830731 - Firefox 18 and up do not pass Acid2 test anymore

Status: RESOLVED DUPLICATE of bug 418235

(Core :: General, defect)

Description

[squalou.jenkins]

Attachment: Capture.PNG

User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/20100101 Firefox/18.0
Build ID: 20130104151925

Steps to reproduce:

connect to :
http://acid2.acidtests.org/#top

with Firefox 18 on windows 7 64


Actual results:

The picture displayed is wrong. (see screenshot)


Expected results:

the reference picture should be displayed
http://acid2.acidtests.org/reference.html

Comment 1

[Boris Zbarsky [:bzbarsky]]

Hmm.  I can reproduce on Mac, with http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=c09a0c022b2e&tochange=a680fd777c3b as the regression range, but it seems to work for others...

Comment 2

[Boris Zbarsky [:bzbarsky]]

That's the day hidpi support landed on Mac, which is very likely to be part of what's going on.

But that's unlikely to be affecting Windows.  Reporter, just to make sure, you're seeing this on Windows?

Comment 3

[Jonathan Kew [:jfkthame]]

If you disable HiDPI mode (set gfx.hidpi.enabled to 0), the problem no longer happens on Mac, which confirms it's related to HiDPI support. And whether it fails or works will depend whether you're on a Retina MacBook or not.

Not sure offhand why it'd be broken on Windows, though. Is it still broken on Aurora (FF 20) or Nightly (21) builds?

Comment 4

[Alice0775 White]

looks like Bug 418235

Comment 5

[Frank Yan (:fryn)]

(In reply to Jonathan Kew (:jfkthame) from comment #3)
> Not sure offhand why it'd be broken on Windows, though. Is it still broken
> on Aurora (FF 20) or Nightly (21) builds?

Yes.

Comment 6

[squalou.jenkins]

(In reply to Boris Zbarsky (:bz) from comment #2)
> That's the day hidpi support landed on Mac, which is very likely to be part
> of what's going on.
> 
> But that's unlikely to be affecting Windows.  Reporter, just to make sure,
> you're seeing this on Windows?

I did see this on windows,
to be more specific : a windows Seven 64 bits Enterprise. (French)

Now I've just checked on an older XP 32bits : the problem does NOT exist there.

Comment 7

[squalou.jenkins]

also for fun:just tested on android 4.1.1. doesn't work either, but slightly different result.

Comment 8

[Damian Shaw [Quan]]

Hi, I can see the screenshot from the original report is much bigger pixel for pixel than the actual Acid2 test. Can the original reporter reset their zoom settings or set them to 100% (should just be "ctrl + 0" unless you have an addon which defaults them to something else) and refresh the Acid2 test.

I think original report is exact duplicate of Bug 418235, I think it's just a happy coincidence that bz spotted this causes a knock on effect on hidpi. I don't have a Mac so I don't test this, how do webkit browsers fair for the same problem?

Comment 9

[Jonathan Kew [:jfkthame]]

Attachment: acid2 test on retina macbook - from left to right: Safari, Firefox, Opera, Chrome

On my retina MacBook, Safari fails similarly to Firefox (though not identical), and Opera looks even worse (though the failure is in the same area of the image); only Chrome passes.

Comment 10

[Jonathan Kew [:jfkthame]]

(In reply to Damian Shaw [Quan] from comment #8)
> Hi, I can see the screenshot from the original report is much bigger pixel
> for pixel than the actual Acid2 test. Can the original reporter reset their
> zoom settings or set them to 100% (should just be "ctrl + 0" unless you have
> an addon which defaults them to something else) and refresh the Acid2 test.

Ah, well spotted. Another factor that might be relevant: has the setting "layout.css.devPixelsPerPx" been changed from its default value of 1.0 (for Windows) in about:config?

Comment 11

[Damian Shaw [Quan]]

Attachment: Chrome 150% Zoom on Windows

How is Google Chrome handling retina displays? Do they use "normal zoom" like Safari, Opera and Firefox all clearly do and therefore get these artifacts or are they using some other methodology entirely?

I think either Chrome have implemented a "better" method and produce more consistent results or they are test fixing. Attached is a screenshot of loading the Acid2 Page at 150% Zoom using Version 26.0.1384.2 canary Chrome on Windows. If they were using the same "normal zoom"  this should look like the expected smiley face, it does not.

Comment 12

[Hixie (not reading bugmail)]

The Safari and Mozilla renderings here are plausible if the images are being scaled with something other than nearest-neighbour. What's going on is that there's three layers, a red background, then two layers consisting of bitmaps of the form:

    XO
    OX

...where X is yellow and O is transparent, with one of the layers being offset by 1 pixel horizontally relative to the other.

Comment 13

[Henri Sivonen (:hsivonen)]

(In reply to Hixie (not reading bugmail) from comment #12)
> The Safari and Mozilla renderings here are plausible if the images are being
> scaled with something other than nearest-neighbour.

http://hsivonen.iki.fi/last-html-quirk/ is bad enough an effect of Acid2. I really hope Acid2 doesn’t end up locking the Web into nearest neighbor scaling.

> What's going on is that
> there's three layers, a red background, then two layers consisting of
> bitmaps of the form:
> 
>     XO
>     OX
> 
> ...where X is yellow and O is transparent, with one of the layers being
> offset by 1 pixel horizontally relative to the other.

What’s the type of legacy engine bug that this testing scheme is trying to reveal?

Comment 14

[squalou.jenkins]

(In reply to Damian Shaw [Quan] from comment #8)
> Hi, I can see the screenshot from the original report is much bigger pixel
> for pixel than the actual Acid2 test. Can the original reporter reset their
> zoom settings or set them to 100% (should just be "ctrl + 0" unless you have
> an addon which defaults them to something else) and refresh the Acid2 test.
> 
> I think original report is exact duplicate of Bug 418235, I think it's just
> a happy coincidence that bz spotted this causes a knock on effect on hidpi.
> I don't have a Mac so I don't test this, how do webkit browsers fair for the
> same problem?

You're absolutely right !
I completely forgot I was zoomed by default (small screen, big resolution).

With the default settings, it looks perfect.

Comment 15

[Hixie (not reading bugmail)]

Henri: I think it was looking for off-by-one positioning bugs, but I forget the details (it was years ago).

For what it's worth, I am skeptical that filtering a 2x2 bitmap of this form is going to give you better results in general, even outside the acid tests. But I agree it's a poor way to test this for the modern age. I have no problem saying Acid2 doesn't apply to high-density displays.

Comment 16

[Damian Shaw [Quan]]

Reported bug is is duplicate of Bug 418235, bz's case is a non-issue as stated by Hixie, closing as duplicate and original bug can probably be marked as invalid.

Comment 17

[Dave Garrett]

It would be really nice if an eventual release of an Acid4 test would include the parts of the older Acid tests that we still cared about so we could stop worrying about them. ;)