SVG WG London F2F 2014 -- 22 Aug 2014

Agenda review

<birtles> scribenick: birtles

<scribe> scribe: birtles

ed: one suggestion was to split up the spec editing over all the days
... e.g. one hour per day for editing instead of a full day

shepazu: just a suggestion, e.g. 4-5pm

heycam: maybe not the end of the day
... just after/before lunch?

shepazu: after lunch is fine

(shuffling of PM agenda to put spec editing after lunch)

ed: any other topics missing from agenda?

krit: I'd like to give a progress report about CSS layout properties
... and discuss fill/stroke properties
... any day other than today is fine

Tav: I'd like to get some review on my talk for the conference if there's time at the end

(conference = graphical web conference)

ed: no one calling in?

heycam: I don't think so

The topic index

https://www.w3.org/Graphics/SVG/WG/wiki/Topics

heycam: one problem we seem to have is that we easily lose track of previous discussion about topics
... we type up our minutes, send them to the ml, tracker picks up issues/actions but apart from that there's no indexing/tracking of topics
... often you want to find out the latest on an issue and you have to trawl through the ml
... so I wrote a couple of scripts to manage links to discussions on particular topics
... they harvest mails sent to the mailing list
... the above link, links to pages from telcon minutes etc.

<heycam> https://www.w3.org/Graphics/SVG/WG/wiki/Topic_database

heycam: if you follow the first link ^
... that's where it's getting its information from
... I've only added a couple of topic lists just to give you an idea of the kind of format

<scribe> ... new ones will automatically get added as we send out more minutes to the list

UNKNOWN_SPEAKER: it extracts out the names of topics and resolutions
... since tracker doesn't do that for us

krit: are resolutions also tracked?

heycam: they're automatically added but they're not automatically copied to the topic page

shepazu: is this a tool that you can make available to other WGs?

heycam: it's on my github but there may be some SVG-specific code

krit: so we have to use the same topic every time?

heycam: that is a problem but you can go back and edit the topic names
... the new minutes will get added automatically and then myself or someone else will adjust the titles so they coalesce

nikos: I've got a similar record on our wiki if you want to use that data for historical records

heycam: I could also run the script on old mails
... hopefully that's useful
... let me know if you have problems/suggestions

shepazu: w3c is looking now at exposing group/spec data through an API
... so this might be another tool
... we're finally abandoning the RDF structures we use internally
... it's quite likely that this information would be useful.. to be able to drill down into topics that a WG discussed
... for example we manually maintain a list of our specifications
... there's no way for us right now to get all our specifications and their statuses

krit: what
... what's the problem with that?

shepazu: it's not nicely maintained, e.g. what WG is producing what specs etc.
... which revision each spec is etc.
... if there's any data you feel you want from w3c then let us know

krit: resolutions are really important

shepazu: I think this is good, and if it's in a wiki that good too
... can we edit to add notes?

heycam: it expects a precise format the moment but I can adjust that

shepazu: we use various bots
... I've never been very happy with the way we handle minutes
... it might be worth fixing those bots

heycam: that's something I've often wanted to do, have links to etherpads etc.

krit: can we add it to svgwg.org as well?

heycam: yes, of course

nikos: it might be good to highlight which telcons have resolutions

heycam: I have this infrastructure to munch emails from the list now

Putting SVG 2 spec on GitHub

heycam: so Dirk asked about this recently...

krit: other WGs are experimenting with this
... HTML is one of them
... it's helping others to contribute to the specification
... some developers would be interested in submitting PRs for specs
... I think we could accept PRs under the W3C license

heycam: I put the WebIDL spec on github and have accepted a few PRs
... but it was a bit unclear if PRs from non-w3c members was ok

krit: you could put a license on the repo

(discussion about how hard it is/isn't to learn GitHub)

heycam: one thing is that the mercurial repo does automatic building of the spec when you push changes
... would that work with github?

ed: you can do that with git

ChrisL: with git or github?

ed: because we have our own server, we can use that

(discussion about how we could set up automatic building with github)

krit: the CSS testsuite runs some scripts on every push

<ed> http://stackoverflow.com/questions/19041220/how-to-run-post-receive-hook-on-github

ed: using webhooks we can trigger actions from github

heycam: if someone's willing to do the work for that I don't see a problem

krit: I don't have the experience yet to do that

ChrisL: who was pushing for that?

heycam: I think Anne was looking to make some changes

ChrisL: is he ok to pushes changes under W3C license?

shepazu: it's not just about him, a lot of people are doing this and seeing good results

ed: would we also push to the w3c mercurial repo?

shepazu: I'd prefer to just push to the github

(general agreement)

jwatt: so we'd dump the w3c repo?

shepazu: MikeSmith has some experience with doing this

ChrisL: are we moving all the specs for this WG?

