239701 - Some gif and png images appear corrupted, with black and shifted lines

Status: RESOLVED FIXED

(Core Graveyard :: Image: Painting, defect)

Description

[Mike Calmus]

User-Agent:       Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.7b) Gecko/20040316
Build Identifier: Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.7b) Gecko/20040316

Many images are distorted when rendered in browser. For instance, the layers on
http://us.i1.yimg.com/us.yimg.com/i/tv/tv2.gif appear to not be synced properly.
I have tried setting up a new profile and clearing the cache. This same image
works fine in NS 7.1 on the same machine and using Mozilla 1.6 in Windows XP
(using Virtual PC). The display of the image is distorted regardless whether it
is displayed in a page or by itself.

Reproducible: Always
Steps to Reproduce:

Comment 1

[Olivier Cahagne]

seems to work fine here using FF 20040402 on Win2k, can you post a screenshot ?

Comment 2

[Mike Calmus]

Attachment: Screen shot of distorted image

I believe this may be a MacOS-only issue.

Comment 3

[Mike Calmus]

I don't know whether this is related or not, but the toolbar buttons are not
displaying correctly in the mouse-over and inactive states. This is visible in
the attachment by looking at the foward and back buttons.

Comment 4

[Stefan [:stefanh]]

Seems to work fine with Mozilla 2004040505 on Mac OS 10.2.8

Comment 5

[Dave Miller [:justdave]]

I can reproduce this... it's been annoying me for a few weeks now.

This is something in the core libraries...  Thunderbird and Firefox both have
this problem.  In Thunderbird, the signed and encrypted icons in the message
header (either from the built-in SMIME stuff or from Enigmail) have this
problem.  In Firefox, various images have this problem on various sites.  On the
ones I've checked so far, the problem images always seem to be PNG format
(though the attached screenshot shows it happening to a GIF).  Not all PNG
images display incorrectly.  The attached screenshot shows almost exactly what
I'm seeing - it looks like the images are interleaved, and the alternating rows
are out of sync with each other.

I can reproduce this on 3 different Macs (all three are using Mac OS X 10.3.3)
with varying versions of Firefox and Thunderbird (all nightlies, pulled within
the last few weeks)

Comment 6

[Asa Dotzler [:asa]]

Can someone track down when this regressed? There are archived nightly builds
availbable at http://archive.mozilla.org/pub/  Figuring out when it regressed
would go a long ways to getting it fixed. I don't see this in the tv2.gif linked
in the initial comment. Are there other examples you all could attach?

Comment 7

[Mike Calmus]

It works in the 2004-03-01 trunk build but not in the 2004-03-02. Based upon the
changed files in CVS, it looked like the problem was due to the fix to bug
#113406. I can also confirm that backing the changes between:
revs 1.5  and 1.6  for mozilla/gfx/src/mac/nsIImageMac.h,
revs 1.66 and 1.67 for mozilla/gfx/src/mac/nsImageMac.cpp and
revs 1.34 and 1.35 for mozilla/gfx/src/mac/nsImageMac.h
seems to fix the problem with a CVS build. However, as expected, doing so once
again causes bug #113406 to manifest. I'll take a look at the code to see if I
can see what is wrong.

Comment 8

[Olivier Cahagne]

Simon, not sure if you're aware of this one, this bug appears to be a regression
from bug 113406 as per commment 7.

Comment 9

[Christian :Biesinger (don't email me, ping me on IRC)]

*** Bug 244171 has been marked as a duplicate of this bug. ***

Comment 10

[Simon Fraser [no longer active]]

Can someone post a screenshot that's in a format that the browser can render?

Comment 11

[halim.boussoualine]

Attachment: Icon uncorrectly displayed by camino 0.8b

Comment 12

[halim.boussoualine]

Attachment: Icon correctly displayed by mozilla 1.6

Comment 13

[Simon Fraser [no longer active]]

that looks like some rowbyte weirdness.

Comment 14

[Asa Dotzler [:asa]]

Can those of you experiencing this comment on your connection speed and whether
or not a shift reload changes things? Is it always reproducible on the same
image? Can you link to some more examples?

Comment 15

[Asa Dotzler [:asa]]

While I'm not seeing this, I suspect that it's probably worse than the bug we
tried to fix. I'd really hate to return to failure on really large images.
Simon, do you think you'll have any time to look into this?  I think we have to
do something for 1.7 so I'm plussing this.

Comment 16

[Simon Fraser [no longer active]]

I need a sample image that shows this, and also info about whether it occurs for
different screen depths.

Comment 17

[Mike Calmus]

The most obvious images for me are the button bar images (reload, stop, etc.). I
just tried different screen depths. It seems to work fine at either Millions or
256. It only occurs when the depth is set to Thousands of Colors. Shift-Reload
obviously does not help.

