[CSSWG] Minutes San Francisco F2F 2016-05-09 Part II: Counters, Values & Units [css-lists] [css-values]
Dael Jackson <[email protected]>
| Newsgroups | gmane.comp.web.css.general |
|---|---|
| Message-ID | <CADhPm3vWONw==4XPcr-473bbzLDhoqKuYhcNbf9C=EWYcGnSuA@mail.gmail.com> |
=========================================
These are the official CSSWG minutes.
Unless you're correcting the minutes,
Please respond by starting a new thread
with an appropriate subject line.
=========================================
Counters
--------
- RESOLVED: Add new syntax (TBD) to control counter scoping and
consider reversing too.
- TabAtkins will write a proposal for this new syntax.
Values & Units
--------------
- plinss will file an issue with TAG to fix the general pushState/
url reference problem in the DOM.
- RESOLVED: Add hidden state to hash-reference-only URLs so that
they can always resolve locally in presence of
pushState
- RESOLVED: Adapt the element() function so that it can be used to
refer to always-local references
===== FULL MINUTES BELOW ======
Agenda: https://wiki.csswg.org/planning/san-francisco-2016#proposed-agenda-topics
Scribe: Shane
Counters
--------
<astearns> https://lists.w3.org/Archives/Public/www-style/2016Apr/0364.html
TabAtkins: Ran into a fun example, author commented on what he
thought was a mistake in list spec.
[TabAtkins] draws a diagram
<fantasai> Tab writes:
<fantasai> <ol>
<fantasai> <li>
<fantasai> <li>
<fantasai> </ol>
<fantasai> <div>
<fantasai> <ol>
<fantasai> <li>
<fantasai> </ol>
<fantasai> </div>
TabAtkins: interesting thing is if you manually set up counters
TabAtkins: and use the counters() function
dbaron: You're supposed to counter reset on ol:before, not ol. I
think that avoids the bug you're about to describe.
TabAtkins: IT DOES!
[the crowd goes wild]
<SimonSapin> dbaron: even if ::before has display: none?
<SimonSapin> or content:none
TabAtkins: So we should just adjust the spec to say ol:before
instead of ol?
dbaron: Assuming :before exists though.
TabAtkins: Counter established by counter-reset on the first ol
counts its siblings until the first one that resets
but the next ol is a cousin, so it nests instead.
TabAtkins: Few ways around it. (1) instead counter-reset on first
li or on :before. Then scope only covers the list.
TabAtkins: (2) explicitly resetting a counter when it came from
something that wasn't in the ancestor chain then reset
instead of nesting.
Florian: Wouldn't that fail with headings?
TabAtkins: No?
dbaron: Yes.
dbaron: The rule is that a later sibling replaces an earlier
sibling's counter, but not if it's a descendant of the
later sibling.
TabAtkins: If you're using heavy sectioning content then you've
got nesting and it'd probably break.
dbaron: With sections you should be fine, as long as sections &
headings line up.
dbaron: There was another problem, solution was counter-set
property.
dbaron: I don't remember the problem.
TabAtkins: The value attribute on li. Already in spec.
dbaron: I was trying to remember if other problem relates to this
one, it doesn't.
ekimber: Question: what's the use case for having the counter
apply to following siblings?
TabAtkins: Headings. You set up one counter at the top of the
document and run it right down through the whole thing.
If you have a flat document structure where headings
are siblings then you can reset H1 and H2 counters,
which will operate until next H1, next H2, etc.
dbaron: You reset your H2 counter *at* your H1.
ekimber: Given that, response would be "HTML is broken, so what."
ekimber: So this is user error.
ekimber: This is an unavoidable consequence of having to support
this flat list.
TabAtkins: Not necessarily, only if you want to support *all* of
what HTML can do.
TabAtkins: If you have an H1 as a child of body, and a div wrapper
around the H2 that sets up a subsection.
TabAtkins: Then that would break with the suggestion I'm making.
ekimber: Implication seems to be - be consistent, otherwise things
will mess up.
Florian: Also the header element, can use for header of page OR
title with subtitle.
