W3C home > Mailing lists > Public > www-style@w3.org > August 2011

[CSSWG] Minutes and Resolutions Seattle F2F 2011-07-26: Writing Modes, Gradients, Grid, Positioning

From: fantasai <fantasai.lists@inkedblade.net>
Date: Tue, 02 Aug 2011 22:50:49 -0700
Message-ID: <4E38E1B9.20003@inkedblade.net>
To: "www-style@w3.org" <www-style@w3.org>
Writing Modes
-------------

   - RESOLVED: Remove the concept of glyph-size tolerance from this level of
               the spec for text-combine
   - Discussed scaling methods for tate-chu-yoko effects
   - Discussed name of 'writing-mode' property

Gradients
---------

   - RESOLVED: when the UA has knowledge of the output resolution, it's allowed
               to substitute an average color for the repeating color when the
               device does not have the resolution to capture the gradient
               correctly.
   - RESOLVED: Use premultiplied colors for gradients and transitions
   - RESOLVED: Split css3-images into level 3 and level 4 with level 3 containing
               the more stable things.
   - Discussion of removing keywords from level 3 for more work: corner-to-corner
     gradients need a change of definition to work right, and other keywords are
     confusing to some people. No objections to deferring corner-to-corner gradients.
     However, no consensus on removing horizontal/vertical keywords.

Grid Layout
-----------

   Phil summarized changes since March F2F. Brief discussion on 'fr' unit and relation
   to flex() and flexbox.

CSS Positioning
---------------

   - Arron and Rossen presented draft for CSS3 Positioning, which is mainly old stuff.
   - Proposed 'position: page', but some problems with its model were noted.
   - RESOLVED: put CSS3 Positioning draft on dev.w3.org

====== Full minutes below ======

Writing Modes
-------------

   <glazou> jdaggett: can you hear us?
   fantasai: I think there are several issues open right now.
   fantasai: One is the name of the property.
   fantasai: Second is the default orientation at a codepoint level.
   fantasai: The third is whether we're doing context-based resolution of
             punctuation, and if so, how?
   fantasai: The fourth is if we're giving additional controls at level 3
             (beyond what you can currently do with text-orientation).
   jdaggett: Let's start with sylvain's issue.
   jdaggett: I think #3 and #4 are dependent on a concrete proposal for #2.

   szilles: Two more issues are Nat's comments on TCY
   szilles: I think you suggested separating the text-combine prop into one
            that did manual TCY and one that set up automatic TCY.
   szilles: They may want to be over different ranges.
   jdaggett: Why do you need that distinction, Nat?
   <dbaron> (TCY == Tate-chu-yoku)
   nat: Right now we have text-combine:horizontal, then a bunch of values
        for controlling the auto case, I guess.
   nat: Is that what we want?  Just specify it on, and then specify the
        details (do letters, do digits, combine X many)
   nat: And then, if the letters in the span meet that criteria, combine them?
   florian: One case was that, or a date, you don't wnat to have to add extra
            spans around the day and month and year, but rather just one around
            the year.
   nat: The other issue I had was the default implied value.
   fantasai: The values are "none", "all" (tcy the whole span), and then
             various options that control what things are combined.
   <jdaggett> nat's post: http://lists.w3.org/Archives/Public/www-style/2011Jul/0405.html
   <dbaron> http://dev.w3.org/csswg/css3-writing-modes/#text-combine
   nat: I thought if I tried to implement this, it would be easier if I didn't
        thave to worry about having a state machine go through the text.
   fantasai: You can't combine, say "all" and "latin 4" - it's not syntactically
             valid.
   nat: I think some examples would help here.
   nat: So when it's not all, you can do auto TCY in any span.
   <fantasai> ACTION fantasai: Add syntax examples
   <trackbot> Created ACTION-357

   nat: A second point, why do you have, in the alphanumeric tolerance section,
        a "1.1em" figure.  That seems arbitrary.
   fantasai: koji was talking with hyatt about that issue.  Hiragino apparently,
             if you ask for half-width glyphs, won't fit within 1em if you put
             them side-by-side.
   szilles: if that's true, why not just do it when they ask for it?
   fantasai: This is for whether you scale or not.  This number is close
             enough that you usually won't have to scale.
   kojiishi: The idea of scale is to compress glyphs so that they fit into 1em.
   kojiishi: But there are some fonts that get slightly gretaer than 1em when
             you set two half-width next to each other.
   kojiishi: So if two chars are much larger than 1em, authors would likely
             want to compress them.  But if they're just a tiny bit over
             (maybe accidentally, like described here), you'd want to just
             leave them alone.
   nat: Okay.  I think having the built-in tolerance is orthogonal to the
        scaling issue.
   nat: If you're tolerant of the idea that you're using glyphs that dodn't
        fit into 1em, you're probably okay with not scaling, so you should
        just say no scaling.
   kojiishi: Even if you do scale, though, you probably still don't want to
             scale if you're just very slightly over 1em.
   nat: But you're saying to scale it to 1.1em, not 1em.
   nat: You can have an "auto-scale" that has some tolerance in there.
   florian: Can we just have a scale with no tolerance, and let you add the
            tolerance in specifically?
   szilles: If they're images, you really notice the artifacts.  If they're
            outlines you won't notice as much.
   szilles: Can we say something other than "scale"?
   squish/compress/scale-x?
   <bradk> condense, horizontal scaling
   Nat: condensed is bad because it implies a different font selection
   fantasai: compress
   jdaggett: If someone wants this tolerance, they should be involved in the
             discussion.
   florian: I don't think it's useless, but it seems to be going a bit too
            far right now.  Push it to level 4.

