00:03
<zewt>
hero ku uses git for pushing code; you need to force push to that a lot too
00:03
<zewt>
also heroku; ios'a new habit of autocorrecting words when you type the *next* word is terrible
00:06
<TabAtkins>
Huh, Android just autocorrets when you hit space.
00:19
<zewt>
that's what ios used to do; now it tries to figure out phrases
00:19
<zewt>
which means you never really know when it'll insert a typo for you
00:35
<Domenic>
TabAtkins: lookin' good. https://whatwg.github.io/streams/
00:36
<Domenic>
Next up I think I need to steal some CSS and/or JS to get the anchor links working
00:36
<TabAtkins>
Happy to help.
00:36
<TabAtkins>
The parts of the CSSWG stylesheet are pretty easy to find.
00:37
<TabAtkins>
http://dev.w3.org/csswg/default.css
00:37
<TabAtkins>
Look for a.self-link
00:37
<TabAtkins>
Also: whatwg specs should use the CSSWG toc styling, it's the best.
00:39
<Domenic>
sweet. will play with this more tomorrow maybe. although at some point i have to consider this yak shaved and keep working on the actual text :P
00:56
<zewt>
yak shaving considered harmful
01:04
<Hixie>
TabAtkins: last time you said that css spec styling was the best, i picked a random css spec, and it looked horrible. do you have a specific spec in mind?
01:08
<Domenic>
was the day April 1?
01:10
<zewt>
one of them was
05:28
<annevk>
TabAtkins: dom-core.html is not too important I guess; https://github.com/whatwg/dom/ should be canonical; I use two definitions of throw, one is from DOM, one is from IDL
07:17
<zcorpan>
Hixie: yoav wants to put a link in his patch to my recent img spec changes. could you regen the spec?
07:21
<krit>
annevk: I am back online
07:22
<annevk>
krit: cool
07:23
<krit>
annevk: I checked mask-image and filter and think that both do not cause any problems if the resources are not allowed to load further resources
07:23
<krit>
annevk: both do not affect layout
07:23
<annevk>
krit: having written down the algorithms it does seem like SVG as image has a lot in common with the other cases
07:23
<annevk>
krit: it's just that rather than returning an element by ID, in that case you simply take the root
07:23
<krit>
annevk: I presented the problem with clip-path to the WG in Tokyo and the WG decided that the attack pattern is not suitable enough
07:24
<annevk>
krit: which WG?
07:24
<krit>
annevk: CSS WG
07:24
<annevk>
krit: what makes them good at security?
07:25
<annevk>
You don't go to the CSS WG for security recommendations
07:25
<krit>
annevk: All I was able to do as spec author is to present the problem, present alternatives and ask for approval… which I did
07:25
<annevk>
You ask abarth, bz, ...
07:25
<krit>
annevk: we discussed it
07:25
<annevk>
WebAppSec copied
07:25
<krit>
annevk: bz was in the discussion
07:25
<annevk>
In Tokyo?
07:26
<krit>
annevk: He did not approve the outcome of the discussion but did not object either
07:26
<krit>
annevk: no, we discussed the issue before the F2F on the mailing list
07:26
<annevk>
o_O sounds like a bad outcome to me
07:26
<krit>
annevk: but irrelevant for the other properties that I care more about right now
07:27
<krit>
annevk: we can bring up clip-path at any time again
07:27
<annevk>
k
07:28
<krit>
annevk: what that means is that mask-image, fill, stroke and filter do not need to check the resource after downloading again
07:28
<krit>
annevk: something that was important for roc
07:29
<annevk>
but you do, no?
07:29
<krit>
annevk: even though he agreed that this is not a big issue if it would be necessary
07:29
<annevk>
you need to at least do a MIME type check?
07:29
<krit>
annevk: no, we don't
07:29
<annevk>
o_O
07:29
<krit>
annevk: well, this is done AFTER downloading the resource
07:29
<krit>
annevk: the MIME checking
07:29
<annevk>
yes
07:30
<krit>
annevk: the policies prevent downloading the resource in the first place
07:30
<annevk>
What policies?
07:30
<krit>
annevk: The CORS policies as implement
07:30
<annevk>
I have the feeling you're compounding several issues onto one
07:30
<krit>
annevk: that is common between blink, gecko and WebKit
07:31
<krit>
annevk: you mean?
