| 00:15 | <Dashiva> | hober: Sensible stuff |
| 00:31 | <mgrdcm> | hober: nice. (hi) |
| 02:07 | <ezyang> | Just making sure I'm not crazy: If I have <math><p>, the <p> node gets placed *outside* of <math>, correct? |
| 02:27 | <ezyang> | Should active formatting elements list every contain a single element which is null? |
| 02:27 | <ezyang> | *ever |
| 02:30 | <ezyang> | Oh, that's their internal representation of Marker |
| 02:48 | <ezyang> | Where in the spec does it specify that, after I encounter the </foreignObject> tag, I'm supposed to somehow switch back into IN_FOREIGN_CONTENT? |
| 02:50 | <ezyang> | It looks like the spec makes a half-hearted attempt to keep the mode in "in foreign content", but as soon as we have some sort of complicated code inside that, we move to another mode and consequently lose the ability to "go back" |
| 02:58 | <ezyang> | Also, assuming that this is a spec error (I'd put it at an 80% chance it's not, but I'm hoping it is), PHP html5lib now has 100% tests passing |
| 02:59 | <ezyang> | erm, s/tests/specced tests/ |
| 02:59 | <ezyang> | And has, maybe, three nasty workarounds (one for empty doctypes, and two for namespaced attributes) |
| 07:29 | <hsivonen> | ezyang: The table taint thing is the only case where the tests deviated from current spec intentionally because of me |
| 07:30 | <zcorpan_> | hsivonen: you can flip a setting in ie8 to disable the xss filter |
| 07:30 | <hsivonen> | zcorpan_: where? |
| 07:33 | <zcorpan_> | hsivonen: internet settings, security, custom level, scripting, enable xss filter |
| 07:34 | <MikeSmith> | zcorpan_: you still around? |
| 07:34 | <zcorpan_> | MikeSmith: yep |
| 07:35 | <MikeSmith> | about "If the area element has no href attribute, then the area represented by the element cannot be selected, and the alt attribute must be omitted." |
| 07:35 | <MikeSmith> | does v.nu not already handle that correctly? |
| 07:35 | zcorpan_ | wonders why the labels are "(o) Disable-ON ( ) Enable-OFF" |
| 07:36 | <zcorpan_> | MikeSmith: dunno, that was re the spec bug you filed |
| 07:36 | <MikeSmith> | oh |
| 07:36 | <MikeSmith> | remind me which bug that was? |
| 07:37 | <zcorpan_> | http://www.w3.org/Bugs/Public/show_bug.cgi?id=6976 |
| 07:38 | <MikeSmith> | ah |
| 07:38 | <MikeSmith> | I didn't file that one, it just got auto-assigned to me |
| 07:38 | <zcorpan_> | oh |
| 07:38 | <MikeSmith> | because dude filed it against the h:tml draft |
| 07:39 | <MikeSmith> | anyway, I'll reply to it and close it out |
| 07:39 | <zcorpan_> | i didn't look carefully enough at the labels |
| 07:40 | <MikeSmith> | np |
| 07:40 | <MikeSmith> | it did make me think some about how v.nu error-reporting could be improved for this case |
| 07:41 | <MikeSmith> | because right now, v.nu just says, "Error: Required attributes missing on element area." for cases like this |
| 07:41 | <zcorpan_> | MikeSmith: hack jing to emit more useful messages? :) |
| 07:42 | <MikeSmith> | well, I think hsivonen's already done some improvements that way |
| 07:42 | <MikeSmith> | I was thinking about possibility adding it in the separate-pass assertions-checking |
| 07:42 | <MikeSmith> | the assertions.sch Schematron file and the Assertions.java code |
| 07:43 | <MikeSmith> | could have that specifically report, "An area element without an href attribute must not have an alt attribute." |
| 07:43 | <MikeSmith> | or whatever |
| 07:43 | <zcorpan_> | as another message? |
| 07:44 | <MikeSmith> | yeah, it would need to be another message |
| 07:44 | <MikeSmith> | if we did it that way |
| 07:44 | <MikeSmith> | (as opposed to hacking jing) |
| 07:45 | <zcorpan_> | i think it would be a better long-term solution to hack jing |
| 07:45 | <hsivonen> | zcorpan_: thanks. not the most discoverable thing |
| 07:45 | <zcorpan_> | but then again i'm not the one who's hacking |
| 07:46 | <zcorpan_> | hsivonen: indeed, ie's preferences are even worse than opera's :) |