   fantasai: The tolerance is used in the scaling (removed now), and also in
             the "used glyphs" section.
   fantasai: Which says to use half/third-width glyphs if available; if not
             available, scale.
   <bradk> Adobe calls it "horizontal scaling" in their programs, IIRC, FWIW
   fantasai: What might happen is that you say you want to use third-width
             glyphs, but it ends up being rendered with a font without
             third-width glyphs.
   fantasai: The default action, then, should be to go ahead and scale, as
             that's minimally invasive to the rest of the page design.
   nat: Is that how it's supposed to work?  I didn't get the idea that things
        had fallback.
   fantasai: You can combine "scale" and "use-glyphs".
   szilles: Why wouldn't you scale to 1em?
   florian: You'd scale to 1em, but you wouldn't scale if you're "close enough"
            to 1em.
   szilles: [an example with "1.3" and TCY]
   <bradk> Is it tragic to allow the glyphs to overflow if they are slightly
           too wide?
   jdaggett: In the original document this was based, there's an example for
             that.
   <jdaggett> original document == jis 4051 spec
   szilles: I know that stuff like TCY'd "1.357" is actually done.
   nat: I think we should just have a "use glyphs" option, and if glyphs don't
        exist, just do fallback.
   fantasai: We can say that if you find the right glyphs, just don't worry
             about compressing.  Assume that the glyphs are the right width.
   kojiishi: If the first font has the glyphs (slightly off the right width)
             but the second font doesn't, the first font would grab the glyphs
             and be slightly off-width, while the second would synthesize and
             be exactly right.
   nat: I think you're far off in edge-cases here.
   szilles: I think it's pretty simple.  If it has half-width glyphs, it has
            half-width glyphs.  Just use them.
   nat: Why are we doing anything if the half-width glyphs don't exist?
   szilles: CSS generally tries to best capture the authors' intent.
   kojiishi: If the author says "digits 3", and the half-widths are slightly
             more than half an em, you'll get slightly more than 1em-width TCY.
   fantasai: Is it possible to know if the font has third-width glyphs?
   nat: There's a feature in OT that the UA can query.
   jdaggett: The problem with these features is that for different features,
             different sets of glyphs are available.
   jdaggett: For third/fourth-width, often only digits are available.
   szilles: What about numeric punctuation (periods and commas)?
   nat: Generally not.  Most only have half-width punctuation.
   szilles: So you could do years and integers, but not decimal numbers.
   nat: Right.
   szilles: I think we just need a clear processing model of when things
            are decided.
   szilles: I don't think we want to say "let's make as tring, and then try
            to compress it".  I think you want to first check for half-width
            glyphs or whatever, and then build out of what exists.
   szilles: So I want a clear statement of what you're looking for from the
            font, and under what conditions.
   fantasai: So what I'm hearing is that we should use the correct glyphs if
             the font has them, otherwise synthesize.
   fantasai: So if I'm doing "122", if the font has third-width glyphs, I
             use them.  If it doesn't, I scale the "1" and "2" glyphs and
             then combine them, rather than making "122" and then scaling.
   florian: Also, the current spec specifies just that they should fit into
            1em, which allows cherry-picking half and quarter-width glyphs
            and combining them, which is not what you want.
   <bradk> Doesn't kerning and tracking affect if two 1/2-width glyphs fit
           into one em?
   fantasai: We can tell if the font has a "third-width" feature, but we
             don't know if individual characters have third-width glyphs.
             So if I wanted "IBM" in third-width glyphs, I'd request those
             glyphs from the font as third-width glyphs, but I may or may
             not actually get third-width glyphs back.
   fantasai: So once I get them back, if they're wider than 1/3em, I need
             to scaled them to 1/3 em.
   fantasai: I think we can make an exception that, for half-width glyphs,
             don't measure and just use what you get.  You may get proportional,
             which may be close enough.
   bradk: Can kerning affect this?
   nat: You don't kern monospace CJK fonts.
   jdaggett: In theory, it's an issue; in practice, you don't ever kern these
             types of fonts.  I don't think we need to worry about it.
   <jdaggett> ... you don't generally kern these *glyphs*
   <jdaggett> i.e. third-width glyphs
   kojiishi: I think it may still be useful to have an option to not scale
             glyphs, as it may be ugly sometimes.
   florian: Perhaps if you say "use-glyph no-scale", you just use exactly
            what's given back by the font, no measurement.  It may be too
            big, but shrug.  Then "use-glyph" will scale if they're too big,
            and "compress" will just always scale the full-size.
   fantasai: Makes sense.
   <fantasai> ACTION fantasai: edit as above
   <trackbot> Created ACTION-358
   RESOLVED: Remove the concept of glyph-size tolerance from this level of
             the spec.
   <jdaggett> whatever you spec out, you need to try it on some simple
              examples and include them in the description
   <jdaggett> the markup should be easy and the implementation should be
              fairly easy
   nat: yeah, you don't want multiple passes to support TCY

   fantasai: So for fonts with propotional, but not half-width glyphs, if
             you say "use-glyphs", should we just use the glyphs directly?
   szilles: I think, since the user can say "no-scale" if they really don't
            want it, you should have a single default that works decently
            enough for everyone.
   kojiishi: The tolerance came in to address the common case where the TCY
             want to use two digits without scaling.
   nat: If we have different scaling values (depending on number of digits),
        it's too complicated for the value you get.
   nat: I agree that it may be useful in some cases, as Koji says, but I
        don't think it's worthwhile to address.
   nat: Koji brings up the possibility that some people may want to scale
        only when it looks terrible, and let it not scale if it's "almost
        there".
   nat: But I don't think it's worth dealing with that explicitly. That
        should be something a UA could decide.
   florian: Or we could go with steve's idea of adding a parameter to
            no-compress, rather than using tolerances
   szilles: I mainly just think we're spending too much time on this.
   fantasai: And the default is the UA tries to do whatever it can to make
             things fit.  No limitations.

