1570363 - transform-origin incorrectly calculated for fill boxes that have zero width or height

Status: RESOLVED FIXED

(Core :: Layout, defect, P2)

Description

[Paul LeBeau]

User Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:68.0) Gecko/20100101 Firefox/68.0

Steps to reproduce:

If an element's bounding box has zero width or height, then the transform origin is incorrect for transform-box: fill-box.

Try this test case: https://jsfiddle.net/17ym4zev/

<svg width="500" height="500">
<g class="small">
<path d="M 0,250 L 500,250" style="stroke: red"/>
</g>
<g class="small">
<path d="M 0,250 L 500,250.1" style="stroke: green"/>
</g>

<g class="small">
<path d="M 250,0 L 250,500" style="stroke: red"/>
</g>
<g class="small">
<path d="M 250,0 L 250.1,500" style="stroke: green"/>
</g>
</svg>

path {
stroke: black;
stroke-width: 8;
}

.small {
transform: scale(0.25, 0.25);
transform-origin: 50% 50%;
transform-box: fill-box;
}

Actual results:

The red lines show the incorrect behaviour. The green lines have been given a tiny width (or height) of 0.1 to show where they out to be.

It's as if the origin is being calculated as (width/2) instead of (x + width/2).

Expected results:

The red lines should be centred in the SVG. They should be where the green lines are.

Tested as working correctly on Chrome and Safari.

Comment 1

[Paul LeBeau]

Attachment: Screenshot showing incorrect positioning

Added screenshot.

Comment 2

[Alice0775 White]

Regression window:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=a675c5d7eb76887a3e4b24548d621c9cc05a1545&tochange=22caf5b03354b71c909e67ec6bc71133494d3137

Suspect: e7732e4cb0c9abe6823f4f02313b0557ed33feab Matt Woodrow — Bug 1456111 - Make sure an empty pixel rectangle gets converted to an empty app unit rectangle, regardless of scale factor. r=tnikkel

Matt, Your patch seems to cause the problem. Can you please look into this?

Comment 3

[Patricia Lawless]

Hey Sean, Can we get a priority on this one? (regression triage)

Comment 4

[Sean Voisen (:svoisen)]

Looks like we run into this since aRect.IsEmpty() will be true for 0 width OR height ... I'll mark as P2, but unsure how common the use case is to have 0 width or height bounding box with a transform. I suppose we mostly see this for SVG.

Matt, thoughts here?

Comment 5

[Paul LeBeau]

It's a significant problem for SVG, because vertical or horizontal lines are common. And they have zero-width or height bounding boxes. Even if the stroke width is large.

Comment 6

[Paul LeBeau]

Is it possible to get a reason why this is a "won't fix" please? The current behaviour is counterintuitive and deviates from the behaviour of other browsers.

Comment 7

[Timothy Nikkel (:tnikkel)]

It's wontfix for Firefox 69 only which is going to ship on Tuesday, there is no reasonable path to fixing this on a release coming out in 6 days. It's just a way for release drivers to track bugs.

Comment 8

[Paul LeBeau]

Thanks. Sorry I didn't realise how the flags worked. I do now.

Comment 9

[Liz Henry (:lizzard)]

Following up in email with Sean and Matt since this missed the 70 train as well.
We could still take a patch for 71/72.

Comment 10

[Pascal Chevrel:pascalc]

Marking as fix-optional for 71 in case a patch materializes before RC week.

Comment 11

[Liz Henry (:lizzard)]

Likely too late for 72 at this point but we could still get a fix in for 73.

Comment 12

[Liz Henry (:lizzard)]

Following up in email, this clearly slipped into the backlog.

Comment 13

[Sean Voisen (:svoisen)]

Yes, this slipped. Getting it into our quality backlog.

Hey Jessie — Might we be able to get help from your team on this? (I think MattW is swamped at the moment.) The regressing patch is related to a web painting optimization where an empty gfx rect gets converted to an empty app rect, but looks like we can't always do that.

Comment 14

[Jessie [:jbonisteel] pls NI]

Miko - would you be able to take a look?

Comment 15

[Miko Mynttinen]

Attachment: 1570363.html

Simplified testcase, with an additional path that Chrome and Safari render differently.

Comment 16

[Miko Mynttinen]

Attachment: Bug 1570363 - Do not discard width and height for empty SVG reference rects r=mattwoodrow

Comment 17

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/6fcfdbf5137c
Do not discard width and height for empty SVG reference rects r=mattwoodrow

Comment 18

[Daniel Varga [:dvarga]]

https://hg.mozilla.org/mozilla-central/rev/6fcfdbf5137c