07:31
<annevk>
I don't see what CORS has to do with this
07:31
<krit>
annevk: right, lets say fetching policies … where CORS is still part of
07:32
<annevk>
Sure, are you saying fetching mode is CORS for mask/fill/stroke/filter?
07:32
<krit>
annevk: http://src.chromium.org/viewvc/blink/trunk/Source/core/fetch/ResourceFetcher.cpp
07:32
<krit>
annevk: no, “image"
07:32
<annevk>
There's no "image" mode
07:33
<krit>
annevk: that is because implementations use their own terms
07:33
<krit>
annevk: we check what type we expect
07:34
<krit>
annevk: font, image, stylesheet, script, SVG document (for <embed>), media
07:34
<abarth>
i've been reading along, but I need to go to sleep soon
07:34
<abarth>
can you summarize what the CSS WG decided?
07:34
<annevk>
You're not making much sense. But I guess what you mean is that mask/fill/stroke/filter use no CORS as mode and that the CSS WG decided that was okay.
07:35
<krit>
annevk: yes
07:35
<annevk>
Even if the eventual thing was an SVG resource
07:35
<krit>
annevk: right
07:35
<abarth>
that's almost certainly insecure
07:35
<abarth>
you can't load XML across origins without CORS
07:36
<krit>
abarth: why wouldn’t you?
07:36
<abarth>
because XML is a confidential mime type
07:36
<abarth>
these features also let you probe inside the XML for ids and such
07:36
<abarth>
which isn't good
07:37
<krit>
abarth: IIRC we discussed that earlier and you even opened a bug report and closed it right afterwards
07:37
<krit>
abarth: we had this discussion before Blink fork
07:37
<abarth>
do you have a link to the discussion?
07:38
<abarth>
I remember this issue came up before
07:38
<krit>
abarth: it takes me some time to find it
07:38
<abarth>
but I didn't remember the details
07:38
<abarth>
maybe there's some specific thing that makes it safe, but in general its unsafe
07:38
<krit>
abarth: the point is, that you can not do ID sniffing with CSS
07:39
<krit>
abarth: if the resource exists, then you download it… there is no way for anyone to check if the ID was found or not
07:39
<abarth>
presumably you can tell if the mask was applied, right?
07:39
<krit>
abarth: how?
07:40
<krit>
abarth: it doesn’t apply, the url is still part of property style
07:40
<krit>
s/it/the mask/
07:40
<abarth>
sure, but the pixels on the screen will be different
07:40
<annevk>
abarth: for <img> I think we decided it was safe
07:40
<krit>
abarth: yes
07:40
<abarth>
if the pixels on the screen are different, the web site can learn that by interacting with the user
07:40
<annevk>
abarth: <img> does no CORS, and if the return MIME type is image/svg+xml we decode that as a bitmap and render it
07:41
<abarth>
there was a cute proof-of-concept on hacker news the other day
07:41
<abarth>
for history sniffing
07:41
<abarth>
using this technique
07:41
<abarth>
annevk: yes
07:41
<krit>
abarth: that one needed visited pseudo selector
07:41
<abarth>
krit: you don't think you could build the same thing using mask?
07:41
<krit>
abarth: which sadly is not specified enough
07:41
<abarth>
meaning make a game
07:42
<abarth>
where the player makes different clicks depending on if a mask applied?