| 07:46 | <zcorpan_> | at least opera has opera:config |
| 07:46 | <hsivonen> | maybe I should put software.hixie.ch into my "intranet" and set the pref only for that |
| 07:47 | <hsivonen> | hmm. but then the compat mode thing could do intranet weirdness and confuse my live DOM testing |
| 07:48 | <zcorpan_> | hsivonen: you could make the compat mode not depend on intranet, no? |
| 07:48 | <hsivonen> | yes |
| 07:49 | <zcorpan_> | i just disabled the xss filter since all i use ie8 for is for viewing the live dom viewer :) |
| 07:50 | <hsivonen> | I don't use it for much else, either, but sometimes I browse to a test case somewhere else or seach the Web in IE8 if I'm already in the IE8 window |
| 07:50 | zcorpan_ | noticed that opera's new mac native skin looks better than the previous skin |
| 07:50 | <hsivonen> | is there a new public build of Opera? |
| 07:51 | <zcorpan_> | my opera just auto-updated |
| 07:51 | <hsivonen> | does Opera have different auto-update channels? |
| 07:51 | <zcorpan_> | though i have flipped the "download all snapshots" pref |
| 07:52 | <hsivonen> | ah. nice |
| 07:52 | hsivonen | didn't know about that pref, either |
| 07:53 | <hsivonen> | maybe I should put Chrome on dev channel, too |
| 07:53 | hsivonen | wonders if WebKit nightlies on Windows have autoupdate these days |
| 07:53 | <zcorpan_> | in chrome it's even harder to discover |
| 07:53 | <zcorpan_> | although i guess it depends on how you go about to find out how to do it |
| 07:54 | <zcorpan_> | if you google for it it might be easier in chrome |
| 07:58 | <hsivonen> | hmm. has Chrome 2 even not been pushed to stable channel yet? |
| 07:59 | <hsivonen> | or is my instance of chrome just slow to update because I don't leave the Windows VM idle for long enough times with Chrome open? |
| 08:00 | <hsivonen> | hmm. looks like Chrome defaults to SSL cert CRL off |
| 08:00 | <hsivonen> | cert security FTW |
| 08:09 | <othermaciej> | I don't believe Chrome 2 is in the stable channel |
| 08:09 | <othermaciej> | I think it's "beta" |
| 08:09 | <othermaciej> | stable is 1.0 |
| 08:09 | <othermaciej> | dev is 3.0.x |
| 08:09 | <hsivonen> | ok. |
| 08:09 | <othermaciej> | yay for burning through version numbers |
| 08:10 | <abarth> | othermaciej: 2 is in stable |
| 08:10 | <hsivonen> | it seems to me that Chrome 3 is slow to start playing Ogg files |
| 08:10 | <abarth> | othermaciej: 1.0 is dead |
| 08:10 | <othermaciej> | abarth: I guess I was out of date |
| 08:10 | <hsivonen> | I wonder if they wait for the whole file in order to show the time track |
| 08:10 | <abarth> | othermaciej: i'm not sure whether it's been pushed out for auto update or just for folks who ask for it |
| 08:15 | <othermaciej> | "plugin prison" seems to be an effective buzzword |
| 08:15 | <othermaciej> | it's much more hostile than I would go for in explaining the benefits of <video> and such... |
| 08:32 | <ojan> | abarth: 2.0 has not been pushed on autoupdate yet. it's just there for new installs. |
| 08:33 | <abarth> | ojan: thanks |
| 08:39 | <Mrmil> | I have a <h1> element with <img ... alt="Some Company"> logo but this outliner (http://gsnedders.html5.org/outliner/) doesn't show the alt text in the outline. Do you think it would be possible to make it reads the alt text and show it if no other text is present? |
| 08:44 | <gsnedders> | Mrmil: zcorpan_ was trying to get me to implement that :P |
| 08:45 | <zcorpan_> | i think Hixie is wasting people's time with making </head>, </body> and </html> do something other than being ignored |
| 08:46 | <Mrmil> | gsnedders: Hehehe, did he succeed? :) |
| 08:46 | <gsnedders> | Mrmil: I guess I probably will |
| 08:46 | <gsnedders> | (get around to doing it) |
| 08:47 | <Mrmil> | gsnedders: Ok. *grins* |
| 08:48 | <annevk42> | zcorpan_, he's right about whitespace roundtripping though |
| 08:48 | <annevk42> | zcorpan_, but supposedly editors have to deal differently with whitespace anyway |
