SVG Working Group Teleconference -- 30 Aug 2012

<trackbot> Date: 30 August 2012

<krit> Zakim: aaaa is me

<scribe> scribenick: nikos

CSS4 image type as paint value

<heycam> http://www.w3.org/mid/FBBDBDBC-F3CD-404E-AFC1-0369A75DAA89@adobe.com

krit: Doesn't matter about the version. I'd just like to use CSS image
... then we get linear gradient and other gradients
... If we use CSS image as a paint server then it's possible to use images or filters or anything that is defined by CSS image

birtles: What about making paint servers part of images?
... a lot of CSS specs refer to image values, why not do it the other way around
... then you could use an SVG gradient in CSS

krit: That is in CSS4 image

<ed> so <rect fill="linear-gradient(bottom, rgb(149,227,138) 47%, rgb(179,255,166) 74%)" .../> for example

krit: we need to figure out how it works because it could lead to circular dependency
... The element function can reference paint servers from SVG

heycam: Do you want paint server references just inside that element or do you allow urls?

<krit> http://dev.w3.org/csswg/css3-images/#image-values

krit: I am just asking if we can use the css image type in SVG fill and stroke properties
... in the beginning I'd be fine with gradient

<heycam> ack

krit: image would be cool but if we just have gradient that would be good

cyril_: for gradients or patterns you have the gradient limits that tells you how the gradient space maps to the image space
... how would you do that

krit: you need to define how the bounding box is defined for svg and it would maybe be object bounding box

cyril_: would the svg objects crop the image or would the image be entirely contained in the svg object?

krit: you mean referencing an image with a url?

cyril_: any image. css3 image fills an object with an image
... you might end up with an image size different from the svg object size so you need to crop, or reflect, or something

krit: at the beginning we might just define gradient. its easier

<krit> http://dev.w3.org/csswg/css3-images/#gradient-type

cyril_: Using percentages or integers or fixed values, you can say the image will take more or less than the bounding box of the object
... you have to define if you crop, or what you do

krit: I agree with that
... At the beginning I'd be fine to just use gradients as this isn't a problem for them
... you just need to define a box, and we'd just use the object bounding box

heycam: I think cyrils concern was that it's not clear whether gradients cover the entire box or whether you pad or repeat
... I think we need a way to specify whether you repeat in horizontal or vertical directions

nikos: I think dirk is saying you use the object bounding box so the dimensions always match up

cyril_: in css gradients I don't think you can specify where the 2 end points of the vector defining the gradient are - except for top or bottom

heycam: you can give a length value

cyril_: I'm seeing left, right, top, bottom, thats it
... you can't say I want the points to be 10% inside
... then you don't have to specify pad or repeat
... there is a repeating linear gradient

krit: that's a limitation in css gradients in general, why is it a problem for svg?

cyril_: I'm just saying it needs to be specified how the object is filled when the gradient is not big enough

heycam: from what I can see it will always be big enough
... you have stops that go from 0 to 100%

cyril_: Yeh with gradient it seems you will always fill the object

heycam: With images I think you're right about fitting.
... Are the only types images and gradients?

cyril_: url image_list and gradient

heycam: I'm sure we could define how the image renders without adding more controls like various background repeat properties
... I don't know how useful it is to solve that here
... maybe in that case you need to use a pattern for example

krit: Can we resolve for linear gradient and solve the other problems later?

heycam: I'm happy with it

ed: I agree

cyril_: what would be the benefit compared to using svg gradient?

krit: you can apply a linear gradient to a class of object
... and use the same gradient for html and svg content

cyril_: ok

heycam: it's also an easier way to specify a gradient

Resolution: Paint values will allow gradients from CSS3 image

krit: CSS image 3 is already a REC

<heycam> http://dev.w3.org/csswg/css3-images/#repeating-gradients

heycam: I see you can do repeating gradient styles in CSS

cyril_: in any case it's going to fill the whole object

krit: will it be in the SVG2 FPWD or is it for the future?

heycam: I think we've discussed this before in the context of referencing as many CSS specs as possible

krit: so SVG 2 then
... next draft release

heycam: If you have time

<scribe> ACTION: Dirk to edit SVG 2 draft to add CSS 3 gradients as a paint value [recorded in http://www.w3.org/2012/08/30-svg-minutes.html#action01]

<trackbot> Created ACTION-3346 - Edit SVG 2 draft to add CSS 3 gradients as a paint value [on Dirk Schulze - due 2012-09-06].