07:42
<abarth>
annevk: we wouldn't do that today, but we're stuck with it because of the past
07:42
<krit>
abarth: hit testing is not applied on mask, but like opacity, you can “hide” objects
07:42
<abarth>
krit: you're not understanding
07:42
<abarth>
mask changes the pixels on the screen
07:42
<abarth>
users react to pixels on the screen
07:43
<annevk>
abarth: yeah, we could have disallowed the cross-origin case there though and require the crossorigin attribute
07:43
<abarth>
meaning, they react to whether the mask applied
07:43
<abarth>
meaning they tell you whether your ID probe worked
07:43
<abarth>
annevk: exactly
07:43
<annevk>
abarth: e.g. if the response image/svg+xml but also tainted/opaque, act as if there was a network error
07:43
<krit>
abarth: sure you can do that
07:43
<abarth>
hence, you're creating a security vulnerability
07:43
<abarth>
unless you use CORS
07:43
<annevk>
abarth: that seems like what we should do for mask et al
07:44
<annevk>
abarth: if we want to keep the "tainted" cross-origin fetching for non-SVG resources
07:45
<abarth>
https://news.ycombinator.com/item?id=7855168
07:45
<abarth>
is a recent demo of this approach
07:45
<abarth>
you could build the same thing with mask
07:45
<abarth>
to probe cross-origin resources rather than history
07:47
abarth
needs to get to sleep
07:47
<abarth>
security's no fun
07:47
<abarth>
:(
07:47
<annevk>
thanks abarth, I was mostly trying to figure out the overall processing model, but this helps greatly with fetch specifics that we also need to nail down
07:48
<annevk>
abarth: nn
07:49
<krit>
abarth: we should discuss that on the mailing list again
07:50
<krit>
abarth: IIRC the issue was that you can figure out if the ID exists even without CSS.. pure JS
07:51
<krit>
abarth: the URL of the resource is known at the point
08:00
<annevk>
krit: I don't understand the aversion to better safe than sorry
08:01
<krit>
annevk: we always need to find the right balance between security and the expressiveness of authors… We don’t need to make the web as secure as possible but as secure as necessary to protect users
08:02
<annevk>
Yes, and we have repeatedly learned that not using CORS is bad for users
08:05
<krit>
annevk: and I think it is good to have people on the other side that challenge the need of overprotection :)
08:06
<annevk>
Overprotection?
08:09
<annevk>
krit: anyway, I updated the algorithms in the etherpad
08:09
<annevk>
krit: they are now generic enough to cover both SVG as image, and whereever else we might need to take hold of a set of nodes
08:09
<annevk>
krit: if you don't pass in an ID it simply returns the root, otherwise it returns the element identified
08:10
<annevk>
krit: there's various entry points as well, as e.g. the mask/filter/ case will do its own fetching and then only if type is image/svg+xml hand it off to SVG
08:24
<krit>
annevk: ?
08:25
<annevk>
krit: what is unclear?
08:30
<krit>
annevk: no, wanted to know if this is what you would like to see for mask-image/
08:30
<annevk>
krit: yeah, it seems we should actually write out mask image processing
08:30
<annevk>
krit: because that is more complicated
08:31
<annevk>
krit: I also wonder if TabAtkins will be in a more convenient timezone at some point so we can fix this stuff in CSS
08:31
<krit>
annevk: fill and stroke inherit the same behavior… so this could be ironed out for all SVG properties then
08:31
<krit>
annevk: including clip-path and filter
08:31
<annevk>
yeah, the algorithm "To get an image or SVG element given a /url/:"
08:32
<annevk>
it seems that should also take an environment URL and maybe an environment base URL
08:32
<krit>
annevk: Tab is working on more parameters for url() that allows authors to specify CORS rules
08:32
<tobie>
Fwiw, tend to agree with jgraham on the to('json') vs. asJSON.
08:33
<annevk>
krit: sure, again, that is not really important
08:33
<annevk>
krit: what is important is the basic setup
08:33
<annevk>
krit: without additional features
08:33
<annevk>
please try to focus on that
08:34
<annevk>
tobie: talk to JakeA or Domenic I guess
08:43
<annevk>
If people want to follow the SVG discussion, we moved to W3C, #svg
10:12
<MikeSmith>
zcorpan: so, about srcset, I have question that maybe I should hesitate to ask since I'm not sure I want to implement it, but I'll ask anyway
10:13
<MikeSmith>
which is, should it be a conformance error if the same URL is specified multiple times in the same srcset value
10:14
<MikeSmith>
in other words, is there ever a use case for specifying a URL with different descriptors in the same srcset value
10:14
<zcorpan>
MikeSmith: not sure
10:14
<MikeSmith>
e.g., srcset="image1.png 1X, image1.png 2X"
10:14
<MikeSmith>
(though probably not that example)
10:15
<zcorpan>
MikeSmith: can you file a bug?
10:15
<MikeSmith>
yeah
10:15
<zcorpan>
thx
10:17
<MikeSmith>
zcorpan: also I told you before I wasn't storing the URLs anyway. But I changed that and I am now, because I use them for reporting purposes in the error messages
10:17
<MikeSmith>
and I do that because I can't report the exactly column number of where the error is
10:18
<MikeSmith>
which is a general deficiency in the way the validator datatype/microsyntax error-reporting is designed
10:19
<MikeSmith>
most of the time it's not a problem to no report the exacty column number because the attribute values are generally short and not complex
10:19
<MikeSmith>
but the case for srcset is a bit different
10:21
<zcorpan>
MikeSmith: yeah i guess you'll highlight the entire value right?