Florian: Not consistent, so nesting will be varying,
esprehn: This change would be observable, right?
esprehn: ol resets the counter. If you put the value attribute on
an li which lets you explicitly number, then there's a
reset there too.
dbaron: That's a set.
TabAtkins: Something something bug.
esprehn: Does anyone implement?
esprehn: We support counter but both value and reset are the same.
So doing this changes the behavior.
TabAtkins: But current behavior is very unexpected
dbaron: <li><li><li value=5> - counter value inside each one,
without set you'd have 1, 2, 2.5
TabAtkins: So yeah it'll fix it to 1, 2, 5, which is actually good.
Florian: Given earlier thing with header, does that rule out
solution?
TabAtkins: Makes it less attractive. Another option: add a
property to counter-reset or another property that
changes scope
TabAtkins: to descendants only instead of descendants & siblings.
TabAtkins: Problem with :before is that it may not get created.
dbaron: I like the new value or property idea
plinss: Putting the :before rule with reset property doesn't
trigger :before?
TabAtkins: Nope.
<zcorpan> "This specification does not yet define the CSS-specific
rules for rendering li elements, because CSS doesn't yet
provide sufficient hooks for this purpose."
https://html.spec.whatwg.org/multipage/rendering.html#lists
zcorpan: Question: is it possible with counter spec to specify how
<ol> and <li> are supposed to work?
TabAtkins: You can as long as nesting isn't observable.
TabAtkins: Except for reverse
TabAtkins: but if nesting is exposed somehow then the naive
mapping is wrong and can't be done in CSS.
zcorpan: I think we should make the counter spec usable by <ol>,
<li>
TabAtkins: Yup.
fantasai: Yup.
TabAtkins: Shall we go with the idea of a new value or property
that scopes only to descendants?
<TabAtkins> ol > :first-child
<zcorpan> including <ol reversed> (ideally)
Florian: but we can just ol :first-child { counter-reset: blah }.
[Tab draws a waaay better diagram]
<shane> https://usercontent.irccloud-cdn.com/file/HwvTalBE/IMG_20160509_115341.jpg
Scribe: fantasai
Florian: So one proposal was use ::before if it exists, or
:first-child of the list
Florian: to reset the counter.
TabAtkins: It works for HTML, but it is weird overall, because you
have to be concerned with the first child who is being
set.
TabAtkins: Setting on the container is much cleaner.
TabAtkins: So one proposal is to cut counter [...]
TabAtkins: Better one from dbaron is we have some syntax for
indicating on counter-reset that counter will only
scope to its descendants.
TabAtkins: The counter would then end its scope at the end of the
</ol>
Florian: Seems a lot cleaner, but worried it will be very obscure
and nobody would know about it.
Florian: Is it better to have a simple system with ::before or a
complex system with lots of switches?
Scribe: shane
TabAtkins: I wouldn't have designed counters this way.
fantasai: One use case that should work - sometimes you start a
list, then close and want to continue later.
<zcorpan> <ol start>
dauwhe: This is super important for us.
Florian: Can't just use a class?
TabAtkins: You need to establish counter in ancestor.
dbaron: dauwhe, does it sound OK if browsers implement list
numbering using counters, you could do it by having a div
higher up that contains all the lists you want to number
together
dbaron: explicitly resetting list counter on that div, then
removing on individual lists. Would this work for you?
dauwhe: That makes sense to me and I think I can explain it OK to
people too.
<TabAtkins> https://drafts.csswg.org/css-flexbox/#main-sizing
TabAtkins: This is explicitly what we use in the flex algorithm.
TabAtkins: Multiple OLs crossing sections that number
subsequently, put a counter on body and use that one
instead.
Florian: Either if we have a simple but surprising system, or a
complex system, people aren't going to understand and
will need to look at stack overflow.
TabAtkins: Can put into UA stylesheet and make prominent in spec
TabAtkins: if you're doing headings, do it like this, etc.
Florian: Probably cleaner with complex but unsurprising system.
fantasai: If it's a syntactic switch then it'll be in the spec.
fantasai: People who are using counters are going to look it up.