shepazu: yes, I think so

ed: on a related note, I noticed that web-platform-tests is on github and I suggest we drop the svg2-test repository and switch to web-platform-tests

RESOLUTION: We will move SVG WG's specs from the W3C mercurial server to Github

Tav: are we going to use github exclusively or if we still need our own server

ed: we'll still need our own server to run the build scripts since we can't run arbitrary scripts on github

jwatt: but as for the WG members, they'd just be using github

Tav: and I'd still be able to build it locally

ed: yes, sure

Tav: so all the tools are included too

<scribe> ACTION: Doug to talk to Mike Smith about migrating the SVG WG repository to Github [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action01]

<trackbot> Created ACTION-3638 - Talk to mike smith about migrating the svg wg repository to github [on Doug Schepers - due 2014-08-29].

ed: who's going to look at webhooks from github to get a ping when someone pushes
... so we can tell svgwg.org to rebuild the spec

<scribe> ACTION: Dirk to actually perform the migration of SVG specs to Github in cooperation with Mike Smith [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action02]

<trackbot> Created ACTION-3639 - Actually perform the migration of svg specs to github in cooperation with mike smith [on Dirk Schulze - due 2014-08-29].

jwatt: we need to decide where it lives in github

shepazu: Mike says he's happy to help and it all looks do-able

<ed> https://github.com/w3c/csswg-drafts

ed: regarding tests, there's web-platform-tests

<ed> https://github.com/w3c/web-platform-tests

ChrisL: regarding that, we previously decided we want to work with CSS WG's test infrastructure
... and their tools do lots of linking between specs
... but none of that's going to be on web-platform-tests

<ed> there's also https://github.com/w3c/csswg-test

ChrisL: so we should choose wisely

shepazu: the way that CSS has done things is idiosyncratic to the rest of the consortium
... I think we're probably better off aligning with the rest of the consortium
... I think there will be other tools built around web-platform-tests

heycam: the advantage of doing things in the CSS repository is that Peter Linss is very active, and works very hard to get things going right

ChrisL: he is very active and we do do a lot of work together with the CSSWG

heycam: it seems like the CSS tests are primarily reftests, where do their scripted tests run?

ChrisL, krit: same repo

krit: the JS tests can still have metadata

birtles: the web-platform-tests can have a fair bit of metadata

shepazu: from the perspective of web-platform org is that we're looking for something like icanuse.com
... MDN's equivalent is just a table that someone inputs
... but for webplatform.org we have scripts that do that automatically now
... caniuse is good but it's based on very few tests and we want to be more rigorous than that
... and have a clickable field where you can drill down to individual tests
... so you have a confidence score about how well a feature is supported

ed: do we still want the svg2-tests repository or use one of the others?
... do we want a separate one?

krit: shepherd doesn't care

ChrisL: what you (shepazu) described about drilling down to individual tests sounds great
... but how far away is that

shepazu: what we have now is a JSON structure that generates the tables
... so we can plug anything into those tables we want
... going from there to drilling down is probably a couple of weeks work
... it hasn't been a priority until now but it will be in the coming few months

ChrisL: that's good for developers, but what about the WG who are looking for "how far are we from CR?" etc

Tav: how can I put Inkscape's test results in?

shepazu: I guess you would have to do it manually

heycam: are you referring to the boxes in the CSS spec?

krit: you'd have to ask Peter

ChrisL: I think you'd have to fill in this data and pass a pointer... it's been asked before
... I think the answer is to mail Peter and ask

shepazu: it sounds to me like people are leaning towards shepherd
... for webplatform.org we'd rather the data be all in one place and rather the CSS WG joined in what everyone else did

krit: we can join these test repositories later
... i.e. go to CSS repo first and then merge to web platform later

birtles: can we go the other way?

ChrisL: no

Rozvan: what about tests which are only supported behind a flag?
... we had this problem for crowdsourcing tests for CSS shapes
... users wouldn't know that they had to enable these flags so they'd report incorrect results

shepazu: for webplatform.org we just want to get the results of tests (not create them or run them)

birtles: I know we're doing some work to better integrate web-platform-tests into the Mozilla build system

krit: we have that for the CSS repo as well

heycam: we have some scripts for the CSS repo too

shepazu: I'd like to avoid going to the mercurial of tests systems

ed: what about user-submitted tests?
... do we want to decide a structure for that

birtles: so TTWF normally submits tests to web-platform-tests?