10:21
<zcorpan>
or the whole tag even
10:22
<MikeSmith>
right
10:23
<MikeSmith>
zcorpan: it's less than ideal but it'd be a lot of work to change given the current design of the datatype-checking code
10:23
<zcorpan>
yeah
10:23
<zcorpan>
i guess people can figure out what the error is anyway
10:26
<zcorpan>
maybe you can do a hack and give a relevant extract in the message?
10:27
<annevk>
Lovely, mask:url(#something) references an element in the document the stylesheet is associated with
10:28
<annevk>
However, they also allow mask:url(/elsewhere#something) which would parse relative to the stylesheet's URL, then be fetched, ...
10:28
<smaug____>
what is an opposite of composed? incomposed seems to have also some other meanings.
10:28
<zcorpan>
MikeSmith: something like srcset="lala 1.5x, foo,, bar 2x" -> Bad value for attribute srcset on element img: Found empty image candidate in `foo,,`
10:29
<annevk>
I guess you could parse against the stylesheet's URL and then treat it as a document-reference if it's the same as the stylesheet's URL minus the fragment thing... But oh boy
10:29
<annevk>
smaug____: excited per Google
10:30
<annevk>
smaug____: also agitated, boisterous, disturbed, excited, fierce, frantic, frenzied, furious, heated, passionate, raging, roused, ruffled, stormy, turbulent, violent, wild, wrathful
10:30
<smaug____>
boisterousDocument
10:30
<smaug____>
sounds perfect
10:30
<annevk>
:-)
10:30
<MikeSmith>
flustered
10:31
<annevk>
ruffled as technical term would also be fun
10:31
<MikeSmith>
unsettled
10:32
<annevk>
I should probably post about IDNA being updated at some point
10:32
<annevk>
Now that http://www.unicode.org/reports/tr46/ is published with the revised algorithms
10:32
smaug____
is trying to figure out names for the trees where document is the root node, and doesn't contain shadow dom, and the tree which does contain shadow dom. ComposedDoc for the latter
10:33
<MikeSmith>
raw?
10:33
<annevk>
smaug____: why does the former need to be different from Document
10:33
<MikeSmith>
untainted?
10:34
<annevk>
smaug____: document and composedDocument seems fine
10:34
<smaug____>
annevk: this is mainly for Gecko internal stuff
10:34
<smaug____>
in order to force the API user to think whether shadow dom should be handled
10:34
<MikeSmith>
smaug____: PreComposed?
10:35
<smaug____>
well, it is not pre
10:35
<smaug____>
it is Document
10:35
<MikeSmith>
then Un-
10:35
<MikeSmith>
UnComposed
10:35
<smaug____>
perhaps that
10:37
<MikeSmith>
anyway it seems analogous the character composition, so maybe there are already some names that could be repurposed from the code around that
10:38
smaug____
uses un-
10:42
<MikeSmith>
I naively set out on srcet checking hoping I'd be able to get by without needing to create strings and store strings
10:42
<MikeSmith>
so much for that plan
10:46
<MikeSmith>
zcorpan: ah I remember now that's why I was asking you about the URLs. So far the spec has no requirement that would make me need to compare the URLs to one another, so so far I don't need to create strings for them or store them as strings
10:52
<zcorpan>
you need to store and compare the parsed descriptors though
10:53
<darobin>
MikeSmith: I had a case not long ago in which better datatype error reporting would have been really nice, it was for a pretty long data: URL
10:53
<zcorpan>
(but they're not strings, so yeah)
10:53
<darobin>
and all it told me was that there was a character in there it didn't like (and that I should be careful with quotes and spaces, which wasn't the problem)
10:55
<annevk>
MikeSmith: still no schemaless validator?
10:56
<zcorpan>
annevk: that'd be a pretty big leap
10:57
<annevk>
one small step for the validator, ...