| 08:49 | <annevk42> | zcorpan_, though that would indicate the current parsing algorithm is not suitable for editors |
| 08:49 | <zcorpan_> | i guess editors would want to preserve whitespace between attributes too |
| 08:49 | <annevk42> | yeah |
| 08:49 | <zcorpan_> | which means that the dom is not a suitable model for editors |
| 08:50 | <gsnedders> | Hmm, Sims 3's first release date is tomorrow |
| 08:50 | <hsivonen> | I think Dreamweaver preserves all sorts of crazy parse errors |
| 08:50 | <hsivonen> | the suitability of the parsing algorithm depends on the kind of editor |
| 08:50 | <hsivonen> | if you want to make a GUI-only editor, it's fine |
| 08:51 | <zcorpan_> | yeah i guess you'd want to preserve weird doctypes and bogus comments etc too |
| 08:51 | <hsivonen> | if you want to make a dual GUI/text editor like Dreamweaver, you are in the world of hurt and complexity |
| 08:55 | <Hixie> | man, jd really has his job cut out for him when his customers go to lengths like these: http://www.reddit.com/r/programming/comments/8opdq/html_5_and_web_video_freeing_rich_media_from/c09yhr2 |
| 09:02 | hsivonen | wonders when accessibility enters the Flash v. HTML5 comparisons with more force |
| 09:02 | <othermaciej> | Hixie: his new story appears to be that HTML is just the next Silverlight |
| 09:02 | <hsivonen> | How's Flash accessibility doing on Mac and Linux these days? |
| 09:02 | <Hixie> | othermaciej: wow, that's quite the compliment he's giving us |
| 09:03 | <othermaciej> | hsivonen: that comparison is why I try to discourage people from citing Bespin as an HTML5 poster child... |
| 09:03 | <Hixie> | bespin is a poster child for us providing <textarea> syntax highlighting :-) |
| 09:03 | <Hixie> | maybe in the next version... |
| 09:03 | <Hixie> | (i mean seriously, they must REALLY want that feature to do what they did in bespin) |
| 09:04 | <Hixie> | (it's a technological tour de force) |
| 09:09 | <annevk42> | Hixie, I've the feeling that's not enough |
| 09:10 | <annevk42> | Hixie, how would you e.g. implement code-auto-completion? |
| 09:10 | <Hixie> | we'd probably need that too, yes |
| 09:11 | <Hixie> | maybe textarea needs to be defined as an xbl2 binding to contenteditable content |
| 09:13 | <hsivonen> | there's also the issue that the DOM is rather heavy-weight as the model of a text editor view |
| 09:13 | <othermaciej> | so far I've had people tell me I should add more disclaimers to the design principles, I should either make them stated absolutely without disclaimers or delete them, or that I should state them without disclaimers but with an unstated expectation that they are not actually absolute |
| 09:13 | <othermaciej> | rarely have I seen a more clear example of the You Can't Please Everybody principle |
| 09:14 | <hsivonen> | maybe we should document that We Can't Please Everybody |
| 09:14 | <annevk42> | maybe we should have a principle: "Design Principles are no Magic Wand" |
| 09:14 | <othermaciej> | hsivonen: but what if there is a case where you can please everybody? |
| 09:15 | <hsivonen> | othermaciej: these absolutes are hard |
| 09:16 | <othermaciej> | I also can't believe that "Tables" was cited as an example of a good design principle |
| 09:17 | <MikeSmith> | I think the principle should be stated, For Any Given Proposal|Decision, We Will Have at Least One Jackass Formally Object to It, Verbosely |
| 09:18 | annevk42 | wonders how bad the fallout will be from Hixie's latest email |
| 09:19 | <othermaciej> | MikeSmith: I think we may have actually had some decisions that occurred without a Formal Objection |
| 09:19 | <othermaciej> | can't remember though |
| 09:20 | <MikeSmith> | I think if we had a proposed decision to publish the statement, "It's very likely that the sun will rise tomorrow morning.", we'd have objections . |
| 09:20 | <othermaciej> | IE is down to 65.59% |
| 09:21 | <othermaciej> | that puts it at down 2.56% for the year to date |