   nat: Moving on, it looks like you can omit the integer in "digits" etc.
        and it defaults to 2.  I think it would be better to be explicit.
   fantasai: Okay.
   kojiishi: You're not requiring an integer for "all", right?
   fantasai: Right.
   nat: We should just make "all" be very simple.  It does *all*.
   <szilles> Add Usecases for 2 or 3 digits in auto t-c-y and have an example
             with mix of digits and non-digits; e.g., "1.2"

   fantasai: If you say "use-glyphs" and there's only a single char, the
             current spec says to use a full-width glyph if one is availab.e
   nat: And I think you shouldn't do that.
   nat: glyphs substituation may cause problems (undefined?)
   nat: Are you supposed to scale up the glyphs to 1em, or what?
   fantasai: I'd just use proportional

   florian: I think we've said that TCY and text-transform interacts.
   fantasai: Spec says that any text-transform features are turned off for
             combined text of more than 1 char.
   florian: Then that's fine.  You can use text-transform to get single
            large chars, but have smaller chars for more TCY.

   fantasai: jdaggett you were saying that for text-orientation we should
             write up proposals first and then discuss?
   jdaggett: Yeah, I did some experiments.  Spec says currently that you have
             to use the VR2 feature of the font, but in practice that doesn't
             really work.
   <jdaggett> http://lists.w3.org/Archives/Public/www-style/2011Jul/0401.html
   jdaggett: In the post I summed up the key points:
   jdaggett: Latin will rotate, greek and cyrillic won't, and there are many
             other cases where it's not clear that what exists is what you want.
   <jdaggett> http://lists.w3.org/Archives/Public/www-style/2011Jul/0402.html
   jdaggett: In discussion with Erik Muller, we thought it made more sense to
             specify a property (possibly going into unicode) that says whether
             a character is upright or rotated.
   jdaggett: Then we can use that property to define the gray areas here.
   jdaggett: symbols, punctuation, currency, etc.
   fantasai: That sounds reasonable.
   jdaggett: I just don't think this can be a derived property from the
             existing values.
   fantasai: Agreed - but until it's a unicode property, we'll have to have it
             be derived with a large list of exceptions.
   jdaggett: And then the logic of text-orientation will be to use that
             property, then apply the "vert" opentype feature to the upright
             runs.
   nat: And you do that so fonts which disagree with the feature can achieve
        that?
   jdaggett: Yes.
   nat: I'm comfortable with that.
   jdaggett: This property effectively already exists, but it's just custom
             right now in different tables.
   jdaggett: IE already has it (though clearly out of date; it doesn't do
             non-BMP). Webkit has a similar table, similarly wrong.
   <ChrisL> has Unicode consortium been asked to add this property?
   jdaggett: It would be great to pull together our individual dbs and help
             make a proposal to unicode.
   nat: InDesign has something similar too.
   <ChrisL> (I take what jdagett said to mean "not yet but we will")
   <TabAtkins> chrisl, I believe so, yes.
   jdaggett: I don't think we need to get into the individual heuristics yet.
   nat: When unicode didn't provide two codepoints for something that was
        two codepoints in shift_jis, we have to be able to differentiate.
   nat: Some fonts have different designs for the two.  In some fonts you
        want to rotate, and in others you don't.
   fantasai: We can get info for Japanese fairly easily, but I'm concerned
             about other languages like Chinese where we don't have much
             if any information.
   nat: It's different, and unfortuantely the standards are even more loose.
   nat: We don't use font metrics, but it's very complicated to deal with this.
   jdaggett: to handle that sort of thing, we could call out those specific
             codepoints and deal with those in a separate category.
   jdaggett: So CSS3 can deal with japanese well, but the separation lets
             us make better rules for Chinese and such later.
   szilles: You and I, jdaggett, have proposed being able to provide a delta
            on the unicode table.
   jdaggett: I think it would be better to wait and propose the property
             first, then went through and discussed how you might want to
             modify it.
   szilles: Right; I brought it up as a reminder of the concept.
   jdaggett: Right. it would be nice to eventually let authors say exactly
             what defaults they want ("never rotate english text"), so they
             didn't need to put a bunch of markup around specific areas.
   szilles: It would also allow all the default tables out there to be
            incorporated.

   fantasai: I have a concern that the rest of the draft (the layout parts)
             really ought to be stabilized somehow.
   fantasai: We have a lot of work left to do on text-orientation, but I
             think the layout bits should be locked down and not held back
             by this work.
   fantasai: Another month or two I can handle, but if this'll take us until
             next year, I'm concerned.
   jdaggett: I don't think it'll take that long.

   fantasai: One more item now.
   fantasai: Sylvain was talking about the name of 'writing-mode' property.
   sylvaing: The writing-mode property says it defines block-progression
             (then why isn't in named "block-progression").
   sylvaing: then it says that the writing mode is deifned by three properties,
             one of which is called "writing-mode", which is confusing.
   fantasai: I think we shouldn't change the name, but if you have a better
             term...
   sylvaing: There used to be a block-progression property.
   fantasai: Yes, because at the time writing-mode was a shorthand that
             overrides "direction", which was a very bad thing. We've
             had this discussion a few times.
   fantasai: And it's incompatible with SVG and IE6.
   sylvaing: What is?
   fantasai: We could call it block-progression, but then writing-mode would
             be an alias.
   dbaron: Is this a naming discussion, or a substantive one?
   fantasai: Kinda both.  Given legacy and other reasons, I think the name
             should stay, but I'm willing to change the terminology.
   szilles: I agree that this would be better called "block-progression".
            But history/SVG means that it should stay consistent and be
            "writing-mode".
   <ChrisL> svg also uses the term 'block progression direction' (from xsl) btw
   fantasai: I'd just like to avoid property aliases.
   <ChrisL> but it does use the writing-mode property, yes
   * jdaggett egads
   dbaron: value aliases are much easier than property aliases.