11:42
<MikeSmith>
darobin: yeah I noticed that same data: URL error-reporting problem
11:44
<darobin>
MikeSmith: I understand that giving the column number is nigh impossible (and I can imagine why), but if you could somehow surface the faulty character it would help lots
11:46
<MikeSmith>
darobin: so that's possible at least. I mean obviously we have the problem character available at that point. we just need to add it to the error message
11:46
<MikeSmith>
the data: URL reporting code is some relatively old part
11:47
<MikeSmith>
it probably needs some attention in other parts too
11:49
<MikeSmith>
annevk: I've been moving away from the schema more and more when possible at least. e.g., for srscet the basic srcset-allowed-for-source-and-img constraint is all that the schema specifies, and the schema says the datatype for srcset is just "string". I put the actual real datatype association-and-checking for it into other code
11:50
<MikeSmith>
in part because srcset is unique in that it basically has two different dataypes depending on whether the "sizes" attribute is also specified
11:51
<MikeSmith>
and also because I tried it first in the schema and it exposed a bug in jing schema-checking code that I couldn't be assed to try to troubleshoot and fix
11:52
<MikeSmith>
it actually exposed a case in jing where the behavior was different depending on the *order* of the attributes -- different if sizes was specified before srcset, or after
11:52
<MikeSmith>
which, I didn't even think that would be possible given the way the parser and jing work
11:53
<MikeSmith>
but I guess it actually use
11:53
<MikeSmith>
*is
11:54
<MikeSmith>
darobin: if you have time to file a bug about the data URL thing, please do. It's an easy fix and I could do it this weekened
11:56
<darobin>
MikeSmith: I was already on the bugzilla page :)
11:57
MikeSmith
and darobin putting the MPAA dollars to good use
11:58
<darobin>
:)
12:15
<zcorpan>
MikeSmith: note that srcset with w is allowed without sizes
13:01
<MikeSmith>
zcorpan: yeah I know that :) but the thing is internally in the vnu design I essentially need to handle it as a separate datatype when sizes is there too
13:01
<zcorpan>
MikeSmith: k :-)
13:03
<annevk>
zcorpan: http://dev.w3.org/csswg/cssom/#css-style-sheets is somewhat wrong
13:03
<annevk>
zcorpan: location should be named href and you should have a field url or some such that's the actual location against which urls are resolved and such
13:03
<annevk>
zcorpan: is the plan that everything else in CSS references this?
13:03
<zcorpan>
annevk: please file a bug
13:04
<zcorpan>
annevk: not sure, but it needs to be well-defined at some point
13:05
<annevk>
https://www.w3.org/Bugs/Public/show_bug.cgi?id=26116
13:05
<annevk>
I'm trying to figure out fetching for SVG
13:05
<annevk>
Seems CSS could use some cleanup there too
13:06
<annevk>
And SVG requires this thing where you load an external SVG document, but don't actually fetch anything from it that isn't a blob or data or other local URL
13:06
<annevk>
Styles need to be resolved however so that property needs to be supported by CSS too
13:07
<annevk>
However, it's unclear where in CSS such things would be defined
13:07
<annevk>
TabAtkins: ^^
13:08
<zcorpan>
earlier today i was pondering how <img src=svg> works when it has a late well-formedness error
13:09
<annevk>
I guess that's not too different from a GIF or some such that has an error late?
13:09
<annevk>
Ill-defined :/
14:02
<zcorpan>
annevk: GIF either goes from unavailable -> error or unavailable -> partially available -> completely available (even if it has late errors, it doesn't make the whole thing corrupt). i think
14:07
<Domenic>
MikeSmith: zcorpan: maybe <img srcset="pixelized.png 1x, vectorized.svg 1.5x, pixelized-2 2x, vectorized.svg 2.5x"> would be a use case for the same URL multiple times?
14:08
<zcorpan>
Domenic: maybe, but would you actually do that? instead of <img src="pixelated.png" srcset="vectorized.svg"> ?
14:10
<Domenic>
i kind of envisioned people creating specific breakpoint images for certain resolutions/densities, and then saying that they'd rather use a vectorized version in between for perfect scaling, instead of letting the browser do it fuzzily.
14:10
darobin
throws some lowsrc at zcorpan, just for funs
14:12
<zcorpan>
darobin: https://www.w3.org/Bugs/Public/show_bug.cgi?id=26072
14:13
<darobin>
zcorpan: oh man, I was actually joking — but you mean business
14:13
<Ms2ger>
Good old lowsrc
14:14
darobin
recalls the jokes, "Hey, unfortunately my ISP is France Telecom, could you add lowlowsrc to the spec please?"
