1026926 - [Flame][Calendar] Dots in month view do not show correctly (not a complete circle)

Status: RESOLVED INVALID

(Core :: Graphics, defect)

Description

[Peko Chen [:pekochen]]

Attachment: month view screenshot.png

please see attached file, the dot was cropped.

Comment 1

[Peko Chen [:pekochen]]

Hi Gareth,
Can you please help on this issue?
thank you so much~

Peko

Comment 2

[Miller Medeiros [:millermedeiros]]

this is some weird rounding issue, probably a Gecko bug related to the way fonts are rendered. If we change the left:-1.5rem; to any even number (eg. -1.4rem, -1.2rem, -1.6rem) the dots are displayed as a full circle.

Comment 3

[Evan Tseng [:evanxd]]

Hi Jerru, 

It is a Gecko issue.
We could reproduce the issue with the icon-dot app[1] in Flame device.
Could you help here?
Thanks.

[1] https://github.com/evanxd/gaia/tree/bug-1026926/apps/icon-dot

Comment 4

[Evan Tseng [:evanxd]]

Attachment: not-a-complete-circle.png

The screenshot of icon-dot app shows the incomplete circle.

Comment 5

[Jerry Shih[:jerry] (UTC+8) (inactive)]

I will check this problem.

Comment 6

[Jerry Shih[:jerry] (UTC+8) (inactive)]

Thanks, Evan.
I can reproduce this issue from comment 3 at flame.
checking...

Comment 7

[Evan Tseng [:evanxd]]

NP :)

Comment 8

[u459114]

Attachment: 1 device unit = 40 app units

Comment 9

[u459114]

Attachment: WIP position rounding before sumbit a cairo painting task

Here is a WIP to fix this bug

Comment 10

[u459114]

Attachment: Position rounding before submit a cairo glyph rendering call

Comment 11

[u459114]

The root cause of this bug:
Only flame, 1 css pixel is equal to 1.5 device pixel.
Let's said: 
1. you want to draw a 2-css-pixels-wide object at (11, 10) css position
   2-css-pixels-wide = 3-device-pixels-wide
2. it maps to (11 * 1.5, 10 * 1.5) = (16.5, 15) device position.
3. We expect 4 pixels draw on the screen, from (16, 15) device position
   16(blur pixel) 17 18 19(blur pixel)
   but, in fact, cairo only renders 3 pixels into render target, without pixel "19", your object looks like been cut.

Comment 12

[u459114]

Attachment: Position rounding before submit a cairo glyph rendering call

Comment 13

[u459114]

Theoretically, there should be no test case impact by this change, but since gfxFont is a high fan-in module, submit a full try: 
https://tbpl.mozilla.org/?tree=Try&rev=623f5ea3ebbf

Comment 14

[u459114]

Bad luck, many test case failed because of this change
REFTEST TEST-UNEXPECTED-FAIL | http://10.26.132.20:30472/tests/layout/reftests/bidi/bidiSVG-02.svg | image comparison (==), max difference: 253, number of differing pixels: 76
REFTEST TEST-UNEXPECTED-FAIL | http://10.26.132.20:30472/tests/layout/reftests/bidi/bidiSVG-03.svg | image comparison (==), max difference: 253, number of differing pixels: 229

Comment 15

[u459114]

There are two "round" in ClampAndAlignWithPixels 
http://dxr.mozilla.org/mozilla-central/source/layout/generic/nsGfxScrollFrame.cpp#1925
http://dxr.mozilla.org/mozilla-central/source/layout/generic/nsGfxScrollFrame.cpp#1930

I think it's the main reason of this failure

Comment 16

[u459114]

https://tbpl.mozilla.org/?tree=Try&rev=fb7df423ed90

1. round off right before cario_show_glyphs.
2. All transform had been apply at this moment.

Comment 17

[Miller Medeiros [:millermedeiros]]

this is not blocking the calendar visual refresh anymore since we used a hack to fix the problem (changed position to an even number).

Comment 18

[Sylvestre Ledru [:Sylvestre]]

No assignee, updating the status.

Comment 19

[Sylvestre Ledru [:Sylvestre]]

No assignee, updating the status.

Comment 20

[Sylvestre Ledru [:Sylvestre]]

No assignee, updating the status.

Comment 21

[Sylvestre Ledru [:Sylvestre]]

No assignee, updating the status.

Comment 22

[Gregory Pappas [:gregp]]

Closing old B2G bugs