It'll be fairly obvious
fantasai: so add a keyword to counter-reset?
TabAtkins: We can't extend counter-reset. It's multi-valued
without commas between each value.
TabAtkins: Can't tell whether a keyword is a second counter or not.
Florian: So multiple counters, might want to scope each one
differently. So need a list in the parallel property too?
fantasai: Can have a second property which does the same as
counter-reset but with the other behavior. OR a property
that switches the behavior.
dbaron: Want other options too, in particular reversing.
TabAtkins: Currently can't be expressed in the counter model.
RESOLVED: Add new syntax to control counter scoping and consider
reversing too. (Syntax TBD.)
<dbaron> (Tab to propose syntax)
<zcorpan> (This needs to be directly usable by HTML spec's UA
stylesheet for <ol>, <ol start>, <li value>,
<ol reversed>...)
[discussion about resolutions, resolution reversion, and
contradictions]
<dbaron> (and counter-resolutions)
Values and Units
----------------
TabAtkins: Say you have a background image 'url(#svgfrag)'.
TabAtkins: Clearly points to something local.
TabAtkins: Once you absolutize the url
TabAtkins: you notice it points to same doc, go find it,
everything's cool.
TabAtkins: But say you do push-state to update the URL, the page
changes it's URL
TabAtkins: we don't go an re-absolutize the images
TabAtkins: (in all browsers, interoperably).
TabAtkins: So that's an issue. pushState means the reference is no
longer local.
<zcorpan> history.pushState()
<zcorpan> or insert a <base href>
<SimonSapin> this is an inline stylesheet, right?
<SimonSapin> <style> not <link>
dbaron: Why didn't that change when the document's URI changes?
TabAtkins: Because everyone absolutizes at parse time and nobody
changes it.
fantasai: Can't we fix it?
TabAtkins: No. It's good behavior.
TabAtkins: If the ref is relative to to the slash path then do
pushState that just changes local position in doc tree
would result in local references being probably wrong.
TabAtkins: If you start out at http://example.com and reference
foo.svg
TabAtkins: then pushState to http://example.com/startMyApp/
TabAtkins: then we're now referencing
http://example.com/startMyApp/foo.svg, which probably
doesn't exist.
TabAtkins: So don't want to update in this case.
fantasai: This is super messed up. Could have 2 different refs to
same target that point to different things because of
statefulness of pushState.
zcorpan: I think it's too expensive to re-resolve all the urls in
the universe when a pushState happens.
plinss: If you pushState to a new URL then you should be able to
reload from that URL and get back to exactly the same state.
TabAtkins: In practice doesn't always work, and can use middleware
to fix the problem anyway.
plinss: I'm not concerned about case where middleware masks a bug.
TabAtkins: I don't think it's a bug that you should have to
rewrite references on pushState.
plinss: But it is a bug that you can't reload a pushState and get
back to the same state.
plinss: I want to see data that we can't fix that bug.
TabAtkins: Someone should say we *can* do it first.
TabAtkins: It's an HTML problem.
plinss: It's a cross-spec problem really.
zcorpan: From browser vendor's POV the burden of proof is in other
direction. Won't risk breaking pages without data saying
it won't break pages.
plinss: If it was a minor change that's fine. But fundamental
architectural issue so different stance.
hober: Would love to see data on compat issue, but given that
absent data implementors are unlikely to change, who's
volunteering to collect data?
TabAtkins: I want to move on to issue. Not absolutizing is bad for
local references that become remote.
TabAtkins: So e.g. pushState from http://example.com to
http://example.com/foo. #svgFrag no longer resolves as
local because the URLs don't match.
TabAtkins: My proposal: '#' is a very clear indication that you're
doing a local reference. So if a url is JUST a hashref
then we store state that indicates the reference is
definitely local, and never absolutize.
TabAtkins: If we can solve the more general problem then we don't
need to do this.
plinss: Sounds like a hack on a hack.
* glazou agrees with plinss
[A general rehashing occurs. Rehashing. Hah]
astearns: If there is a compat issue with fixing this the right
way, isn't this going to trigger that too?