14:15
<odinho>
:P
14:19
<yoav>
zcorpan: FWIW, I think that any data URI/cached image can be used as a replacement image until the real resource is downloaded
14:19
<yoav>
See https://code.google.com/p/chromium/issues/detail?id=382591
14:19
<yoav>
zcorpan: And AFAICT, this is inside the "UA resource selection optimization" domain
14:20
<yoav>
So no need to spec it (even though we can add a note or something)
14:22
<zcorpan>
yoav: intredasting
14:25
<zcorpan>
ok wontfixed the lowsrc bug
14:28
<zcorpan>
annevk: thx for the bug
14:55
<Domenic>
http://www.openwebfoundation.org/legal/the-owf-1-0-agreements/owfa-1-0 redirects to bing??
14:56
<Ms2ger>
https://sites.google.com/site/openwebfoundation/
14:56
<Ms2ger>
Wat
14:59
annevk
*blinks*
14:59
<annevk>
mathiasbynens: https://code.google.com/p/v8/source/detail?r=18683 does not seem good, writing invalid utf-8 should not be an option...
15:00
<jgraham>
annevk: Traitor
15:00
<annevk>
jgraham: ?
15:00
<jgraham>
annevk: Don't worry
15:00
<annevk>
k
15:00
<jgraham>
It wasn't funny anyway :)
15:10
<mathiasbynens>
annevk: felixge said it’s because the V8 team wants to keep the API stable :/
15:11
<mathiasbynens>
(note that it’s not just an option – it’s the default)
15:13
<annevk>
they are crazy
15:21
<darobin>
V8? Crazy? Noooo
15:25
<annevk>
heard it here first
15:31
<zcorpan>
Hixie: i've added picture, i tried running anolis locally and i think the xrefs are OK now. let me know if something's still broken.
15:37
<TabAtkins>
annevk: What is "this thing where you load an external SVG document"?
15:37
<annevk>
TabAtkins: I'll need some more context :-)
15:38
<TabAtkins>
The thing you mentioned me about 2 hours ago.
15:38
<annevk>
TabAtkins: oh, SVG as image
15:39
<TabAtkins>
Oh, now I understand your sentence.
15:40
<TabAtkins>
Referring to a "property" confused me, as it looked like you meant a CSS property, not an abstract property.
15:55
<Domenic>
Hixie: I'm happy to buy a SSL cert for resources.whatwg.org
15:58
<jgraham>
https://www.globalsign.com/ssl/ssl-open-source/ might apply
16:03
<Domenic>
The main use case is avoiding mixed-content warnings on HTTPS sites, like github.io or github.com
16:44
<mathiasbynens>
Domenic: might as well go for a *.whatwg.org cert then
16:45
<Domenic>
mathiasbynens: if we can get a free one, sure, but we'd need a $300/year one to cover all the subdomains and sub-subdomains
16:45
<Domenic>
mathiasbynens: turns out wildcard certs ($100/year) only cover a single level
16:45
<mathiasbynens>
Domenic: TIL
16:53
<annevk>
TabAtkins: I've been working with krit on figuring out how we should define those features that can reference both an SVG element and an image and it seems we're closing in on a solution
16:53
<annevk>
TabAtkins: I guess at some point we should discuss what it'll take to define fetching for CSS properties in general
16:54
<annevk>
TabAtkins: it seems part of the infrastructure for that is defined in CSSOM today (the concept of a CSS style sheet and its various subconcepts)
17:15
<TabAtkins>
Yeah, but possibly not enough. More than happy to help figure this out whenever you have time, or to apply whatever you and krit bang out.
17:16
<TabAtkins>
Dunno if you heard, but we figured out how we want to allow passing flags and whatnot with urls.
17:17
<TabAtkins>
Within our existing budget of lookahead, I can change parsing to make *quoted* url() functions actually a FUNCTION token, so they can take additional values after the url string. (Unquoted urls would still be the magical weird URL token.)
17:17
<TabAtkins>
Passed dbaron's initial sanity check, so I'm happy about it.
17:18
<TabAtkins>
annevk: Also, would appreciate your feedback on the WHATWG bikeshed header, over in my github.
17:18
<TabAtkins>
Domenic has a couple of questions about variations between your headers and his.
