598890 - -moz-crisp-edges can result in a ghost line being wrapped around.

Status: RESOLVED WORKSFORME

(Core :: Graphics: Canvas2D, defect)

Description

[Grant Galitz]

User-Agent:       Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:2.0b7pre) Gecko/20100922 Firefox/4.0b7pre
Build Identifier: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:2.0b7pre) Gecko/20100922 Firefox/4.0b7pre

When -moz-crisp-edges is used on the canvas element, a line maybe appearr as the first or last single-pixel row (not scaled or resized, but actual one-pixel height) in which that row seems to have been wrapped from the opposite side of the canvas. If this ghost line is the last line, it appears to contain pixels that should really be in the first row. vice versa.

Does not occur with normal filters, only nearest-neighbor by -moz-crisp-edges.
Also, this is a bug of whchI've seen in FF3.6 on Mac Intel as well.

Reproducible: Always

Steps to Reproduce:
1.Open a rom file
2.Wait for the rom file to load and start looking at the top and last lines of the canvas.
3.A ghost line will appear, and it will shift its location when the canvas has been resized by dragging the bottom right corner of the container.
Actual Results:  
A wrapped / ghost line of pixels appear on the canvas upon close inspection when certain roms provide the right contrast to easily discern the ghost lines.

Expected Results:  
There should be no ghost line of any sort on the top or botom row of pixels.

Some gameboy and gameboy color rom images do not work properly yet. You may have to try multiple roms before a rom image plays.

Comment 1

[Grant Galitz]

Seems that there's a flawed nearest-neighbor filter.

Comment 2

[Grant Galitz]

Attachment: A screen shot of Pokemon Yellow running with -moz-crisp-edges set, the bottom pixel line is bugged from NN scaling

Comment 3

[Joe Drew (not getting mail)]

It's a little tricky to reproduce (you have to resize the gameboy element from the bottom-right corner; play around with it until you get the repeating), but I can definitely reproduce it. We probably put our canvas, or the images inside it, in repeat mode?

Comment 4

[Grant Galitz]

I should note that this site sends the webpages as application/xhtml+xml to browsers that support that mime type, while others, like IE8, get a recompiled webpage as html. There are differences between the two types being sent. Though this should have no bearing.

Comment 5

[Grant Galitz]

Attachment: A screen shot of Super Mario Bros. Deluxe running with -moz-crisp-edges set, the TOP pixel line is bugged from NN scaling

Now the ghost line is on the top.
You'll see that what should be part of the bottom tiling is now the first row of pixels for this instance.
This is a problem with how the canvas is rescaled when nearest-neighbor is in use (I've only seen this on Macs as of now).

Comment 6

[Grant Galitz]

Should be changed to WILL result in a ghost line being wrapped around. Sucks that the scaling is pretty much ruined by this small bug.

Comment 7

[Grant Galitz]

Tested Ubuntu 10.03
There's no ghost-line on Ubuntu, but it does appear Ubuntu is not setting the size right and making the line where the ghost line would have been as a transparent line, as if the height was rounded off.

Comment 8

[Grant Galitz]

I meant Ubuntu 10.04, lol

Comment 9

[Grant Galitz]

https://bugzilla.mozilla.org/show_bug.cgi?id=490997 seems to be related to this bug (Slightly different problem, probably caused by the same bug).

Comment 10

[Grant Galitz]

Also https://bugzilla.mozilla.org/show_bug.cgi?id=444990

Comment 11

[Grant Galitz]

Attachment: Rendering artifact on jsgameboy app on FaceBook

Black vertical line on the right side of the canvas should not even exist!
The emulator on facebook is the same one used in the other screenshots, it's just being loaded in as an iframe.

Comment 12

[Grant Galitz]

In the newer nightly builds, there isn't a ghost line being wrapped around, but some weird black or white artifact line being drawn now instead.

Comment 13

[Grant Galitz]

any update on this (Will it even be looked into before the FF4 release)? It really has been a boog since 3.6 at least.

Comment 14

[Mardeg]

Can you retest this please now that bug 468496 is fixed?
If it still happens, then it may be related to bug 450140

Comment 15

[Grant Galitz]

Ah, it was fixed a few months ago, it seems someone ninja-fixed it. :D