<br/>

Gradients
---------
Scribe: fantasai

   Tab: 3 issues to resolve
   Tab: First one is repeating gradients when the distance between the first
        and last stop approaches zero
   TabAtkins: zw gradients are not a problem without repeating, but is a
              problem for repeating because you can't repeat them on the
              same point infinitely
   TabAtkins: Right now I punt on the issue by making repeating gradients
              required to have minimum 1px width
   TabAtkins: That guarantees we can tile the space like we should
   TabAtkins: Don't know if it's ideal, but it stops the problem of infinity
   * glazou loves when Tab explains that n * 0 is always 0 whatever is n :-)
   TabAtkins: Other possibility that preserves continuity is that when we
              hit zero width, we pick the average color you'd get
   * glazou loves when Tab explains that infinity / n is always infinity
            whatever is n :-D
   TabAtkins: Seems kinda complicated for an edge case, but I don't realy
              care, just want a decision
   TabAtkins: And to be consistent with svg
   Florian: Pixels aren't px
   TabAtkins: I'm fine with saying it's a hairline wide
   Florian: So you do fallback at device pixel
   Brian: WD says to use first color stop
   TabAtkins: Yeah, but that's not continuous behavior
   <ChrisL> device pixels is better. because svg scaling can mean that a 1px
            is really big
   [fast talking]
   Brian: Non-repeating radial gradients folow last color stop rule
   dbaron: But with those you always fill the area outside with that color
   TabAtkins: As you approach the limit, you get that result
   Brian: So it's continuity issue.
   TabAtkins: I'm ok with either way, whatever way implementers would like to
              do, let's do
   shepazu: SVG, when it hits zero, says that nothing is rendered
   shepazu: That kinda punts on that
   shepazu: I don't know that SVG says anything about approaching zero
   <ChrisL> it does not, no. its either a zero bounding box or it isn't
   shepazu: The averaging thing doesn't seem like what ppl want to do
   fantasai: I think that makes the most sense; as you zoom out that's what
             you'd get
   shepazu: It'd be mixed with other color.. start color end color harmonious
            with page
   TabAtkins: If you do 1px red blue and repeat it, it looks purple. If you
              make it smaller, can assume it'll look purple
   Brian: 1px seems too large to me. But making the spec normative at 1px,
          but allow better resolution
   TabAtkins: So the clamp is at 1px or below.
   <ChrisL> px or device pixel?
   <fantasai> CSSpx
   <ChrisL> eww
   Florian: So this would allow them to go to device pixel, but not force it
   <ChrisL> device pixels is better. because svg scaling can mean that a 1px is really big
   Brian: Do you want inconsistent rendering across zoom
   Brian: If you'll get purple pixels sooner at one zoom level than another
          zoom level
   ...
   * glazou waited for "infinity * 0 is undefined", here we are :-)
   TabAtkins: Gradients are a vector format in any case
   shepazu: If you zoom in until 1px is the whole screen, then what?
   TabAtkins: You're allowed to clamp at a smaller size
   Brad: Makes sense. Like with text you get better resolution as you zoom in.
   TabAtkins: It ends up being a quality-of-implementation issue.
   dbaron: You could just specify that as the length of the repeating gradient
           approaches zero, you should average the color.
   Florian: Need to make sure it's not going to do that at 20px
   <ChrisL> <g transform="scale(1000,1000)"><rect width="1px" height="1px"
                                                  fill="url(#gradient)"/></g>
   <ChrisL> I don't want the 1000 by 1000 device pixels rect above to be a
            solid colour
   plinss talks about billionths of a CSSpx
   dbaron: I think it should be device pixels, because if you zoom out ...
   plinss: Say that the UA can substitute an average color if the gradient
           length is small [...]
   fantasai: Just define what happens at zero. Everything above zero is handled
             by pixel-rounding, which we don't speicfy
   TabAtkins: So should I specify averaging color?
   dbaron: Need to specify color space to average in
   shepazu: So what happens when you zoom in?
   TabAtkins: That depends on implementation
   <ChrisL> @dbaron yes, you do
   TabAtkins: If you zoom enough you hit rounding issues
   <ChrisL> @tab that is always the case. welcome to the world of non-financial
            computing which uses fixed precision arithmetic
   TabAtkins: I want to make sure ChrisL's case is handled
   TabAtkins: is allowed to handle, don't know if I can require it
   ..
   TabAtkins: I can't must without being precise
   shepazu: It's not as important to specify the behavior at 1px when it's been
            defined as 1px, it's what the rendering is.
   <dbaron> I think Peter's suggestion was good.
   shepazu: Important part was ... as you zoom in that 1px width is now 50
            pixels it should have the whole gradient
   shepazu: You can test that -- chris just wrote a test for that.
   shepazu: You haven't been dealing much with things like scale, but you're
            going to be, so you're going to run into this problem
   <ChrisL> css transforms
   Dean: We all know the implementers are going to do the best they can.
   * fantasai peterl should type his proposal
   * fantasai didn't catch the details
   * ChrisL missed the nanoPx also
   plinss: This is very simply specified.
   plinss: You try to not overspecify it.
   plinss: When the UA has knowledge of the output resolution, it's allowed
           to substitute an average color for the repeating color when the
           device does not have the resolution to capture the gradient
           correctly.
   plinss: It lets everybody do the right thing to the best of their ability.
   RESOLVED: Accept plinss's proposal.
   * ChrisL is okay with that
   <arronei> I want my infinite resolution monitor