17:20
<annevk>
TabAtkins: looking now
17:20
<krit>
TabAtkins: if you mean CORS arguments as part of URL, we didn’t touch it… that would still be up to you and annevk
17:20
<TabAtkins>
krit: Yeah, that's what I'm talking about.
17:20
<TabAtkins>
Ok.
17:20
<annevk>
As far as I can tell we need CORS mainly for @import
17:21
<TabAtkins>
Yeah, so we can expose the stylesheets.
17:21
<annevk>
But it might be good to make url() in general forward compatible with it at least
17:21
<TabAtkins>
But also we'll need them if we ever expose more resources that are cors-restricted by default.
17:22
<TabAtkins>
annevk: URL good enough for me to refer to it from V&U to define url parsing now?
17:22
<annevk>
TabAtkins: yeah, think so
17:22
<annevk>
TabAtkins: but note that you need access to a base URL
17:22
<annevk>
TabAtkins: you need some kind of environment setup too
17:22
<TabAtkins>
Right, which the stylesheet should provide.
17:22
<TabAtkins>
Interesting.
17:23
<TabAtkins>
I'll read more into it.
17:23
<annevk>
TabAtkins: well for now it's mostly base URL; but if we expose these properties in script somehow they might want to make blob URLs work somehow
17:24
<Hixie>
zcorpan: regenning...
17:24
<Hixie>
Domenic: cool
17:24
<Hixie>
i wonder how you can get it to me safely
17:24
<Hixie>
let's see
17:27
<Ms2ger>
Hixie, pgp, clearly
17:27
<Hixie>
Domenic: you around?
17:27
<Domenic>
Hixie: ya just got back
17:41
<Hixie>
zcorpan: Possible xref problems:
17:41
<Hixie>
+ density</var> be the URL and pixel density that results from <span>selecting an image source</span>, respectively.</li>
17:48
<zcorpan>
Hixie: fixed
17:49
<Hixie>
zcorpan: cool, will integrate that too
17:49
<Hixie>
zcorpan: sorry i haven't been regenning much recently, i'm updating my pipeline
17:49
<Hixie>
zcorpan: hopefully gonna make it sooo much quicker
17:49
<Hixie>
i've got the parsing of the spec down to four seconds
17:49
<zcorpan>
Hixie: it's ok
17:49
<zcorpan>
nice
17:50
<Hixie>
takes 90MB to hold the parse tree, but...
17:51
<SamB>
Hixie: does that involve a lot of redundancy?
17:51
<zcorpan>
might be less memory to use the v.nu parser in streaming mode
17:51
<Hixie>
memory is cheap
17:51
<SamB>
does that support Object Pascal?
17:51
<Hixie>
it's time that i'm worried about
17:51
<zcorpan>
SamB: i think only java and c++
17:52
<zcorpan>
actually dunno if the c++ version has the streaming mode
17:52
<Hixie>
SamB: most of the cost is my DOM, i think. The source is only 6MB, so even if i have two or three copies of it, I wouldn't get to 90MB just from that.
17:52
<SamB>
zcorpan: that's one more language than I expected you to mention
17:52
<Hixie>
if i used validator.nu's parser i'd do it in Java or C++
17:52
<SamB>
Hixie: I didn't expect the source to even be included in that figure
17:53
<zcorpan>
SamB: firefox uses the c++ parser
17:53
<Hixie>
i dunno how fast validator.nu is though
17:53
<Hixie>
SamB: 90MB is the memory usage of the process after it's parsed the spec
17:53
<Hixie>
so it includes everything
17:53
<Hixie>
though i recently changed how i read the file to just be an mmap, dunno how that affects it
17:54
<Hixie>
medium-term i plan to move to a world where i don't copy the strings at all, since that's about 10% of the total time the process takes right now
17:54
<Hixie>
maybe even 20%
17:54
<SamB>
Hixie: what proxy for "memory usage" are you using?
17:54
<Hixie>
i forget the most recent numbers
17:54
<zcorpan>
https://hsivonen.fi/cost-of-html/ - dunno if it tells you anything you can compare it to
17:54
<Hixie>
SamB: resident size
17:54
<zcorpan>
might also be out of date
17:54
<Hixie>
hm, interesting
17:54
Hixie
looks
17:55
<SamB>
and, er, which OS is this?