ed: yes

ChrisL: but there are some tests in shepherd from TTWF

heycam: at some point we need to convert our SVG 1.1 test suite to reftests etc. (i.e. automated tests)

shepazu: that might be a good topic for TTWF

(discussion about how to write reftests for things like paths)

ed: another thing to consider about dropping svg2-tests is that it's not being pulled into blink's repo

shepazu: at some point I'd like to add to the agenda discussion about using annotations in the SVG spec
... we should decide between the CSS test repo or web-platform-tests

heycam: I'd like to know if web-platform-tests supports reftests

https://github.com/w3c/web-platform-tests/

scribe: it would be good to use the structure from CSS for reftests and the boxes with test results

shepazu: that's why I'd like to talk about annotations
... since we're planning on building that functionality
... if that kind of feature is useful, it will get added to web-platform-tests

ed: it doesn't seem like we're arriving at consensus over which repo to use

ChrisL: I feel like I don't have enough information to decide

Tav: I'd like to see a demonstration of both

RESOLUTION: We will migrate tests from svg2-tests to either the CSS test repository or web-platform-tests (or both)

(discussion about who to demonstrate the test repositories)

(break 15mins)

<heycam> ScribeNick: heycam

marker, symbol: refX and refY shorthand anchor points: 'top left' 'center center', etc.

Tav: if you remember we decided to extend refX and refY to apply to symbols, like markers
... Andreas Neumann had requested that
... he's also now requested for us to consider adding shorthands: top, left, center
... because this is typically what people doing mapping stuff use

krit: center would be the bbox center?

Tav: yes

krit: do we support % on these attributes?

Tav: don't know

ed: they are length in SVG 1.1

krit: which means it includes percentages
... what's it resolved against?

ChrisL: so we could make left,center,right mean 0%,50%,100%
... that's the point of having a symbol rather than g, it has its own viewport

krit: but no viewBox?

ChrisL: yes

Tav: right now it's the viewport that's being used

krit: of the marker or the path?
... for refX and refY, is this % resolved against the viewport of the marker?

Tav: yes
... I wasn't aware that % was already supported there

<ChrisL> actually it does have a viewBox attribute http://www.w3.org/TR/SVG/struct.html#SymbolElement

heycam: if the %s mean what Andreas is requesting, then I would say just use the %s

shepazu: this is what I wanted to roll these all up in one

Tav: I was looking at that...
... these different elements differ enough in their uses that I think it would be hard?

ChrisL: I think there is an advantage to having separate names for these things

Tav: e.g. markers change depending on stroke-width, or not

shepazu: that's sort of like a use? you can change some aspect of it, or not

Tav: there's an attribute that controls whether it scales according to stroke width

<ChrisL> advantage is you can optimise based on typical use eg caching patterns

shepazu: if they haven't different behaviours, that's one thing, but if they have additional behaviours, I think we should try to do this
... consistency of the model, understandability
... bugs, etc. if you only have to do one different thing for this other use...

Tav: I can try to find the work I did looking at this

shepazu: maybe it doesn't make much sense for Inkscape but it does for the spec

ChrisL: if we had a new DOM we could make all of these inherit from the one interface

ed: your question is whether to allow the keywords?

Tav: I didn't know percentages were already allowed

ChrisL: so if we did support 'top left' it would be a shorthand for '0% 0%' etc.?

Tav: it would be
... I'll ask what Andreas thinks about using the percentages

ChrisL: one advantage to the shorthands is that CSS does that for thing like boxes, tiling of background images it has this center top syntax
... but it's just syntactic sugar
... the one thing to remember is to switch off the clipping
... otherwise you only get the first quadrant of your symbol
... don't understand why overflow isn't visible by default

krit: so this is overflow: visible for symbol?

shepazu: by default

ChrisL: the only downside I can think of is if they have hidden things in the other quadrants
... I suggest we do that and have authoring guideliness to define symbol around (0,0) to make your life easier

Tav: Andreas pointed out that in mapping you often have different anchor points
... a tower would be aligned to the center bottom, for example

RESOLUTION: symbol will be overflow:visible by default by the UA style sheet in SVG 2

<scribe> ACTION: Chris to add |symbol { overflow: visible; }| to the UA style sheet and add authoring suggestion to say to design symbols around (0,0) [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action03]

<trackbot> Created ACTION-3640 - Add |symbol { overflow: visible; }| to the ua style sheet and add authoring suggestion to say to design symbols around (0,0) [on Chris Lilley - due 2014-08-29].

ed: what about the keywords?
... transform-origin has these

Tav: that's where I got the names frmo
... it might be useful if that's how it's done in CSS already

ed: I think it's fine to add that

heycam: I would prefer not adding the keywords

ed: do you see refX/refY becoming presentation attribtues in the future?

krit: to x and y maybe

Tav: I don't think they need to be presentation attributes

krit: I think for authors it's more important to have the basic shape geometry attribtues as properties

nikos: I don't see any need to add the keywords; I think the meaning of the percentages are clear

shepazu: no strong feeling

krit: I don't think we should add the keywords

Tav: ok I'll get back to Andreas with that

<scribe> ACTION: Tav to get back to Andreas about refX/refY top/left/etc. [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action04]

<trackbot> Created ACTION-3641 - Get back to andreas about refx/refy top/left/etc. [on Tavmjong Bah - due 2014-08-29].

krit: for overflow:visible I thought we were talking about marker, not symbol

ChrisL: I think we should do it for both

ACTION-3640: Actually this should remove the existing UA style rule to set |overflow: hidden|

<trackbot> Notes added to ACTION-3640 Add |symbol { overflow: visible; }| to the ua style sheet and add authoring suggestion to say to design symbols around (0,0).

krit: did IE make overflow:visible by default on root <svg> of inline SVG?