TabAtkins: No. This definitely always wants to be local.
Florian: But it's not clear for files because you don't know
whether they want to be relative or relative relative
hober: Are you proposing that hidden state be exposed to CSSOM? Or
via syntax to explicitly opt in? Or via URL object?
TabAtkins: Currently just if URL string starts with '#'. Can
detect by asking for computed value because it should
serialize back to just a '#' instead of abs value for
round tripping purposes.
plinss: In general, TAG is facing another issue where we need to
assign additional information to a URL to make it useful.
Tim hates this. Fundamentally URLs should be standalone,
without magic. This is going to <elided> Tim.
TabAtkins: We could introduce new function that is identical to
url but only accepts local references
TabAtkins: but that doesn't fix existing pages
TabAtkins: or upon parsing this could computed value to local only
function.
plinss: Yuck.
plinss: Shouldn't make an assumption based on presence or absence
of parts of the url.
TabAtkins: There's a couple of options; we need to fix it somewhere.
TabAtkins: (A) get DOM to fix everything
TabAtkins: (B) hidden state on urls
TabAtkins: (C) new local-only url function
fantasai: How is that better than fixing A?
TabAtkins: Because A will break things
TabAtkins: e.g. your URL is "foo.svg" and you pushState to
something with a different URL segment then your page
will break.
plinss: But right now it'll break on reload anyway.
TabAtkins: We should decide on (A) assume DOM will fix everything
and make no change or (B) make *some* sort of change.
fantasai: What about (D) try to get (A) to work and in the
meantime do (B).
fantasai: I prefer (D)
astearns: Me too, but I don't know how fixing the actual problem
is going to happen?
hober: Comes back to: "who's going to provide the data"? Haven't
heard any volunteers.
fantasai: If you do it correctly, relative URLs aren't going to
change unless you wanted them to change. If you're
depending on the broken-ness then it's going to change
and you'll need to fix your URLs.
plinss: I'll file (A) as an issue with the TAG. It's bigger than
CSS.
astearns: Does anyone object to pursuing a temporary solution
while that happens?
<fantasai> Proposed resolution: Make local fragment URLs magically
always relative; file issue against TAG to fix this
generally.
Florian: (B) is a temporary solution that silently disappears if
we fix (A) so it's better than (C)
fantasai: yeah, agreed.
plinss: I'd rather have a different function - i.e. (C)
Florian: Why would you prefer a different function?
plinss: So you can say what you mean. getElementByID should be a
new thing, not the existing thing,
plinss: and the parameter should be an ID, not a URL fragment.
hober: I think esprehn's saying that functionally, fragment
references currently are getElementByID.
esprehn: Right. It's already live in that we can swap out elements
and the reference updates.
zcorpan: Special casing URLs with only a fragment already exists.
Navigating special-cases fragment changes vs. other
changes.
Florian: In this case we specialize fragment behavior to do the
right thing. So fixing behavior in the subcase. i.e. like
(B).
dbaron: One of the other cases for having something that isn't url
function is to be able to move styles that reference an ID
in a document into an external stylesheet.
dbaron: We added an element() function as an image value. Is
basically this.
TabAtkins: element() might be a reasonable choice then.
hober: This is a garbage fire. Browsers screwed up AND authors
screwed up. (B) seems reasonable then but both plinss and
dbaron have made reasonable arguments for adding (C) *as
well*.
ChrisL: SVG has a concept of a URL that's restricted to the local
document. Originally had idref but the TAG told us to stop.
TabAtkins: Current proposal: Add hidden state to URL *and* make
sure element() can be used to just refer to local
references.
hober: I'm a little unsure about expanding element() beyond
existing use as an image source. Probably fine.
ACTION: Peter Linss to file an issue against the TAG to fix the
general pushState / url reference problem
<trackbot> Created ACTION-769
RESOLVED: Add hidden state to hash-reference-only URLs so that they
can always resolve locally in presence of pushState
RESOLVED: Adapt the element() function so that it can be used to
refer to always-local references
<br type=lunch>