   TabAtkins: Image values is becoming divergent in implementation stability.
              Some features like gradients widely implemented, others have no
              impls or are just beginning implementation.
   TabAtkins: In the interest of getting gradients unprefixed as soon as
              they're sufficiently stable, I'd like to either pull gradients
              out into a Gradients spec, or go through and kick a bunch of
              stuff into css4-images
   TabAtkins: specifically, the image() function, the cross-fade() function,
              and image-* properties
   <ChrisL> which is larger, the bits you cut out or the bits you leave in?
   JohnJansen: I think pulling out Gradients would make it easier.
   sylvaing: If you move the others, you're still splitting the document.
   <fantasai> ChrisL, about equal
   <ChrisL> oh
   fantasai: There's some things that should move just as fast as gradients.
             Also you need to define <image> type so we can reference it.
   TabAtkins: CR can reference WD, so it wouldn't delay CR, just REC
   TabAtkins: I just want to get Gradients to a point where we can kill prefixes
   TabAtkins: I don't care either way; WG decide for me.
   <sylvaing> http://dev.w3.org/csswg/css3-images/
   TabAtkins: So, we would keep 4.1. 4.3 is implemented in Mozilla, so keep
              that there. We can test it for now.
   TabAtkins: 4.2 and 4.4 would go
   TabAtkins: Gradients stay
   TabAtkins: 6 stays; object-* implemented already, rest is image sizing algorithm
   TabAtkins: Rest can be punted to level 4
   <ChrisL> I suggest splitting into css3 gradients and css3 images. because
            css4 images sounds like its years away, just from the name
   <fantasai> ChrisL, we'll publishing Selectors 4 next month ;)
   <ChrisL> my point about perception stands, regardless of any actual facts :)

   dbaron: Things without 2 impls should be at-risk
   TabAtkins: 9 is a separate issue we need to discuss
   smfr: Seems we need cross-fade() to interpolate gradients
   Florian: We could pull back into 3 if it stabilizes before CR
   <ChrisL> @smfr you mean to interpolate as in a transition, not to draw
            the gradient?
   plinss: Any objections to splitting along the lines described?
   <smfr> ChrisL: my point is that interpolation of gradients and images
          should be in the same spec
   RESOLVED: Split css3-images as described

   TabAtkins: 3rd issue is about gradients directly
   TabAtkins: I've been messing with keyword definitions for gradients
   TabAtkins: Right now you can specify linear gradients by keyword or angle
   TabAtkins: Some very informal polls on twitter, if you ask someone what a
              gradient with top should do, they agree with the spec
   TabAtkins: If you ask them about angles, they agree with the spec
   TabAtkins: If you ask them both at the same time, they say they should be
              the same, despite that being inconsistent with their previous
              answers
   TabAtkins gives statistics for his sample size of < 20
   TabAtkins: Given this, I think the angles are very easy to see, but the
              way keywords work right now is confusing
   TabAtkins: So I'm thinking we should drop keywords and work them out
              better for level 4

   <TabAtkins> data:text/html,<div id=one></div><div id=two></div><style>div {margin: 50px; border: thick solid black;width: 
520px;height: 300px;}#one {background: -webkit-linear-gradient(top left, red, white, blue);}#two {background: 
-webkit-linear-gradient(-60deg, red, white, blue);}</style>
   TabAtkins: We have an issue that corner-to-corner gradients, for example,
              are not sufficiently pretty.
   <bradk> here's mine: http://www.bradclicks.com/cssplay/linear-gradient/corner-to-corner-gradient.png
   TabAtkins: The top one is what corner-to-corner gradients currently do
   TabAtkins: The gradient bands are perpendicular to the diagonal gradient line
   TabAtkins: But the lower picture is what people actually expect/want
   TabAtkins: The two are different by the angle being reflected over the
              line y=x
   <dbaron> data:text/html,<div class=one></div><div class=two></div><style>div {margin: 50px; border: thick solid 
black;width: 520px;height: 300px;}.one {background: -moz-linear-gradient(top left, red, white, blue);}.two {background: 
-moz-linear-gradient(-60deg, red, white, blue);}</style>
   TabAtkins: one is 30deg other is 60deg
   <dbaron> (for those who don't want to use WebKit to look at it :-)
   Brad: Another way to say is that for corner to corner, the hypothetical
         midpoint connects the other two corners
   TabAtkins: With the current spec, the further the rectangle is from a square,
              the more it looks like a sideways gradient rather than a
              corner-to-corner one
   TabAtkins: Given this issue, I want to drop keywords for now and address
              them in level 4