| 09:21 | <othermaciej> | at this rate it will still take years to get under 50% |
| 09:22 | <jgraham> | othree: In any case "You can't please everybody" would be rejected on the grounds that it was just a thinly veiled excuse not to care about accessibility |
| 09:23 | <jgraham> | Or rather people would complain about it citing that as a reason |
| 09:23 | <Hixie> | i care about accessibility |
| 09:23 | <Hixie> | it's self-styled accessibility experts i don't care about :-) |
| 09:23 | <jgraham> | Hixie: I care about accessibility too |
| 09:24 | <Hixie> | of course many of them don't seem to be able to distinguish the two |
| 09:25 | <othermaciej> | I'm not generally in favor of self-styled experts of any stripe |
| 09:25 | <jgraham> | The point was that given a self-evident principle like "you can't please everybody", I predict that we would get complaints that the principle could be used as an excuse to preferentially exclude the disabled |
| 09:25 | <othermaciej> | I prefer the kind that can do without styling themselves |
| 09:25 | <jgraham> | Even though that would, of course, not be the case |
| 09:25 | <Hixie> | people already claim that |
| 09:26 | <Hixie> | so i don't think that would be a new problem |
| 09:26 | <othermaciej> | nowadays people seem more concerned with us excluding critical non-browser HTML implementations |
| 09:26 | <othermaciej> | such as libwww-perl and refrigerators |
| 09:26 | <gavin_> | I don't think jgraham was tying to claim that it would be anew problem :) |
| 09:27 | <jgraham> | Hixie: Sure. I was just pointing out how even the most obvious statement of compromise would be bent by those determined to see HTML5/HATWG/hatever in a bad light |
| 09:27 | <Hixie> | well sure |
| 09:27 | <Hixie> | but as you say |
| 09:27 | <Hixie> | even the most obvious statement would be taken that way |
| 09:27 | <Hixie> | so i wouldn't concern myself with that |
| 09:28 | <jgraham> | wwwww <- please insert any extra w's from this stash |
| 09:29 | <othermaciej> | jgraham: did you break your w? |
| 09:31 | <abarth> | i'm trying to think of some way to direct this discussion so that I can be quoted by Mr. Last Week |
| 09:32 | <Hixie> | you have such low ambitions :-P |
| 09:32 | <Hixie> | btw, thanks to plh's clarifications, i'm glad to say that html5 is now a standard (in development), rather than just a recommendation |
| 09:32 | <abarth> | Hixie: easy for you to say. you get quoted by him all the time :) |
| 09:32 | <Hixie> | at least on the whatwg site |
| 09:32 | <othermaciej> | abarth: your line above might be enough! |
| 09:33 | <othermaciej> | mpilgrim got cited so you can score without your name being on his enemies list apparently |
| 09:33 | <Hixie> | abarth: you just have to compliment easily offended people using long words that he won't understand |
| 09:33 | <abarth> | haha |
| 09:34 | <Hixie> | abarth: or, say things that are insults but are stated in ways that make the insult sound like they are the most obvious thing in the world to you |
| 09:34 | <annevk42> | othermaciej, I think mpilgrim made it to his list |
| 09:34 | <Hixie> | abarth: (e.g. implying that mr last week doesn't understand long words) |
| 09:34 | <annevk42> | considering what he said about him anyway |
| 09:34 | <othermaciej> | annevk42: he has now I guess |
| 09:35 | <othermaciej> | but he wasn't on the canonical list of Enemies of the People |
| 09:35 | <Hixie> | there's an enemies list? |
| 09:35 | <abarth> | I think Mr. Last Week is secretly a fan of Hixie and is trying to discredit the opposition by being rediculous. |
| 09:35 | <annevk42> | not an actual published list |
| 09:36 | <othermaciej> | r*i*diculous |
| 09:36 | <gsnedders> | abarth: I think we need to undermine the IETF HTTPbis WG by saying you shouldn't concat. headers with ", " :) |
| 09:36 | <othermaciej> | (spelling pet peeve) |
| 09:36 | <othermaciej> | annevk42: remember his TPAC checklist? |