17:55
<zcorpan>
Hixie: http://www.whatwg.org/specs/web-apps/current-work/multipage/edits.html#dependencies-0 should be moved. should i file bugs or is irc ok?
17:56
<Hixie>
SamB: debian
17:56
<SamB>
ah, my favorite :-)
17:57
<Hixie>
zcorpan: file a bug, i'll do it when it's more stable
17:57
<Hixie>
zcorpan: doesn't hurt to have it there for now
17:57
<zcorpan>
Hixie: k
17:57
<Hixie>
zcorpan: i'll review the <picture> section in detail at some point
17:57
<zcorpan>
Hixie: great, thanks
17:58
<Hixie>
no, thank _you_! i'm so glad i don't have to do this
17:59
<SamB>
Hixie: anyway, I think I might look at the private size rather than the resident size
17:59
<Hixie>
SamB: if i really cared about memory usage, i'd probably just instrument the internals
17:59
<zcorpan>
Hixie: my pleasure
18:00
<Hixie>
SamB: my main concern is speed
18:00
<SamB>
yeah
18:00
<Hixie>
and four seconds is about 3.99 seconds too slow
18:00
<Hixie>
:-)
18:00
zcorpan
gotta go
18:00
<SamB>
Hixie: maybe you're thrashing the cache
18:00
<annevk>
TabAtkins: so btw, I'm happy to help out with URL / Fetch integration / CSS model setup
18:01
<annevk>
TabAtkins: however, I have a pretty serious timezone problem
18:01
<TabAtkins>
Yeah. ^_^
18:01
<Hixie>
SamB: yeah, need to run it under valgrind-cachegrind
18:01
<Hixie>
SamB: right now i still have a low-hanging fruit in the form of the string copying to deal with
18:02
<annevk>
TabAtkins: I'll see if I can type out some advice soonish and we can do it the very slow way on www-style
18:02
<Hixie>
SamB: not gonna try to figure out cachegrind until i've got everything i can out of callgrind :-)
18:04
<SamB>
Hixie: I'm almost surprised they're distinct frontends, honestly
18:04
<SamB>
they have a hell of a lot of overlap
18:04
<Hixie>
yeah?
18:04
<Hixie>
i've never used cachegrind
18:04
<Hixie>
then again i'd never used callgrind until a few days ago
18:05
<SamB>
not saying that they have enough that you can just use cachegrind for everything
18:05
<SamB>
but they've got more in common than any two other valgrind tools I can think of
18:06
<TabAtkins>
annevk: The very slow way is better for me, because it means I don't have to hold things in my head until I'm ready to do them. ^_^
18:07
<annevk>
TabAtkins: sounds good
18:07
<Hixie>
SamB: it did seem that way from the docs, yeah
18:10
<annevk>
<picture> changes not being announced on @WHATWG seems bad
18:10
<Hixie>
yeah, that's an artefact of my currnet pipeline and how they're being merged in
18:10
<Hixie>
i'll add it to the list of things to fix
18:10
<Hixie>
(or, file a bug to remind me)
18:10
<Hixie>
(put "tools" in the whiteboard)
18:11
<annevk>
https://www.w3.org/Bugs/Public/show_bug.cgi?id=26121
18:13
<Hixie>
thanks
21:04
<aleray_>
hi, I would like to markup seconds (or milliseconds if needed) in rdfa, like <span property="aa:begin" content="5.2">00:00:05,2</span>. how can I specify the content is a float/a second value?
21:05
<Ms2ger>
I doubt you'll find a lot of rdfa enthusiasts here
21:05
<scor>
aleray_: use the datatype attribute, for example datatype="xsd:float"
21:06
<jgraham>
Wow RDFa and XSD? Good thing I left my sense of logic at the door
21:06
<aleray_>
scor, thanks. probably a stupid question, but Is there anything specific datattype for time?
21:07
<SamB>
Ms2ger: because people around here don't like RDF, or because they dislike RDFa in particular?
21:07
<jgraham>
I think yes to both
21:08
<SamB>
aren't the datatypes the only thing worth using from XML Schema?
21:08
<scor>
aleray_: is it a duration, or a time that you want to represent?
21:09
<aleray_>
scor, a timecode: begin and end times
21:09
<scor>
aleray_: either way, #swig is a probably a better channel to ask your questions
21:10
<aleray_>
scor, oups, that the chan I inteded to join
21:10
<aleray_>
thanks