Rossen_: it made the most sense

ed: it's what people would expect

Rossen_: I'd expect it to be visible for foreignObject too

ed: per spec that would be hidden since it's a viewport-establishing element

<ed> https://svgwg.org/svg2-draft/masking.html#OverflowProperty

ed: so I would prefer to drop the last bullet point in that section

krit: I would worry about the inner SVG becoming overflow:visible
... otoh I'm surprised that inline root SVG is not breaking anything for IE

ed: I think people are told to workaround it by putting overflow:hidden
... the less special handling for overflow in SVG the better

krit: that's a benefit, but I'm just noting that people are using inline SVG and I'm not sure if that new behaviour would be better
... maybe we can make this change and see implementation feedback?

ed: pattern too?

heycam: no I think that would break things

<ed> The initial value for ‘overflow’ as defined in [CSS21-overflow] is 'visible', and this applies also to the rootmost ‘svg’ element; however, for child elements of an SVG document, SVG's user agent style sheet overrides this initial value and sets the ‘overflow’ property on elements that establish new viewports (e.g., ‘svg’ elements), ‘pattern’ elements and ‘marker’ elements to the value hidden.

<ed> (this is the last bulletpoint in the link I pasted above)

ed: so pattern would remain
... but viewport establishing elements would all go to overflow:visible
... overflow doesn't apply to mask

krit: mask default has this 10% margin region
... this has changed in the masking specification to be auto
... so it's effectively overflow:visible

RESOLUTION: All viewport-establishing elements will be overflow:visible by default, except for root <svg> of SVG whole documents.

ACTION-3640: Plus more, check the minutes here.

<trackbot> Notes added to ACTION-3640 Add |symbol { overflow: visible; }| to the ua style sheet and add authoring suggestion to say to design symbols around (0,0).

variable stroke width

birtles: last time I put forward the proposals for syntax

<birtles> https://www.w3.org/Graphics/SVG/WG/wiki/Proposals/Variable_width_stroke

birtles: the next action coming out was to make a polyfill
... I did the part that does the parsing, but I didn't do the part which actually draws the stroke
... I don't have the time to do that
... so if anyone wants to see the feature progress, it needs your help to do that

shepazu: I've done a lot of that stuff in the past

ChrisL: did you try and it was complicated? or that's just where you stopped

<birtles> https://rawgit.com/birtles/curvy/master/index.html

birtles: I haven't updated the syntax for what we discussed in Germany
... if you look at the wiki page, the property is now called stroke-profile
... but in the prototype here it's called stroke-widths
... so it's not up to date with the proposal

shepazu: I see it supports inner and outer

birtles: asymmetric stroke yes

shepazu: how does it interpolate between the points?

birtles: we were going to try different ways with the prototype

ChrisL: would be a good option for Catmull-Rom
... the wikipedia page mentions a few different forms of Catmull-Rom, some of which are more well behaved than others
... i.e. doesn't produce cusps/loops

Tav: inkscape has something similar implemented

<scribe> ACTION: Tav to ask Inkscape vsw implementor to add Catmull-Rom interpolation to see what it's like [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action05]

<trackbot> Created ACTION-3642 - Ask inkscape vsw implementor to add catmull-rom interpolation to see what it's like [on Tavmjong Bah - due 2014-08-29].

cabanier: there's an Adobe guy who probably would be eager to help with the prototype

<scribe> ACTION: Rik to ask the Adobe guy about working on the vsw prototype [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action06]

<trackbot> Created ACTION-3643 - Ask the adobe guy about working on the vsw prototype [on Rik Cabanier - due 2014-08-29].

Tav: we have cubic bezier, linear, and spiro as interpolations in Inkscape
... shouldn't be that hard to add Catmull-Rom
... I think this isn't the harder part of the problem
... joins are harder

krit: dashes too

[discussion about various corner cases]

<ChrisL> http://en.wikipedia.org/wiki/Centripetal_Catmull%E2%80%93Rom_spline Centripetal Catmull–Rom spline

<ed> [work on actions until 3.15pm]

<nikos> trackbot, close ACTION-3599

<trackbot> Closed ACTION-3599.