   bradk: Keywords are the most popular way of doing this right now, mostly
          vertical or horizontal gradients.
   Brian: Instead of up top and bottom, use upwards and downwards and instead
          of left and right use leftward rightward
   Brian: And remove the paired keyword corner options
   plinss: If we do partial keyword option, then we're locking in our syntax
           and it'll be incompatible with our future syntax.
   plinss: We'll get into a situation where we dislike our legacy keywords
           and can't change them.
   TabAtkins: Could use a different function
   dbaron: Authors really want gradients. At some point we need to stop fiddling
           with it and declare it ready.
   anne: We did that with border-radius, and by the time we got it sorted out
         they weren't popular anymore. :-)
   <arno1> We could just have "horizontal" and "vertical" as keywords
   Florian: We're not throwing out all of it, just part of it
   Florian: just the part we don't agree on it.
   TabAtkins: We solve the cases we know we need now, get them unprefixed,
              then see if the remaining cases are useful enough to care
   Brad: the only gradient generators online use the keyword types right now,
         and because we changed the angle definition, you can't get something
         that works
   Brad: People use prefixed versions for years.
   TabAtkins: We're not going to change prefixed versions
   Brad: Because we changed meaning of degrees, you can't make a backwards
         compat version
   TabAtkins: You write the prefixed versions with the old syntax, do
              unprefixed version with spec version
   various try to explain that prefixed versions aren't going to change;
   they'll change when they drop the prefix
   plinss: How vendors deal with them and their customers is up to them.
   Brad: I don't think we can pretend what we decide here is not going to have
         ramifications. If we don't have a committment that they won't do that...
         otherwise we wind up with something unusable
   TabAtkins: That's why nobody will do that.
   dbaron: We've done it plenty of times, but we probably wouldn't do it with
           this one
   Anne: It worked fine for border-radius, -moz-opacity, etc.
   Brad: I'd like to keep the keyword going and figure it out before we publish
         it.
   Arron: Do you have a problem with upwards/downwards/etc
   Brad: As plinss pointed out, it'll make it harder for us to get a consistent
         syntax later when we add corner-to-corner gradients
   TabAtkins: The future has potential. I'm confident I'm not blocking my
              ability to expand in the future.
   Brad: I don't see the syntax as being problematic.
   Brad: I don't think it needs to change.
   * fantasai thinks Tab's questions were skewed
   Brad: It's intuitive for a lot of people, and they've learned it, and do
         we really want to hold up the spec for this.
   ...
   smfr: I'm concerned about having multiple gradient functions. Do I need
         linear-gradient or corner-gradient?
   smfr: This will just add to the confusion.
   TabAtkins: I think we can have a syntax that's different enough there won't
              be a parsing ambiguity
   Brad: "from top" would be close to current syntax and resolve ambiguity
   Brad: "left" -> "rightwards" is more different
   plinss: The question was about whether to drop the keywords, not about
           what they should be
   plinss: So do we drop the keywords, or go offline and figure it out
   TabAtkins: I want to drop the corner keywords
   plinss: Sounds like we're don't have consensus
   plinss: Take it to email/telecon

   plinss: Any other gradient issues?
   Brian: Premultiplied? Keep it don't keep it have options?
   TabAtkins: I think premultiplied produces most attractive gradients in
              common cases blending with transparent.
   TabAtkins: I can go either way, depending on implementers
   fantasai: If you go other way, then make 'transparent' magic.
   TabAtkins: I don't want to make transparent magic.
   Brian: Drop 'transparent' keyword, make everybody use rgba()
   Tab: If we're going to go non-premultiplied, then disallow use of
        'transparent' and make everybody use rgba()
   fantasai: That's horrible. You're making the author do the work.
   Brian: You have the same problem with transitions.
   fantasai: gradient from color to tranparent is very common use case
   smfr: Reality is that if we go with premultiplied we won't have it
         working correctly on Mac for another year or so
   anne: If it's just a year, I say head for the future.
   smfr: It could be more than a year.
   Vincent: Example of where premultiplied looks better?
   <nimbupani> http://jsfiddle.net/rK9Pd/11/show/
   <nimbupani> (check in webkit vs opera)
   TabAtkins: yellow to transparent, in premultiplied goes from yellow,
              fading to transparent
   TabAtkins: in non-premultilied, goes through an ugly grayish greeny color
   TabAtkins: If you go red - transparent - blue, if you want to make it
              work correct in non-premultiplied, you have to write red,
              transparent-red, transparent-blue, blue, placing transparent-blue
              and transparent-red at the same spot in the gradient
   plinss: Why don't we want to do premultiplied?
   TabAtkins: It's not natively available on some platforms
   smfr: We'd have to get libraries to add it
   Brian: Alan Gresley was asking about non-premultiplied gradients because
          there are some cases that you'd want that result
   TabAtkins: You could simulate it by manually arc it through the color
              space, it's tricky, but doable, and a very uncommon case in
              comparison
   Someone asks who supports premultiplied today.
   Brian: I had good results with Opera
   TabAtkins: With Opera and IE we'll have 2 impls, so let's stick with that.
   RESOLVED: Use premultiplied colors for gradients and transitions

   shepazu: At the outset of this effort, there was discussion about remaining
            compatible with SVG  gradients.. was that abandoned?
   TabAtkins: No. SVG doesn't have alpha colors.
   shepazu: Talking about geometry
   TabAtkins: Yeah, I don't think that's useful. But I was going to talk
              tomorrow about using SVG paint servers in CSS or CSS gradients
              in SVG.
   shepazu: So an engine supporting both will have to support two different
            things.
   TabAtkins: Yes.
   plinss: Ok, let's discuss that tomorrow. Grids.