Filter subregion clipping

ed: I sent my comments to the public mailing list
... I think it addressed both discussions
... feOffset and how to define input and output clipping

<ed> http://lists.w3.org/Archives/Public/www-svg/2012Aug/0143.html

ed: so in Opera we treat feOffset not exactly as the spec tells us to, because if you consider the example in the mail, feOffset itself doesn't specify a primitive subregion - doesn't have x, y and width. Intuitively for me I wouldn't expect it to clip, I'd expect it to move the previous primitive

krit: I would agree that we don't need to clip the offset, we move it around
... Do you clip if you move a primitive to the outside of filter regions then bring it back in?

heycam: do you clip at each step or just once at the end?

ed: Could you post an example?

<krit> <filter id="feoffset1" primitiveUnits="objectBoundingBox" x="0%" y="0%"

<krit> width="100%" height="100%">

ed: So you move the result out and back again?
... what would you expect in that case?

krit: I think the spec says it's clipped away
... when you move it back you are just moving transparent black back

ed: I'm not really sure what I'd expect in that case
... I think I might expect to be able to move the result back
... it would be interesting to see what the implementations do

krit: it would be interesting to see what gaussian blur does

ed: I think it would be interesting to have some way of letting the user agent find out what the filter region is automatically, but we'd have to be careful not to break content

krit: In this case we definitely should clip, the question is does opera clip
... Do you clip the input or the output?
... that's a general question

ed: I think we do input clipping

krit: filter effects spec requires input and output clipping, which is not useful in my eyes

ed: So do we want to control this, or do we require one particular way?

krit: I think both input and output is not useful
... I don't care if we go with input or output

heycam: what might break if we switched? if we specified just output

krit: we are really discussion an edge case so I wouldn't expect much to break

heycam: it only matters to primitives that do things outside their region
... most filter primitives say their input and output regions are the same
... is that right?

krit: by default, firefox's subregion depends on the previous subregion so if you have 2 filters of different size, it is the union of both
... when you specify x, y, w and h what should be clipped, the input or the output?

ed: do you have a test case?

krit: I have posted something to the mailing list

ed: I'm looking for something that doesn't use feOffset

krit: it's difficult without feOffset

ed: We treat it differently, we don't take the union of the inputs, we use the filter region itself if you don't specify

heycam: apart from whether you do clipping on input or output. Dirk, do you think it makes sense to special case feOffset?

krit: if you don't have input clipping it doesn't matter
... in that case the subregion depends on the clipregion of the feOffset filter
... I'd have to think about it some more
... right now the specification wants to clip to the region always

heycam: the spec only says clip the input and output both
... to make feOffset more useful, would we need to change it to only output or only input? or is that a separate issue

krit: I think it's separate

ed: I think it would be nice to have an analysis of where and when it's useful to have each clipping method

heycam: I think it would be good too, I find it hard to visualise why you'd take one or the other

ed: There's one example in the spec but I don't think it uses feOffset
... There's also the example from the mail and the bug report
... but apart from that I don't think we have many tests for this

<scribe> ACTION: Dirk to investigate and expand on the proposal of sub region clipping for Filter Effects [recorded in http://www.w3.org/2012/08/30-svg-minutes.html#action02]

<trackbot> Created ACTION-3347 - Investigate and expand on the proposal of sub region clipping for Filter Effects [on Dirk Schulze - due 2012-09-06].

heycam: what about for the other issue of whether feOffset behaviour should be special?

krit: I'll look into that as well

ed: They're similar but not quite the same

<krit> nikos: great, thanks. Let us share the examples next week

<krit> haha

heycam: i imagine you could construct the filter to avoid the problem - don't shift outside the region

ed: not many people write such advanced filters that they would run into this issue

heycam: we'll pick up the discussion once dirk has had a look

Behaviour of feOffset dx/dy values with percentages

heycam: does anyone support percentage values for dx and dy?

krit: if you use 50% then it is treated as 50
... on webkit

heycam: we should make percentages work or disallow them

krit: We already support percentage - but indirectly

heycam: why is it that dx and dy unitless values are treated as unit bounding box values and not ...
... if you have unit less values they are just user units
... there's nothing weird and I'd expect percentages to be percentages of the viewport
... if you have primitive space = user space on use, dx and dy should be percentages of the viewport?

krit: if you supported percentages, but we don't

heycam: do we have other primitives that take percentages?

krit: no

heycam: my feeling is that we should make percentages work

krit: and for other cases?
... where percentages could work?

heycam: so the issue with standard deviation is it's a number not a length right?