Comment 18

[Dave Miller [:justdave]]

Hmm, confirmed, I'm set for Thousands.  If you want good examples, go look at
Orkut. :)  Almost every non-photo image on Orkut is experiencing this problem.

Comment 19

[Dave Miller [:justdave]]

Also confirmed, if I switch to 256 or Millions it goes away.

Comment 20

[jhp (no longer active)]

Attachment: patch

Seems to be another bug in CopyDeepMask.  In this case, it's not handling a
mask of alpha depth = 1 correctly when the screen resolution is set to
thousands of colors.  Seems to work fine for alpha depth > 1.  In
nsImageMac::Draw, we only call CopyDeepMask if the alpha depth > 1 (for
performance reasons?), so I put that code in nsImageMac::CopyBitsWithMask.  Now
it works for all cases of varying alpha depth.

Comment 21

[jhp (no longer active)]

Oh, forgot to mention.  The bug in CopyDeepMask seems to be directly related to
passing the 'ditherCopy' flag.  If I change that to 'srcCopy', the problem also
went away.

Comment 22

[Asa Dotzler [:asa]]

Comment on attachment 149419 [details] [diff] [review]
patch

I'm not sure who should do the reviews on this so I've requested from sfraser
and tor. If that's wrong or the r/sr order is backwards and either of you can
help, I'd certainly appreciate it.

Comment 23

[Simon Fraser [no longer active]]

Comment on attachment 149419 [details] [diff] [review]
patch

That makes perfect sense. We've seen other bugs in CopyDeepMask.

Comment 24

[chris hofmann]

Comment on attachment 149419 [details] [diff] [review]
patch

a=chofmann for 1.7..  can someone land today?  thanks

Comment 25

[jhp (no longer active)]

Checked in on branch and trunk.

Comment 26

[TT]

If I switch to 256 colors, the problem does NOT go away for me, but it does go
away if I switch to millions of colors.

Comment 27

[Dave Miller [:justdave]]

I attempted to verify the fix, but there's no Mac builds on the 0.9 branch and
the trunk is busted beyond usable at the moment.  I'll attempt to build my own
0.9 tonight and see what happens.

Comment 28

[halim.boussoualine]

As far as I am concerned, I downloaded yesterday the camino nightly build (build
2004052808 (v0.8+)) and the problem is gone, pictures displaying correctly in
256, thousands and millions of colors.
Thanks for the fix.

Comment 29

[Olivier Cahagne]

*** Bug 245102 has been marked as a duplicate of this bug. ***

Comment 30

[jhp (no longer active)]

verified by comment #28

Comment 31

[Mike Connor [:mconnor]]

checked in on aviary branch by Javier

Comment 32

[jhp (no longer active)]

This patch needed to be backed out, since it led to crashes.  Seems to be an OS
bug that is causing the crashes.  While the final fix may be switching to using
Quartz rather than QuickDraw (bug 245407), we'll leave this bug open in case
anyone can think of a workaround patch.

Comment 33

[jhp (no longer active)]

*** Bug 248756 has been marked as a duplicate of this bug. ***

Comment 34

[Stuart Morgan]

*** Bug 249162 has been marked as a duplicate of this bug. ***

Comment 35

[Stuart Morgan]

Updating summary to be more clear, and to be more findable by words used in
dupes (was "Some images are distorted in browser").

Comment 36

[jhp (no longer active)]

*** Bug 249213 has been marked as a duplicate of this bug. ***

Comment 37

[Guido Zeiger]

The Image on www.sonnenblen.de with the Name
http://www.sonnenblen.de/images/newanimation1.gif

isnt correctly displayed with
Mozilla 1.8a3
Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.8a3) Gecko/20040817

BUT would be correctly rendered with
Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.7) Gecko/20040603

The same problem is occured with the actual releases of FireFox :-((

Comment 38

[Uri Bernstein]

Attachment: minimal tescase

This is from bug 261251 (which should be marked as a duplicate of this bug).

A 96x96 pixels (2-bit) black square GIF.
When viewed with screen set to "thousands of colors" or "256 colors", the right
one-third of the square has interlaced black and white lines.

I'm seeing this on:
Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; rv:1.7.3) Gecko/20040913
Firefox/0.10

Comment 39

[Uri Bernstein]

Bug 260689 is also a duplicate of this bug.

Comment 40

[jhp (no longer active)]

*** Bug 261251 has been marked as a duplicate of this bug. ***

Comment 41

[jhp (no longer active)]

*** Bug 260689 has been marked as a duplicate of this bug. ***

Comment 42

[Floyd Miller]

This bug info seems to describe the problem I have noticed with some
transparent GIF images in Mac OSX.  The problem occurs in the latest
released FireFox (PR 0.1.01)and Mozilla build 2004101105 on Mac OSX
10.3.5 on a 500MHz, 256 MB G3 iBook and 500 MHz 512 MB PowerMac G4.