   * ed wonders what this meant: "TabAtkins: No. SVG doesn't have alpha colors."
        (most if not all browsers already support rgba/hsla syntax in svg
        colors/gradients, also there's stop-opacity in svg gradients)


CSS Grid Layout
---------------

   Phil: My name is Phil and I'm from MS.
   Phil: We recently published new editor's draft. Hoping none of it's
         controversial so we can go through it quickly.
   Phil: With one possible reduction in functionality in grid template property
   Phil: specifically, assinging display types to grid pseudos
   ACTION Phil: Post notes to www-style or www-archive so they can be put
                in the minutes

   <smfr> http://dev.w3.org/csswg/css3-grid-align/
   Phil: 7.2 covered explicitly defined grid cells, creating named grid cells
   Phil: You could place children of the grid into them. We're removing that
   Phil: Removing grid-stacking property, which said which layout this
         explicitly-defined grid cell would be using
   Phil: The other mode it had was layer, which was the default layout type
         for a grid cell so items would layer on top of others
   Phil: When we presented that at MV ...
   Phil: Talked about creating flows inside a grid cell
   Phil: And assigning display types to the grid cell
   Phil: This was about creating presentation-specific structure through
         declaration of these grid cells, trying to remove concept form the
         grid layout spec
   Phil: We also added new paragraph about grid cell concept. We still have
         logical notion of a grid cell, but it's just an alias syntax for
         referring to a region of the grid.
   Phil: But we're saying it's not stylable
   Phil: It's just a way to name a spot on the grid

   Phil: In section 6.4 , repeating columns and rows. We got some feeedback
         that the named lines syntax we had didn't make sense inside repeating
         syntax
   Phil: Since it defined that only the first occurance of the name would be
         honoree, and purpose of repeat syntax is to replay the grid lines over
   Phil: Added issue as to whether to remove that ability from there
   Phil: Any feedback, send it, otherwise we'll remove
   Phil: We also had another isuse on the grammar, for grid columns and grid
         rows
   Phil: So we changed from ... to ...
   Phil: I'll let you figure it out; trust me the new one is better.

   Phil: Section 7.1 anonymous grid cells, we just added some language about
         relation between grid cells
   Phil: Now it's just logical container, what does it do to grid items inside.
   Phil: Define them as containing block
   Phil: And we said how they came into being, added language defining
         dimensions of the grid cell etc.

   Phil: Next is explicitly defined grid cells, 7.2 is gone
   Phil: Defining grid cells with a template is still there
   Phil: Still use the ascii-art syntax
   Phil: Keeping around grid-cell property, just don't have pseudo-element
         selector
   Phil: Can still define grid, and put things in it
   Phil: with that

   Phil: Section 7.5 automatic placement of grid items.
   Phil: It's no longer what we're thinking, so removed note
   Phil: ...
   Phil: Importance of automatic placement, some language about fixup of grid;
         it matches language in flexbox
   Phil: And we noted that if we don't have this fature, the fixup isn't useful
   Phil: Everything just gets dumped into first grid row. We're planning to
         leave in, so this is just an observation
   Phil: I just renamed section 8.1 .. size of grid items.
   Phil: Needs more work; need to specify box model calculations
   Phil: So if going to be stretched, questions like whta if you have a replaced
         element with intrinsic ratio ...........

   Phil: 10 calculating isze of grid tracks. Don't know if anyone is
         implementing, or thinking of implementing, but if so, we've published
         some pseudo-code about sizing these bits of the grid.
   Phil: Some bugs in it still
   Phil: We'll update a few more times before pushing to WD
   Phil: Lastly all these changes captured in Appendix A
   Phil: That is everything we have changed. The biggest piece is obviously
         the inability now to create these grid cells that you can give a
         display inside to control how they layout their contents, does
         anybody have objections to removing concept from the spec?
   TabAtkins: Ok with it, majority of my use cases don't need it.
   TabAtkins: But if I want to use this with regions, I need to now insert
              dummy elements to position them with grid ...
   TabAtkins: I greatly disagree that I should put dummy divs in my doc
   Markus: We think that should be defined somewhere other than grid
   TabAtkins: I hate put junk into my page for the sole purpose of styling.
   Tab tries to explain the difference between semantic markup and stylistic
       presentation to the MS folks
   Steve: Aren't templates a little bit of both?
   TabAtkins: As long as we keep in mind that we might want to do this more
              generalized in the future, then I'm cool.
   <bradk> If the CSS is creating the pseudo-elements, then conceivably more
           regions can be created to accomodate more content.
   Vincent: We have a resolution on this from the morning.
    * fantasai agrees with Tab
   ...

   Alex: One issue that we discovered ... which was that alignment in grid
         is currently different from alignment in flexbox
   Alex: What we discussed yesterday flexbox alignment, we kinda liked the
         idea of what grid is doing now. We're going to come up with a
         proposal that will make something consistent between the two.
   TabAtkins: Don't think they need to be different when both flexing and
              aligning stuff
   Markus: Q for Peter, we experiment with named lines, and ended up with
           very long strings.
   Markus: Is there a way to shortcut this somehow? Authoring becomes awkward.
   Phil: This example uses template
   Phil: It only takes one letter to name a position
   Phil: with named grid lines
   Phil: You end up putting 4 strings for each item that you had
   Phil: It gets a little verbose
   Phil: I think in practice if you have a grid, and have a large number of
         grid items
   Phil: And don't want to renumber them, you probably won't use named lines
         anyway
   TabAtkins: What wins if you use grid-rows, grid-columns
   Phil: If the grid cell exists, then it wins
   TabAtkins: What properties apply to grid template?
   ...
   Phil: You can only use ascii letters in grid template
   Florian: Disallowing punctuation doesn't disallow Chinese characters or
            accented characters...
   TabAtkins: Most reasonable layouts won't use 26, but last year I was
              looking at some very complicated layouts that were pushing it
   Arron: I had 180 in a layout I did
   TabAtkins: I think you should say any single character
   fantasai: grapheme cluster
   ...
   <dbaron> I wouldn't allow more than characters that are allowed to start
            identifiers.
   Phil: Would you use numbers ... ?
   dbaron: That includes most of Unicode except some ascii punctuation and stuff
   plinss: Where it gets unweildy with grid names, it's not so much where
           they're defined as where they're used
   <dbaron> template: "北北北北" "西中中东" "西中中东" "西中中东" "西南南东";
   ...
   plinss: I'm not concerned about the verbosity of the grid columns declaration
   plinss: If you line it up like that, I think it's readable
   plinss: Later when you do the grid-column ...
   Phil: You have up to for lines for positioning, so that's not too bad.
   Phil: But if you have a lot of grid items, that gets quite large
   plinss: Depends on how large your grid is
   Phil: No issues open on it, these are just some comments we got back
   plinss: Maybe you can have something that maps grid cells back into named
           grid lines
   plinss: Right now you can only do that with template.
   plinss: What if you could define a grid-cell "foo" defined by these named
           lines
   plinss: You're not defining the grid template, so not single-letter names;
           use multi-character names
   Phil: We had ::grid-cell() pseudo that could do that
   Phil: It was styleable enough that you could specify the  ...
   Phil: Ultimately we pulled that out
   Phil: Anything else?