<nikos> trackbot, close ACTION-3168

<trackbot> Closed ACTION-3168.

<nikos> trackbot, close ACTION-3170

<trackbot> Closed ACTION-3170.

trackbot, close ACTION-3634

<trackbot> Closed ACTION-3634.

<ChrisL> action-3640?

<trackbot> action-3640 -- Chris Lilley to Add |symbol { overflow: visible; }| to the ua style sheet and add authoring suggestion to say to design symbols around (0,0) -- due 2014-08-29 -- OPEN

<trackbot> http://www.w3.org/Graphics/SVG/WG/track/actions/3640

<ChrisL> <svg><star><star><star><star><star><description>OMG its full of stars</description></svg>

<ChrisL> s/star>/star\//g

<nikos> Scribe: Nikos

<scribe> scribenick: nikos

New SVG DOM

heycam: last time we discussed the DOM stuff, the discussion ended on the point where a couple of people weren't sure about going ahead in this direction
... so we decided to write a polyfill for experimentation
... the polyfill only works in recent FF with a certain pref on
... relies on web components, shadow trees, mutation events, and some other features
... I'll take you through some of the examples that are in the repo
... graphics element causes the new DOM interfaces to be put on elements
... I realised that to avoid breaking the pages that use the old one, we need some syntactic way to opt into the new stuff
... the best way seemed to be to have a different element name for the root
... so that existing content gets the old interfaces
... and new elements can go in html namespace

krit: does polyfill create elements in the html namespace?

heycam: the root graphics element gets a shadow tree. The whole tree gets cloned in the shadow tree with updated names and cases and so on
... and it watches for updates
... there are limitations - events, etc

ed: if you were to have nested graphics elements this wouldn't work right?

heycam: no. inner things are called viewport and outer are graphics

ed: foreignObject?

heycam: I didn't do it
... I rely on html parsing and these elements go in html namespace. there are some tricky bits that require parser changes
... if we had foreignObject it would probably be a good chance to use a different name

birtles: we are going to look at using svg elements in html without foreignObject on Monday

shepazu: rather than be a new element, couldn't the flag be svg element without the ns?

heycam: you don't put the ns already in inline svg in html

shepazu: the key is that things are going in the html ns

heycam: yep

heycam shows some demos

heycam: there are some options for lengths. There are some modes you can set to try different options
... I'm not sure which would be best
... in terms of the namespace. Because we're in html ns. If I want to create a new rectangle I can just use createElement("rect")
... certain things get a lot shorter - interaction with attributes, creation of element

shepazu: why can't we expose these methods on existing svg elements?

heycam: the names are already taken
... currently if you're interacting with a rect you use rect.x.baseVal.value = 100
... if you wanted rect.x = 100, that would work
... but rect.x == 100 wouldn't work
... it doesn't know you want x to be a number
... I have an example showing the different ways of exposing svg lengths - moving-rects-0.html
... this is using the existing dom
... creates 10 rectangles, initially diagonally down the page at 0em, 1em, etc
... then does an animation where a random amount is added each update
... most concise way to set is via setAttribute
... jittering rectangles around means you need .x.baseVal.value to get to a number
... same example using strings. Creating the object and setting initial values is more concise and more familiar for html authors
... doing the numeric stuff is a problem because you need to convert to numeric values
... this is a limitation of reflecting lengths only as strings

ChrisL: could it be reflected multiple ways?

birtles: you can't overload the getter but you can for the setter

heycam: reflecting as strings is one option, but there are a few other options
... next example the properties on assignment can take number or string, but when you get them out you always get the user unit out

ChrisL: so when getting values all numbers are converted to a canonical unit?

heycam: yes
... this is nicer than string only, but you can't use .x to get the string value
... final way you could do it is to have parallel accessors
... e.g. rect.x_px

ChrisL: I don't like that as much as .x.px

shepazu: chaining is a common pattern, I don't know if it's only with methods or if it would also be useful for properties

<ChrisL> no, I was unclear. I don't like either x.px or x_px; prefer just .x returning a number

heycam: not really relevant for properties

krit: I was talking with Dmitry. He doesn't like this kind of magic
... where you set with one value and getter returns a different value because it's in a different unit

heycam: I think that's a valid point and might be reason to pick another proposal over this one
... might be an author expectation that returned value is the same as the set value

krit: Dmitry suggested something like x.getPixel()

shepazu: that's also magical

ChrisL: problem with returning strings is you have to parse them or have utility functions for dealing with them
... I think there's something in css where you can assign but the returned value is a canonical type
... why can't it be the same here?

