342513 - [SVGSVGElement].width.baseVal.value returns 1 for 100% wide SVG

Status: RESOLVED DUPLICATE of bug 1283539

(Core :: SVG, defect)

Description

[Arpad Borsos [:Swatinem]]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1a3) Gecko/20060622 BonEcho/2.0a3
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1a3) Gecko/20060622 BonEcho/2.0a3

[SVGSVGElement].width.baseVal.value returns 1 for a 100% wide SVG.
This bug only occurs in a recent trunk build width gecko 1.9. It works flawlessly width a 1.8 branch build.
[SVGSVGElement].viewBox.baseVal.width returns the correct width however.

Reproducible: Always

Steps to Reproduce:
retrieve the value of [SVGSVGElement].width.baseVal.value
Actual Results:  
1

Expected Results:  
the correct width in pixels, e.g. 1158

Comment 1

[tor]

Do you have a testcase you can attach?

Comment 2

[Arpad Borsos [:Swatinem]]

Attachment: testcase that prints the width of a 100% wide svg

I've just created a simple testcase. This prints out the svgDoc.width.baseVal.value and svgDoc.viewBox.baseVal.width values of a svg with width="100%".
The two values match on a branch build. A trunk build outputs 1 for svgDoc.width.baseVal.value.
Opera does not support the viewBox method at all.
This bug breaks my SVGraph examples at http://swatinemz.sourceforge.net/svgraph.xhtml

Comment 3

[tor]

Attachment: pure svg testcase - click on red

Comment 4

[tor]

Some results of testing - seems like little agreement:

FF 1.5:

        width.baseVal.value   = pixel width of area
        viewBox.baseVal.width = pixel width of area

FF trunk:

        width.baseVal.value   = 1
        viewBox.baseVal.width = pixel width of area

Opera 9:

        width.baseVal.value   = pixel width of area
        viewBox.baseVal.width = 0

Batik 1.6:

        width.baseVal.value   =  pixel width of area
        viewBox.baseVal.width =  <"getViewBox not implemented">

Adobe SVG Viewer 3.0:

        width.baseVal.value   = <"null or not an object">
        viewBox.baseVal.width = <"null or not an object">

Comment 5

[Arpad Borsos [:Swatinem]]

Todays build regressed some more. I don't know which checkin caused it but it now outputs the following:
svgDoc.width.baseVal.value = 1
svgDoc.viewBox.baseVal.width = 0
That means there is currently no way to get the actual width of a 100% wide SVG with scripting :(

Comment 6

[tor]

Attachment: make nsSVGLength2 handle a outersvg context

This doesn't reliably fix the HTML testcase attached to this bug, as the load event appears to fire before the first reflow, which sets up the outer svg dimensions.  Not sure if that is per spec or not (the random values due to an uninitialized variable before the first reflow definitely aren't, but that's trivial to fix once we decide what the appropriate behavior should be).

Comment 7

[tor]

Attachment: missed header file

Comment 8

[Arpad Borsos [:Swatinem]]

Attachment: mochitest for embedded testcase

I wrote a mochitest for the embedded testcase to practice writing mochitests.
Something else I noticed while playing with firebug:
>>> document.getElementById('svg').width.baseVal.newValueSpecifiedUnits(5,100) // set value to 100px
>>> document.getElementById('svg').width.baseVal.convertToSpecifiedUnits(2) // convert to percentage
>>> document.getElementById('svg').width.baseVal.value
100 // value is always in "user units" according to spec. "user units" is always equal to pixels?
>>> document.getElementById('svg').width.baseVal.valueAsString
"10000%" // this is wrong

The SVGLength implementation somehow uses a fixed conversion ratio of 100 from "user units" (pixels) to percentage. The embedded SVG actually grows when I use the convertToSpecifiedUnits method.

Comment 9

[Jonathan Watt [:jwatt]]

Ah, crap. We shipped with this broken? :-(

Thanks a lot for the mochitest Arpad! It looks good to me.

Comment 10

[Arpad Borsos [:Swatinem]]

Attachment: reftest for convertToSpecifiedUnits

This is a reftest which tests the bug that the SVG is growing when using the convertToSpecifiedUnits method. I hope I've done that right, I didn't get the reftests to start :(

Comment 11

[Jonathan Watt [:jwatt]]

The basic problem here is our use of nsSVGElement::GetCtx which returns an nsSVGSVGElement* to the ancestor element that provides the coordinate context for the element on which it is called. Unfortunately there's no such ancestor for <svg> elements that are the root of an SVG document fragment, so GetCtx returns null. GetAxisLength then falls back to returning 1 when it doesn't get a context, so GetUnitScaleFactor returns 100 and we get 100/100 in nsSVGValue2::GetBaseValue.

Comment 12

[Jonathan Watt [:jwatt]]

Arpad: does that reftest actually fail currently? I thought this bug was only for percentage width/height on the rootmost <svg>. Besides that, it's preferable to get rid of the -ref file and just compare to pass.svg if possible. You can hopefully do that if you set the margin, border and padding on the html and body elements to none. Also, better to use convertToSpecifiedUnits(nsSVGLength.<ENUM_CONSTANT>) instead of convertToSpecifiedUnits(2). I've no idea offhand what 2 is. :-)

Comment 13

[Arpad Borsos [:Swatinem]]

I will post an updated reftest if I can get this reftest infrastructure to run correctly.

Comment 14

[Robert Longson [:longsonr]]

I think this would be better as a mochitest. It's the output of convertToSpecifiedUnits that you are trying to test. What gets rendered is a side effect.

Comment 15

[Arpad Borsos [:Swatinem]]

Attachment: updated mochitest

This also tests the behavior of convertToSpecifiedUnits and friends.

Why does mozilla use SpecifiedUnits internaly and converts them to user units on every access to .value and not the other way around?

I also noticed that besides GetAxisLength also GetMMPerPixel always returns 1.

Comment 16

[Robert Longson [:longsonr]]

Patch in bug 361920 fixes this so that we're compatible with Opera.

Comment 17

[Robert Longson [:longsonr]]

Fixed by check in for bug 361920

Comment 18

[Robert Longson [:longsonr]]

Attachment: mochitest updated to tip with invalid test removed

Comment 19

[Robert Longson [:longsonr]]

checked in http://hg.mozilla.org/mozilla-central/rev/bbbe69d822c0

Comment 20

[Robert Longson [:longsonr]]

Use getBoundingClientRect if you want to get the pixel width of an <svg> element.

Comment 21

[Erik Haukjær Andersen]

Being new to svg, I tried to create some pretty basic stuff that didn't work (in firefox),
Now I see it actually works in chrome, and this particular problem has been present (reported) 10 TEN years ago.
And it is not solved ! (at least not in the pretty old firefox 46.0.1 I'm using).
So it is the policy of Firefox/Mozilla not to fully implement the w3c spec - as the SVGSVGElement's width returned Value contains what seems to be garbage - people can just use some other way (getBoundingClientRect) - so firefox specific code.

Also the comments in the bug-report reveals a reason for firefox being so cpu-hungry (poor implementation, of something, which in the standard actually provides reasonable ways to do it efficient, as pointed out by Arpad)

Incredible, no wonder people (users and developers) move away from firefox/mozilla