| 09:36 | <othermaciej> | abarth: it's hard to tell! |
| 09:36 | gsnedders | notes annevk42 signed that and othermaciej didn't :P |
| 09:36 | <othermaciej> | abarth: his "satirical" posts sometimes make the opposition sounds dumb |
| 09:36 | <annevk42> | othermaciej wasn't there... |
| 09:36 | <gsnedders> | annevk42: I know. |
| 09:37 | <gsnedders> | http://www.flickr.com/photos/gsnedders/2979057018/in/set-72157608267037421/ |
| 09:37 | <othermaciej> | abarth: but the outright profane ones, or postings of the most unflattering photos of Hixie he could find on the internet, those seem sincerely mean-spirited |
| 09:37 | <abarth> | gsnedders: I don't really understand the HTTPbis WG. It's unclear to me how to help out there. |
| 09:37 | <othermaciej> | gsnedders: I'll get you at this year's TPAC if it ends up happening and people go |
| 09:37 | <othermaciej> | (I gather the fee is controversial...) |
| 09:37 | <Hixie> | in all fairness, i'm proud of that vampire photo |
| 09:38 | <Hixie> | it was taken at a party at hyatt's when i was an intern at netscape |
| 09:38 | <gsnedders> | othermaciej: I'm _probably_ not going. |
| 09:38 | gsnedders | wonders |
| 09:38 | <gsnedders> | Oh, that vampire photo |
| 09:38 | <othermaciej> | plenty o' cred, then |
| 09:38 | <gsnedders> | abarth: Nor do I understand them. |
| 09:39 | <othermaciej> | abarth: they seem more interested in spitting the RFC into a dozen pieces than in actually fixing anything |
| 09:40 | <abarth> | Hixie: how far back is the vampire picture? i've been scrolling through but don't see it yet |
| 09:41 | <Hixie> | i dunno, i don't actually read lastweek so i've no idea how many posts he's written |
| 09:41 | <Hixie> | (i have far better things to do with my life!) |
| 09:41 | <Hixie> | such as today, going to maker faire and then seeing wicked, which, by the way, is pretty good as a musical |
| 09:41 | <Hixie> | though i prefer into the woods |
| 09:41 | <Hixie> | (and avenue q, of course) |
| 09:41 | <Hixie> | (though anyone who's read the spec realises this) |
| 09:42 | <gsnedders> | Oh, I need to get ticket to go and see Avenue Q/ |
| 09:42 | gsnedders | is reminded :P |
| 09:42 | <abarth> | i missed avenue q, but i did make it to wicked |
| 09:42 | <Hixie> | man when i start writing examples later this year, i hope i don't run out of quotes from these musicals to put in the spec |
| 09:42 | <Hixie> | abarth: did you see the one in SF? |
| 09:42 | <abarth> | Hixie: yes |
| 09:42 | <Hixie> | abarth: i was _really_ impressed by the production value there |
| 09:42 | <Hixie> | holy crap |
| 09:43 | <Hixie> | that's some serious set dec |
| 09:43 | <gsnedders> | Hixie: Do you know anything about the London production? |
| 09:43 | <othermaciej> | I saw Up today |
| 09:43 | <abarth> | Hixie: Yeah, I always worry when I see a play outside of NY or London that I'm missing out, but the production was really good |
| 09:43 | <othermaciej> | I rarely look up MLW unless someone mentions it here |
| 09:43 | <Hixie> | up was pretty fantastic too |
| 09:43 | <Hixie> | though that's to be expected |
| 09:44 | <othermaciej> | but I am flattered that he adopted the name I assigned him |
| 09:44 | <Hixie> | what with it being the movie of the year and all |
| 09:44 | <Hixie> | though i may have to update the example in the spec |
| 09:44 | <othermaciej> | (I believe I coined the term "Mr. Last Week" back in the day) |
| 09:44 | <Hixie> | because up comes before a number of the other movies in the list |
| 09:44 | <gsnedders> | Hixie: what example? |
| 09:44 | <Hixie> | gsnedders: no idea (re london) |
| 09:45 | <othermaciej> | Up was surprisingly not-for-little-kids |
| 09:45 | <gsnedders> | Hixie: Damnit, I was counting on you! :P |
| 09:45 | <gsnedders> | Ah, Up doesn't come out here until October :( |
| 09:45 | gsnedders | always hates the fact that Pixar films don't come out here for months |