krit: not the same in css
... internally css uses a canonical type

ChrisL: I think it will align down the road with css

heycam: depends if css goes down the road where you assign one type and get another out
... but they might not go in that direction
... whatever they do, it would be good to follow the same pattern

birtles: I thought the long term plan was to use value objects

heycam: don't think it's going to happen soon

birtles: I don't want us to paint ourselves into a corner where we can't align

krit: I'm not sure if it's magic, as far as I read the proposal from Tab it's pretty straight forward

heycam: I don't see that happening soon

krit: could svg wg push that forward?
... I added new section in SVG 2 with x, y presentation attributes. We could start working on om for css
... I have a fear with any new OM. it's svg specific

heycam: it's not svg specific in terms of the pattern of interacting with objects
... the pattern of exposing all properties as types appropriate to the property

ChrisL: people like that sort of thing

heycam: think there's a lot of antipathy wrt to getAttribute and setAttribute

krit: I support the part of making everything in html namespace
... don't support new graphics element
... it took 12 years for svg to be adopted. Now they have to learn something new

ChrisL: we can't break all the existing scripts

krit: if we decide to use new dom we can't

ChrisL: only way to move forward is to introduce a new element with the new dom
... so you're arguing for breaking all scripting in svg with new dom?

krit: I'm arguing that we shouldn't introduce new dom because we won't do it right this time
... like last two times
... i prefer to follow pattern of new CSS OM
... rather than creating new SVG model
... implementations won't support elements in two namespaces
... it's duplication of maintenance effort

heycam: I don't think it'll be too much work
... will just be an extra line in tables
... I think this DOM thing is actually rather small
... mostly it's accessor things
... which html elements already have code for reflection of attributes
... if we have string or length then that's a bit different than html but it's not too much extra
... plus a couple of methods for dealing with list type things
... so total code to support new DOM is not that big

krit: for webkit it would be a major change because svg animation uses old dom
... and needs to be mapped to new dom as well

ChrisL: I see your point that we need to consider implementer feedback
... if we allow opting in and promote the new dom
... in a few years we can drop the old method
... and break content that hasn't updated

krit: my point is that supporting new svg dom in the meantime with long term plan to go to css om
... means three things we have to maintain at once

ChrisL: css om is 5-10 years away

krit: new svg dom will be same time frame before it is in mainstream use

ChrisL: don't agree

krit: in this case you said we'd break a lot of content with svg dom, but the adoption of the new dom will be lower than you expect

ChrisL: I think as soon as this is implemented people will want to use it - it's simpler, probably more performant, etc
... everyone hates the existing dom
... number one thing people want is the dom fixed

krit: we are seeing most content comes from tools or scripts
... people aren't using svg dom

shepazu: possibly because it's so crap
... the performance hit for marshalling is bad, this might fix it right?

heycam: for lists it should
... I can show an example - transform-changes-1.html
... graphics element, 3 shapes, each has an initial transform and we're doing some script animations
... instead of having svg transform list and all the items in the list
... we have getTransformItems() which returns an array of objects
... each item in the array is a plain javascript object with dictionary
... it's not a live reflection of the thing
... you can push some new object onto the end of the array and it takes effect

krit: that's pretty neat
... but it would just work on svg elements?

heycam: the model I'm thinking of, for elements that have attributes, for each attribute there's some accessor method
... why couldn't we use that on a div
... ?
... if cssom had some nice list method like this (which it doesn't) then you could

shepazu: why do you think it's important that it's a generic one rather than one that works with svg elements

krit: it's important to have a unified model

shepazu: in what way is this not similar to getting the href of an a element in html?
... I think there's a core difference between svg and html content
... html is light on attributes and attributes don't have complex values
... I think this is about as close as you can get with how a html dom would work

ed: looking at an attribute with a complex structure - say path - there's talk of adding path to css
... would this work similarly on path in css?

heycam: I would want the same dictionary values to be used

krit: with the SVG attribute has the least priority in the cascade, you would constantly override the values

heycam: maybe that's an argument for not having accessors on divs

krit: it's the same with svg attributes

heycam: that's true

krit: I do like this api more than what we have on svg currently
... but it should apply to everything that's transformable

shepazu: for transform you have a point

heycam: maybe it would have been better to show an example on path

jet: I was hoping to see the sprite model on svg
... what I'm seeing is svg is more machine generated
... lists would have many elements and iterating over isn't pratical
... what I'm seeing is that you can't pick out a particular object

heycam: one step in making it faster is the 'willchange' property
... in terms of the dom memory size issue, if you're not going to make modifications, it's been suggested to take a snapshot of the tree as a raster and throw away the tree
... it's kind of possible now with some work using Canvas

shepazu: there's a problem with that technique because you lose quality if someone zooms or does anything similar
... though perhaps you could just reconstruct

heycam: depends if you've thrown away the subtree

birtles: you could use an image element with a data uri to generate a sprite and throw away the dom tree