The corruption or distortion seems to depend on the rendered width of
the image.  I created a test case with examples at

    http://www.studiodust.com/ff_imagetest/imagetest.html

The test includes two similar images with different widths and
each is rendered twice.  Regardelss of which images is displayed,
if it's rendered width is 352 pixels the result is distorted.

I do not currently have a screenshot that shows the problem.
If it would be helpful, I will add one to the test case page
later today.

Comment 43

[Floyd Miller]

Attachment: Screen shot of Firefox 0.10.1 Mac OSX

This is a screen of transparent GIF corruption from
URL http://www.studiodust.com/ff_imagetest/imagetest.html

About Info: Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; rv:1.7.3)
Gecko/20041001 Firefox/0.10.1

Comment 44

[Hank Roberts]

Problem happens for Display Color set to "thousands of colors" but not when set
for "millions of colors"
OSX 10.3.5, on my Powerbook G3 "Pismo" with ATI Rage M3P8 display hardware.

Good example: boingboing.net, where with "thousands" the jackhammer animation at
top of page appears as twins side by side, each made up of a bunch of horizontal
black lines separated by white space, moving as one, up and down.
With color set to "millions" it appears correctly.

Comment 45

[Uri Bernstein]

*** Bug 265570 has been marked as a duplicate of this bug. ***

Comment 46

[Uri Bernstein]

Removed "fixed-aviary1.0" status since this was backed out (see comment 32).

Comment 47

[Greg K.]

Interesting. I never saw this problem until I toggled the option for
font-antialiasing in the General System Preferences options. Now, suddently,
it's everywhere.

Comment 48

[Asaf Romano (gone)]

*** Bug 273317 has been marked as a duplicate of this bug. ***

Comment 49

[:Gavin Sharp [email: [e-mail]]]

*** Bug 269570 has been marked as a duplicate of this bug. ***

Comment 50

[Nick Dimiduk]

I can confirm comment #44; when changing my display settings from "thousands of
colors" to "millions of colors" the bug doesn't manifest in either firefox or
thunderbird.  Switching back to thousands causes the bug to return.

Powerbook G4 12" 1.33 GHz; GeForce FX Go5200; Mac OS X 10.3.6 (7R28); Darwin 7.6.0

Comment 51

[Hiro]

*** Bug 274608 has been marked as a duplicate of this bug. ***

Comment 52

[Mischa Krilov]

Confirming comment #44: Bug still exists for me. I can only confirm that it
exists for animated GIFs at thousands of colors. I've got an iBook G3/600.

Using Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.7.5)
Gecko/20041107 Firefox/1.0 on OS X 10.3.7.

Comment 53

[Jasper]

*** Bug 276410 has been marked as a duplicate of this bug. ***

Comment 54

[Asaf Romano (gone)]

*** Bug 276798 has been marked as a duplicate of this bug. ***

Comment 55

[Uri Bernstein]

*** Bug 277019 has been marked as a duplicate of this bug. ***

Comment 56

[6v84lua02]

*** Bug 279175 has been marked as a duplicate of this bug. ***

Comment 57

[Asaf Romano (gone)]

should be fixed now...

Comment 58

[Mike Calmus]

Looks good to me.

Comment 59

[Asaf Romano (gone)]

assuming the "orginal patch" was backed out from 1.7 as well...

Comment 60

[Jason Bassford]

As far as I can tell, the only patch here was backed out and there's no mention
of what other patch / bug finally solved this problem.

-> WORKSFORME  (But, please correct that if there IS a specifically known
bug/patch that finally solved this - mention it in a comment.)

Comment 61

[Mike Calmus]

This was fixed by the patch for bug #245407.

Comment 62

[Miles Lane]

Works for me now.  YEA!!!

Comment 63

[Asaf Romano (gone)]

*** Bug 285109 has been marked as a duplicate of this bug. ***

Comment 64

[Dave Miller [:justdave]]

correcting resolution based on comment 61

Comment 65

[Hiro]

*** Bug 289349 has been marked as a duplicate of this bug. ***

Comment 66

[Uri Bernstein]

*** Bug 290601 has been marked as a duplicate of this bug. ***

Comment 67

[Anders Conbere]

*** Bug 294849 has been marked as a duplicate of this bug. ***

Comment 68

[Asaf Romano (gone)]

*** Bug 296835 has been marked as a duplicate of this bug. ***

Comment 69

[Simon Fraser [no longer active]]

*** Bug 287373 has been marked as a duplicate of this bug. ***