fr unit and flex
----------------

   TabAtkins: Alex and I were talking about dropping the 'fr' unit and just
              using the flex() function for flexbox
   Alex: Flex function can do everything fr unit can do in flex, but fr unit
         is just a shortcut for one of the use of flex function
   <dbaron> Isn't 'fr' a shortcut for one of the uses of the flex() function
            that requires all three arguments?
   Alex: In a grid fr is a basic of how it is calculated
   Alex: it would be .. also allow flex function in grid, in which case you
         could replace fr unit with it or have both
   Phil: right now we have minmax function, and can mix that with min-content,
         max-content keywords
   Phil: Would we introduce flex() function there?
   Alex: flex function has flexibility and preferred size
   Alex: ...
   Alex: Auto and flexibility at once.. produces the same calculation ..
         maximum, however preferred size of column based on content
   Phil: There's just an issue we haven't resolved in the spec now
   Alex: For flexbox you don't specify 50 values for an items, so flex()
         being longer than a unit is not a big deal, but on a grid probably
         wouldn't want to replace fr unit with flex
   Alex: Make it even longer
   Alex: So if in grid definition fr is shortcut for flex(), we could keep
         it for flex
   TabAtkins: I'm uncomfortable with two of us using different notions of flex
   dbaron: fr right now is equivalent to flex() on a basis of 0, and 0 is
           not the default for that third argument
   TabAtkins: I changed flex() ...
   TabAtkins: If you say flex(1), you start from zero
   TabAtkins: That's not what my spec says right now
   Alex: the omitted preferred width should default to initial value
   TabAtkins: Ok, we need to discuss this
   dbaron: I agree with Alex, but I think that's another reason to keep fr.

CSS Positioning
---------------

   Arron: We submitted our proposal for floats and positioning, and we decided
          to split those. Alex volunteered me to do the editing
   <arronei> http://www.interoperabilitybridges.com/css3-positioning/
   Arron: So you can take a look at it.Not too many differences from 2.1
   Arron: Most of it's what 2.1 says, with some little clarifications of
          handling our new value. So you won't see a lot of differences.
   Arron: Will jump straight to some changes.
   Arron: So our first change that we have here is page positioning.
   Arron: What this defines is a way to go straight for the initial containing
          block, much like fixed positioning does but without fixing on the page
   Arron: So size and positioning based on initial containing block
   Arron: It'll go all the way to the initial containing block, and positioning
          off of that box.
   Arron: That's the first part.
   smfr: How is it different from position: fixed
   Rossen: If we create a region each the size of containing block, you want
           to have one positioned to the current region
   Rossen: Don't want to rely on how many relative and abspos elements in your
           ancestor chain
   Rossen: Different from position: fixed because in scrolling media, it's
           fixed to initial contianing block. So element scrolls along with
           the rest of the content
   Rossen: Similar to abspos content that has no positioned ancestor
   Rossen: Whereas fixed replicates
   Rossen: Position page was specific to design so you can target only the
          current page, and unlike position: fixed; position: paged;
          elements ... you can position negatively, and you can overflow,
          and that's just fine, whereas fixed position that just clips
   dbaron: So it's positioned on the current page relative to its placeholder.
   fantasai: My concern is that you'll get a layout that makes sense on
             paged media, but breaks in scrolled. Opposite problem of fixed
             positioning.
   fantasai: e.g. if you use this on a 15-page document, positioning 15-20
             things throughout the document across the 15 pages, looks fine
   fantasai: Load that into a scrolled view, and everything piles on top of
             each other in the first screenful, and then scrolls away with
             nothing in the rest of the document
   ...
   Alex: We should have discussion of what are different aspects of paged media
   Alex: Some means it's non-interactive, you're on paper
   Alex: Some of it means you're paginated
   Alex: From pov of layout, it's tempting to apply paged media to regions.
         But it's not paged media
   Alex: Should we have a new media type?
   fantasai: I think so.
   Arron: What I'm trying to get here is, we have this defintion right now,
          it may need more work as fantasai points out.
   Arron: What I'd like to do is put it up on W3C as an actual editor's draft,
          so we can start to shape something that works
   plinss: Anybody objecting?
   RESOLVED: Put CSS3 Positioning draft on dev.w3.org

   Arron: That's pretty much it. Please review once it's up
   plinss: Some flexbox issues to get back to?
   TabAtkins: Haven't had time.
   Alex: Directions and alignment; need to ...
   plinss: Meeting closed.
Received on Wednesday, 3 August 2011 05:51:21 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 17:20:43 GMT