heycam: it's a bit like the original promise of use

shepazu: and by having params you could optimise on what would change and what would not
... so you could decide what bits to throw away and what to retain
... I hadn't considered that aspect, but params along with variables could be used as an optimisation hint
... one of the interesting things about svg is that you don't just have a static image - you have something you can change
... that is an advantage over rasters

jet: I'm not just talking about performance, but usage
... vast majority of svg isn't programmed
... so I'd like to be able to opt in to enabling programming on svg content

shepazu: you could specify that on the graphics element

jwatt: you could probably determine that programatically
... you'd need to take note of percentages and layout
... it's kind of the thing that you don't want an author to decide

heycam: if it's inline svg markup you need to have the dom because changes could come from outside

jet: programmable graphics is useful, but we're making assumptions about the dom that come from html
... which is mostly hand written, but svg is generated
... author just wants a scalable image, but it's slow because all the stuff going on under the hood to support the dom

heycam: you can already make that decision as an author by placing the image in line or not
... if you're doing inline svg you've made the decision that the dom needs to stick around

krit: if you scale the webpage, is there a performance difference between scaling an inline svg or an image svg?

Rossen_: you will see differences if you use percentages

krit: if percentages aren't use will images always be faster in IE?

Rossen_: yes
... which goes back to the point before about using the svg as an image

krit: I'm not sure about WebKit, what's FF like?

birtles: we have optimisations on img that will make it faster
... I think we've worked out a way to reduce the cost of the DOM if that's what you want

krit: is there a cost on the dom other than memory?

jwatt: for us there would be - e.g. hit testing on elements

krit: for re-rendering only the render tree would be used (in WebKit)

heycam: back to the polyfill

Rossen_: there was a css value proposal that Tab had - I liked that he hid all the properties behind a CSS object

krit: that's the css om

Rossen_: have you thought of applying the same approach here?

heycam: I think we considered at one point doing something like giving different united access
... e.g. element.px.width

Rossen_: I liked it all being behind css
... related to dirk talking about transform, it's clear which you're working with

<Rossen_> link to Tab's proposal http://www.xanthir.com/b4UD0

ed: we are moving more svg properties to css om, there's less unique svg attributes (or objects) in the dom

heycam: for everything that has an attribute on an element, there's a dot something that lets you access it
... for consistency with html there should be some way of doing that

krit: not sure if we need to be consistent with html at that point

heycam: I think that's a consistency that hasn't been broken in html
... an author can always guess how to interact with an object

ed: I'd rather break the existing dom, phase it out and upgrade what we have

heycam: to achieve that new objects shouldn't have access to the old dom

shepazu: I'm don't think there's too much content using the current dom

heycam: we get bugs filed all the time when things change

krit: there's lots of differences now between Blink and WebKit and we don't get many bugs filed on it

shepazu: I'm skeptical that there's content using the specific part of the dom that we would break
... some content would break obviously, but is there a significant amount?

heycam: didn't we do a search last time and we found that d3 uses some

krit: just transform, with a fallback

shepazu: I think we talked last time about reaching out to those library authors and seeing how they feel

birtles: it's all the sites that have installed it and aren't updating it

ChrisL: I'd like to take up your point, get Dmitry's input for example

shepazu: that's not what I'm talking about. I'm talking about the proposal not having this new graphics element

ed: are there any other ways of opting in?

heycam: I couldn't think of any that are reasonable

shepazu: I'm suggesting that we break backwards compatibility
... just say rather than maintain these two paths, just break it

krit: speaking for WebKit and Apple, we would do it

jwatt: if you do it and get away with it I'd be happy to do it

shepazu: put out a build and see if people complain

krit: we can't break getPathLength, but we can break animval stuff

heycam: I agree that if it's possible to completely replace things then that's the best solution
... but I'm cautious about doing that

shepazu: there will be content that breaks, old documentation that will no longer be applicable
... but I suspect most of that content is not content that is being used
... a lot of the content would be from the old days
... FF broke almost all SVG content when it insisted we include the ns in the svg root
... I think we should be aggressive and try to change it and see what people say

ed: as long as there's an alternative to the functionality

shepazu: you can do it with a shim

ed: that would be difficult with animval

heycam: 1. whether or not new accessors are what we want

2. is backwards compat what we want?

heycam: I still think we should have the nice accessors, but it means we can do it in stages

shepazu: I would be happy with any of the suggestions rather than what we have now
... I couldn't decide on which one we want now

ed: it should be possible to drop parts of the old svg e.g the SVGPathSeg* API

heycam: what do you think about the dictionary things?

krit: everything using dictionaries is more like coding today
... could we map svg namespace to html namespace?

heycam: that's a big dom core change but ....