| 09:45 | <Hixie> | othermaciej: yeah, the trailers had me worried that it was going to be dreamworks-like, but woah was that a mistaken worry |
| 09:46 | <othermaciej> | Dreamworks is not in the same league as Pixar |
| 09:46 | <Hixie> | othermaciej: i guess i should have more faith in ye people of emeryville |
| 09:46 | <gsnedders> | Hixie: "The top 10 movies of all time"? |
| 09:46 | <Hixie> | gsnedders: no comment |
| 09:46 | <gsnedders> | :P |
| 09:46 | <Hixie> | and re the release date, i guess it's because they have to do all the subtitling and so on |
| 09:46 | Hixie | ducks |
| 09:47 | <othermaciej> | I say that having recently seen "Monsters vs. Aliens" and moderately enjoyed it |
| 09:47 | gsnedders | slaps Hixie |
| 09:47 | <othermaciej> | so I wonder if John Lasseter really has full creative control of all Disney animation now |
| 09:47 | <othermaciej> | because apparently they have a new cel-animated feature coming up |
| 09:47 | <othermaciej> | and I am wondering whether I should expect it to be any good |
| 09:48 | <gsnedders> | Hixie: Actually, Cars at least did in parts have a different audio track (and script, and hence animation) of the character's voices at parts here! |
| 09:48 | <Hixie> | wow |
| 09:48 | <Hixie> | i know monsters, inc did |
| 09:48 | <Hixie> | the non-us version had different graphics in a couple of places |
| 09:49 | <Hixie> | e.g. the "walk/don't walk" sign is symbols, iirc |
| 09:49 | <Hixie> | and the alert in the code 21 or whatever it's called was a different symbol |
| 09:49 | gsnedders | only got around see seeing Ratatouille in English over Christmas, having been bullied by French relatives into seeing it in Paris in French when it came out :P |
| 09:49 | <Hixie> | ew! ewewew! |
| 09:49 | <Hixie> | dubbing! |
| 09:49 | <Hixie> | ewwwww! |
| 09:49 | Hixie | backs away from the forever tainted gsnedders |
| 09:50 | <gsnedders> | Hixie: Did you see an undubbed version of Црна мачка, бели мачор the? |
| 09:50 | <gsnedders> | *then |
| 09:50 | <Hixie> | yes |
| 09:50 | <Hixie> | i doubt there has ever been a dub for it |
| 09:50 | <gsnedders> | That is true. |
| 09:50 | <jgraham> | Please do not get me started on the insanity of Pixar release dates. The films are amazing. The delays of months and months and months are inexcusable. |
| 09:51 | gsnedders | wonders where he'll be in October… |
| 09:51 | <Hixie> | gsnedders: actually that movie is the first example of a movie that was so good that by the end i actually forgot i was reading subtitles |
| 09:51 | zcorpan_ | marks all html5 alerts from the weekend as read |
| 09:51 | <Hixie> | lol |
| 09:51 | <zcorpan_> | if there was something interesting, let me know :) |
| 09:52 | <Hixie> | all the interesting stuff was mentioned in this channel i think |
| 09:52 | <zcorpan_> | k |
| 09:52 | <Hixie> | though if you haven't seen the wave keynote video, i recommend that |
| 09:52 | <Hixie> | http://code.google.com/events/io/ |
| 09:53 | <zcorpan_> | i saw that on friday |
| 09:53 | <Hixie> | and the first keynote is worth it just for the one slide |
| 09:53 | <zcorpan_> | or maybe i just saw part 2 |
| 09:53 | <zcorpan_> | s/part/day/ |
| 09:53 | <gsnedders> | Hixie: I must admit I was impressed by how well Ratatouille was done in French, though at parts it was obvious it wasn't the original language (though it was re-animated so the mouth movements all fit), as the language at times got too complex for the age-range trying to stay fairly close to the English |
| 09:55 | gsnedders | tends to think dubbed films aren't very good |
| 09:56 | <annevk42> | Spirited Away was not too bad I thought, but then I never watched the original |
| 09:57 | <gsnedders> | I didn't think the dubbing of that was overly good in that |
| 10:18 | zcorpan_ | tries the custom search element on his site and finds that it is more to it than just copy-and-paste-and-be-done-with-it |
| 10:24 | <zcorpan_> | Hixie: the google web elements snippets aren't valid html5 |
