixes the problem, but I'm concerned about performance. This paint method 
gets called alot when loading pages within Mozilla. 34+ times. 



This bug has been marked "future" because the original netscape engineer working 
on this is over-burdened. If you feel this is an error, that you or another 
known resource will be working on this bug,or if it blocks your work in some way 
-- please attach your concern to the bug for reconsideration. 

Marked bug future based on nsbeta3- designation.

Comment 19

[Kevin McCluskey (gone)]

*** Bug 48854 has been marked as a duplicate of this bug. ***

Comment 20

[Marc Attinasi]

*** Bug 42622 has been marked as a duplicate of this bug. ***

Comment 21

[Kevin McCluskey (gone)]

Attachment: Patch which clears the backbuffer to white before rendering to it

Comment 22

[Kevin McCluskey (gone)]

*** Bug 46669 has been marked as a duplicate of this bug. ***

Comment 23

[Randell Jesup [:jesup] (needinfo me)]

bug 46669 was marked as a dup of this.  However, I'm not 100% certain that it is
- 46669 doesn't appear to fill with garbage, it fills with the content of the
current window.  Now, this may be just another aspect of the same bug, but it
seems like it might be something else (like cloning an object when a new frame
is created).

I would say I'm 90% certain however.  I think 46669 should be retested with the
patch installed or when the problem is fixed before this bug is marked as
closed.

Comment 24

[Randell Jesup [:jesup] (needinfo me)]

Bug 48480 may be a dup of this one

Comment 25

[Kevin McCluskey (gone)]

Marking nsbeta3+. This bug shows up a numerous pages when using a slow 
connection.  The fix is small localized and low risk.

Comment 26

[Kevin McCluskey (gone)]

Fixed in 8/21/2000 9:23PM build.

Comment 27

[Chris Petersen]

Not seeing this problem in the latest build (August 23rd) on Mac or Windows. 
Reporter, check with the latest version to also verify.

Comment 28

[Henrik Gemal]

Verified with build 2000082313

Comment 29

[Randell Jesup [:jesup] (needinfo me)]

20010208xx pull/build

I still see this bug at web.mail.dk, as well as most any framed site (though the
affect can be hard to see on a fast machine/link/display).

See comments in n.p.m.performance

It's also easy to recreate by going to http://www.nerve.com/photography/, going
to mozilla.org, then hitting back.  watch the bottom frame in particular.

OS->All since I see this on FreeBSD.  It's possible this is an X/etc specific
instance.

Adding perf since this extra rendering slows us down, especially on a remote X
display.

Very visually unappealing if you have a site with a bunch of frames.  VERY.

nominating for mozilla 0.9

See also the comments in bug 46669.

Comment 30

[Henrik Gemal]

If you turn the Sidebar on, I also sometimes see parts of the sidebar in the 
places where the frames are loaded!

Build 20010209 on Win2k

Comment 31

[Kevin McCluskey (gone)]

This has been fixed.
marking fixed.

Comment 32

[Henrik Gemal]

seems ok to me too