<krit> https://dvcs.w3.org/hg/FXTF/raw-file/tip/filters/index.html#FilterFunction

heycam: in the corresponding shorthand if we are going to support lengths, we should take lengths in the element
... I'm thinking whether it would be more consistent in general to support that, but I'm not sure that it's important

ed: is there any particular examples?
... I'm looking for some kind of testcase to try out
... just wondering if you had one handy

heycam: I think dx and dy look like they go with width and height, you might expect them to take the same kinds of value

krit: I agree

heycam: I would be happy to say dx and dy take lengths and percentages and forget about the other things like standard deviation

ed: I'd like to see an example before we make any changes

<scribe> ACTION: Dirk to provide examples of percentages with dx and dy in Filter Effects [recorded in http://www.w3.org/2012/08/30-svg-minutes.html#action03]

<trackbot> Created ACTION-3348 - Provide examples of percentages with dx and dy in Filter Effects [on Dirk Schulze - due 2012-09-06].

Commas vs spaces in SVG View fragment identifiers

<krit> http://dev.w3.org/csswg/css4-images/#image-fragments

heycam: Robert brought up in the mailing list some text in the linking chapter that talks about how you must use commas rather than spaces in the viewbox part of an svg fragment. You can use %20 to encode spaces but it's easier to use commas.
... there's a similar issue in preserve aspect ratio, because you can have the slice value after the xMinyMin part
... and that normally takes a space between them
... I think viewbox also would normally take a space, so there's no issue saying just use a comma
... I think Robert was asking for a clarification on the grammar

krit: I don't like spaces in iri
... I pasted a link to css image fragments, they comma separate
... I don't have a really strong opinion but I feel its more natural

heycam: I agree, you don't really want to url encode

ed: I agree with the comma, even though I don't like the svg view syntax.

krit: the css specifications use uri and svg uses iri
... in theory there's a difference but in practice there's not
... do we have a strong opinion that we should use iri?

heycam: I raised something on the mailing list about the differences. I didn't get a clear idea in the end whether one or the other should change or whether we ignore the situation

krit: Filter Effects uses iri and the discussion on exclusions they want to use parts of svg that use iri
... it would be good to harmonise the specifications

<heycam> http://lists.w3.org/Archives/Public/www-style/2012May/0772.html

heycam: here's the mail where I asked how things in brackets are parsed, because I think the spec doesn't really say.
... I didn't really get a satisfactory answer

ed: The Filter Effects spec uses iri because SVG 1 used iri

heycam: If you look at the data url in the email I linked to

krit: iri just supports more characters right?

heycam: yes, you can use non ascii characters
... for uri you would have to escape them somehow
... the test uses a css background image that has some non ascii characters
... and that seems to work everywhere so I assume browsers are interpreting as iri

krit: you tested locally?

heycam: no it's on a server
... if you view my example and 'view source'

krit: it doesn't work in Safari

ed: It doesn't seem to wokr in Opera either

heycam: It may be because I left a bracket out of the example
... I think the issue is that the css spec needs to define how things inside the url brackets are parsed and interpreted

<heycam> http://lists.w3.org/Archives/Public/www-style/2012May/0824.html

<heycam> http://räksmörgås.josefsson.org/raksmorgas.jpg

heycam: That's the test I meant to do... oh wait it doesn't have the brackets either
... I tried to link to that url in background-image
... I think the css spec needs to say that the things in the brackets are interpreted as iris or whatever they need to do to take non ascii characters

krit: Can you bring it up with the css group?

heycam: well I sent the mail, I think if you want to bring it up you can
... back to the topic. I think the question is whether we allow commas or spaces