| 10:25 | <Hixie> | i am shocked |
| 10:25 | <Hixie> | (if you really want me to tell the team, send me a mail to ianh⊙gc and i'll forward it appropriately) |
| 10:27 | <Hixie> | ok i should probably head to bed |
| 10:27 | <Hixie> | nn |
| 10:35 | <zcorpan_> | <feature name="API:example.org.geolocation" - http://www.w3.org/mid/4A1FF4AA.2030409⊙oc |
| 10:35 | <zcorpan_> | reversed dns labels++ |
| 10:37 | <hsivonen> | oh noes. Feature enumeration rides again. :-/ |
| 10:37 | <Philip`> | zcorpan_: Reversed DNS labels-- because they didn't write it in proper reverse order |
| 10:38 | <zcorpan_> | maybe they should be called Labels DNS Reversed |
| 10:39 | <jgraham> | Philip`: it is presumably using the .example tld |
| 10:40 | <gavin_> | instead of the example.org domain? |
| 10:40 | <gavin_> | seems unlikely |
| 10:40 | <zcorpan_> | although he never said it was a reversed dns label, he just minted the API: url scheme |
| 10:40 | <jgraham> | (Although that would be inconsistent with the rest of the message and willfully obscure, so I may not be being entirely serious) |
| 10:41 | <zcorpan_> | i guess my point is that people will misunderstand how to write reversed dns labels |
| 10:42 | <hsivonen> | I seriously doubt the utility of the <feature> feature |
| 10:42 | <hsivonen> | I guess I should send a comment, but I'm probably late considering BONDI's schedule |
| 10:44 | <annevk42> | "better late than never"? |
| 10:45 | <hsivonen> | yeah. |
| 10:45 | <hsivonen> | lunch first |
| 10:56 | <Mrmil> | May I ask for some light feedback? http://server.ebrana.cz/olda/_apps/html5/ |
| 11:00 | <zcorpan_> | Mrmil: looks pretty good |
| 11:01 | <gsnedders> | PHP html5lib is slow. Really slow. |
| 11:01 | gsnedders | wonders if it is in an infinite loop |
| 11:03 | <Philip`> | gsnedders: Wait until it finishes, and then you'll know it's not an infinite loop |
| 11:03 | <Mrmil> | zcorpan_: Yay, thanks :) Now I have a question - imagine that the 3 service float boxes didn't have the annotation paragraph(s), what code would you use? headers wouldn't make sense anymore because there would be no text associated for them. Would you use ul li instead? |
| 11:03 | gsnedders | adds output of current line/column |
| 11:04 | <gsnedders> | Over line 25k now… (86k lines, IIRC) |
| 11:05 | <jgraham> | gsnedders: php-html5lib alone is 86k lines? |
| 11:05 | <gsnedders> | jgraham: No, the spec |
| 11:05 | <jgraham> | Oh I see |
| 11:06 | <gsnedders> | line 38k |
| 11:06 | <gsnedders> | OK, I think it probably will finish… |
| 11:07 | <gsnedders> | 44k… |
| 11:08 | gsnedders | wonders how ever long it'll take with a profiler when it takes this long without |
| 11:10 | <gsnedders> | I guess we take the prize for slowest impl. of HTML 5 parsing algorithm, at least :P |
| 11:11 | <jgraham> | gsnedders: I bet I am slower |
| 11:11 | <gsnedders> | jgraham: Slowest automated impl. |
| 11:11 | gsnedders | would not be surprised if he was slower than jgraham |
| 11:11 | <gsnedders> | line 60k! |
| 11:12 | <gsnedders> | Uh, yeah, I'm bored now. |
| 11:12 | <annevk42> | empirical evidence suggests I'm faster than jgraham and Hixie at getting correct answers :p |
| 11:13 | <jgraham> | annevk42: Hey, that was a microbenchmark if ever I saw one |
| 11:13 | <jgraham> | (although I would bet that you are still faster than me) |
| 11:14 | <gsnedders> | Line 70k! |
| 11:15 | <Philip`> | gsnedders: Using the entire spec for profiling seems like a pretty stupid idea :-p |
| 11:15 | <annevk42> | you know, if I'm really bored one day I think I rather find something else to do than test my performance on the HTML5 parsing algorithm :p |
| 11:17 | <gsnedders> | Has any machine-run implementation taken over 15 minutes to parse the spec before? |
| 11:17 | <Philip`> | That's an easy record to beat |
| 11:17 | <Philip`> | Start parsing, press ctrl+Z, wait 16 minutes, type "fg" |