ed: we had some hacks in presto to let you do that (document.createElement("rect") got you an SVGRectElement if the document had an <svg> root)

heycam: could hack html parser now to put svg elements into html namespace
... and rely on not inspecting the namespace

krit: we got bug reports on that - about inheritance of objects
... right now is the best time to change svg this way

<jwatt> Some modern HTML5 elements that have added numerical DOM properties:

<jwatt> http://www.whatwg.org/specs/web-apps/current-work/multipage/forms.html#the-meter-element

<jwatt> http://www.whatwg.org/specs/web-apps/current-work/multipage/forms.html#the-progress-element

<jwatt> http://www.whatwg.org/specs/web-apps/current-work/multipage/embedded-content.html#media-elements

shepazu: right now people are using svg like an image, in the next few years people will be using it more like something you can program

<jwatt> (so precedent for heycam's proposal)

heycam: I think we have to provide a nice scripted interface. Don't think it's good to require people to use libraries

jwatt: dirk was saying he didn't like having numerical properties - there's precedent for doing that in html5

heycam: I think we have a path to the next step now
... if we can rip the band aid off then that's good
... we would still need to work out the details regarding namespace stuff
... if we're not using a new root element name

shepazu: I wonder if this is something we should talk about at graphical web?
... on the panel thing

heycam: for a couple of the changes here, can you change href and classname to a string?

shepazu: you should also allow href without the xlink

krit: we also need to talk about image vs img

heycam: that was one of the wrinkles with this
... any way we go about it it's going to require changes to parsing
... maybe we should allow img
... we are already switching to object-fit and object-position
... so to summarise the plan:
... 1. can we just remove the existing svg dom? dirk will try
... 2. if so, various parts of this proposal don't need to be done because we don't need to opt in
... we should give people whatever nicer options we can as soon as possible, but it doesn't have to be all at the same time
... 3. if it fails then we re-evaluate

krit: WebKit only releases every year
... need to talk to someone else for Blink

ed: I think as long as there's use counter data then it might be possible

krit: introduce a compile or runtime flag

heycam: decision about switching that flag?

krit: easy on nightly

ed: we can do it in parallel

heycam: dirk, how close are you to being convinced to doing something like this if the plan fails?

krit: my main concern was having a new graphics element that forces people to the new svg

heycam: we didn't talk about lower casing all attributes

shepazu: it might be advantageous for Blink and WebKit if you could point to this conversation and say this is what the svg wg is interested in doing to gather data

jwatt: we should clarify that we're not removing the entire svg dom
... someone should come up with a list

shepazu: we could give an example - say everything that is reflecting something

RESOLUTION: We will remove SVG DOM attribute accessors pending web compatibility checks

<jwatt> all IDL attributes of type SVGAnimated*

ACTIONS: Erik to add run time flag to disable parts of SVG DOM in Blink

<scribe> ACTION: Dirk to add compile time flag to disable parts of SVG DOM in WebKit [recorded in http://www.w3.org/2014/08/22-svg-minutes.html#action07]

<trackbot> Created ACTION-3644 - Add compile time flag to disable parts of svg dom in webkit [on Dirk Schulze - due 2014-08-29].

krit: I'd be interested in whether MS would remove the DOM if others did?

Rossen_: likely would
... would be in favour of better accessors

heycam: dirk your biggest concern is graphics element?

krit: yes, duplicating code paths

heycam: if we can't do removing and I can come up with another method of opting in without root graphics element would you be happier?

krit: yes

heycam: I'm not sure there is another way but I could think about it

ed: the point is to make existing svg elements inherit from HTMLElement

heycam: yes

shepazu: image element isn't as big a worry as a element as far as conflicts go

krit: it's not used very much
... we'd have to discuss with html community
... most of html community would be happy if we move to html namespace
... not sure they realise the cost though

heycam: I wonder if we're at a point where we decide that we want to put things in the html namespace?

krit: I think we should

shepazu: I'm in favour of it

ed: I'd like it if possible

heycam: I propose that we work towards having all svg elements in the html namespace

krit: do we need changes to the content model if we do that?
... e.g. circle having nested elements?

heycam: don't think changes would be needed

shepazu: think it would lead to changes but they're not required

heycam: I don't want to change the structure of elements

krit: if we decide this is the direction we want to go. we can devote next f2f to resolving the issues

Rossen_: moving to a namespace free world would be awesome

RESOLUTION: We want better accessor methods for list type things. e.g. return array of plain javascript values
... All SVG elements should be moved to html namespace pending further discussion about details of doing that

heycam: in my proposal I said that you can put things in html namespace or no namespace
... with the aim that if you're writing xml by hand you can leave off the xmlns
... seemed like the nicest thing to do

- DRAFT -

SVG WG London F2F 2014

22 Aug 2014

Attendees

Contents