<heycam> https://svgwg.org/svg2-draft/linking.html#SVGFragmentIdentifiers

heycam: if you look at the grammar, you can see that the part of the grammar that refers to preserve aspect ratio, aspect params is defined in the list beneath and it's not defined properly
... given that spaces are said to be not allowed, we need to define it so that spaces are allowed or we can allow a comma
... I don't have a strong preference

ed: i was wondering about another issue, url escaping, firefox did allow the url escaped spaces to work
... I couldn't find anything in the spec but it would be nice to have it clarified

heycam: I'm not sure what what level it should do that
... is suspect if we look at the html spec it will define it
... there was meant to be a new url spec, but I don't think it progressed much
... I think it was going to define this sort of thing

shepazu: is anyone working on it?

heycam: I think so but I don't know
... I suspect that that spec would define how to encode the fragments and decode them
... that's the level it should be at
... I agree it would be nice to have a link to somewhere that defines how to parse these fragments

<scribe> ACTION: Cameron to look at the url spec or find out what we need to reference to make sure url fragments have a well defined encoding and decoding [recorded in http://www.w3.org/2012/08/30-svg-minutes.html#action04]

<trackbot> Created ACTION-3349 - Look at the url spec or find out what we need to reference to make sure url fragments have a well defined encoding and decoding [on Cameron McCormack - due 2012-09-06].

heycam: Erik or anyone do you have an opinion on what to do with the preserve aspect ratio part?
... I don't have a strong opinion

ed: this is a special case, I'd prefer if parsing was the same as the attributes
... might not be possible

heycam: we can make it possible

ed: that would be easier implementation wise
... it's not hard anyway but consistency would be nice

shepazu: consistency means a better experience for everyone

heycam: are there any other examples?

shepazu: any attribute can have keywords

<ed> requiredFeatures is space-separated I think

<ed> http://www.w3.org/TR/SVG11/struct.html#ConditionalProcessingRequiredFeaturesAttribute

heycam: I think you can't just have an arbitrary name in there or the property would fail to parse
... I was wondering about non property attributes
... I don't know any off the top of my head
... I'm concerned whether we allow commas in preserve aspect ratio whether that will lead to it happening elsewhere
... there may not be any other cases

shepazu: there's the text, they're not keywords, they're values. Like x or y for text

heycam: we do allow commas there

shepazu: there was a kerfuffle with stroke dash array because it wasn't specified
... we fixed that

<shepazu> http://www.w3.org/TR/SVG/attindex.html

heycam: Unfortunately the type of value isn't in that table
... there's things that take number-space-number and they don't take commas

shepazu: number-delimited-number should allow anything
... should allow comma or space or a combination rather
... oh transforms!
... you can have a space separated list of values and I don't think you can use a comma there

heycam: I think you're right

krit: they can be comma or space separated

heycam: they can in the attribute but not in the property

viewbox and zoomAndPan are the only ones I can see

heycam: let me, suggest we allow commas in preserveApsectRatio

<richardschwerdtfe> have to drop

Resolution: preserveAspectRatio will allow commas in the attribute and that part of the view specification

<scribe> ACTION: Cameron to update preserveApsectRatio in view specification to allow commas [recorded in http://www.w3.org/2012/08/30-svg-minutes.html#action05]

<trackbot> Created ACTION-3350 - Update preserveApsectRatio in view specification to allow commas [on Cameron McCormack - due 2012-09-06].

maskType, and camel case elements/attributes more generally

heycam: We might need to discuss this at the F2F

shepazu: I think that we should not capitalise anything

heycam: I think that's what Simon wanted
... at least on new things

<shepazu> CAPITALIZE NONE OF THE THINGS!

heycam: Maybe allowing lowercase everywhere might be the cleanest solution

shepazu: I'm struggling to think of a place where caplitalisation would matter if we had error correction
... is there anywhere where capitalisation would matter?

heycam: it matters from the perspective of making a dom call
... you need to use the correct capitalisation there
... I want to think about this more deeply
... it might tie in with switching modes in the SVG DOM
... like switching to no namespace
... and in that case everything is in lower case, for example.
... I do think it's an issue that we need to resolve.
... or if we keep inventing camel case names we need to make a concious decision to go that way
... we'll continue the discussion at a later point

- DRAFT -

SVG Working Group Teleconference

30 Aug 2012

Attendees

Contents

CSS4 image type as paint value

Filter subregion clipping

Behaviour of feOffset dx/dy values with percentages

Commas vs spaces in SVG View fragment identifiers

maskType, and camel case elements/attributes more generally

Summary of Action Items

Scribe.perl diagnostic output