| 11:17 | <gsnedders> | (We're now under 10k lines to go!) |
| 11:18 | <Philip`> | gsnedders: Sounds like your performance is slightly sub-linear |
| 11:18 | <gsnedders> | Line 80k! |
| 11:18 | <gsnedders> | No, I think it's linear. |
| 11:19 | <gsnedders> | Just slow. |
| 11:19 | <Philip`> | Hmm, good point, I suppose linear performance would be within the error bars of my IRC timestamp precision |
| 11:20 | <gsnedders> | real 17m16.388s |
| 11:20 | <gsnedders> | user 15m43.416s |
| 11:20 | <gsnedders> | sys 0m12.038s |
| 11:20 | <Philip`> | Is the output correct? |
| 11:20 | <gsnedders> | Dunno. |
| 11:20 | <gsnedders> | It just created a DOM, then deleted it :P |
| 11:21 | <gsnedders> | ezyang: So, only around 16m of CPU time to parse the spec. I guess we're not that bad. |
| 11:22 | <Philip`> | Compared to, like, one second in Java? |
| 11:23 | <hsivonen> | 16 minutes to parse the HTML5 spec? |
| 11:23 | <gsnedders> | hsivonen: yes. |
| 11:24 | <annevk42> | whoa |
| 11:24 | hsivonen | goes back to manually inlining the read() method in the Java tokenizer and tweaking it not to write into the input buffer to allow speculative parsing in C++ |
| 11:27 | <jgraham> | gsnedders: May I suggest profiling on <10% of the spec |
| 11:27 | <jgraham> | Since that will probably contain enough interesting stuff to see where all the time is being taken |
| 11:29 | <hsivonen> | are PHP arrays slow when used as arrays of ints? |
| 11:29 | <hsivonen> | that is, in that case, do they perform like real arrays or like hashtables or like something else? |
| 11:29 | <gsnedders> | hsivonen: We don't use them. Hashtables. And I don't think they're slow, just memory hogs. |
| 11:29 | <gsnedders> | (We're talking around 70 bytes of overhead per value) |
| 11:30 | <thomaslee_> | Why does TextMetrics have a width but no height? |
| 11:30 | <othermaciej> | 70 bytes per value? |
| 11:30 | <othermaciej> | seriously? |
| 11:30 | <gsnedders> | othermaciej: Seriously. |
| 11:30 | <hsivonen> | gsnedders: ok. I was wondering how hard it would be to make the C++ translator to target PHP |
| 11:31 | <gsnedders> | hsivonen: You don't want to, basically. |
| 11:31 | <Philip`> | thomaslee_: Height of what? (Line height? Bounding box height? Em square height?) |
| 11:31 | <hsivonen> | as I understand it, GWT compiler Java char[] into a JS array |
| 11:31 | <hsivonen> | which seems reasonable performant compared to the DOM and stuff |
| 11:32 | <hsivonen> | gsnedders: is indexing into PHP strings O(1)? |
| 11:32 | <gsnedders> | hsivonen: A black box function (in an extension) which took a string and returned a DOMDocument (or something else if you created another object model) would probably be the only way to get sane perf |
| 11:32 | <gsnedders> | hsivonen: yes |
| 11:32 | <gsnedders> | Uh, doing it with a bit less debug stuff it did it in 1m44s |
| 11:32 | <Philip`> | thomaslee_: There are lots of metrics that could be returned (which is why there's a whole TextMetrics structure instead of a single value), but currently it's just intended to address use cases like drawing labels on graphs |
| 11:32 | <thomaslee_> | Philip`: line height I guess. I'm trying to center some text vertically within a known bounds. |
| 11:33 | <Philip`> | thomaslee_: ... and not to address e.g. complex multi-line text layout, and so it doesn't expose any of the metrics you need for that |
| 11:33 | <gsnedders> | hsivonen: Mainly because all arrays, no matter what type of the key, is just put in a hashtable, and AFIAK that has O(1) lookup |
| 11:33 | zcorpan_ | sends an email to pf basically repeating the ToC of html5 when they ask if there are any other areas with a11y concerns in html |
| 11:33 | <Philip`> | thomaslee_: Can't you use ctx.textBaseline = 'middle' for that? |
| 11:33 | <thomaslee_> | Philip`: was hoping to be able to use something like: (y + (cellHeight / 2) - (metrics.height/2)) |
| 11:34 | <thomaslee_> | Philip`: maybe? *reads docs* |
| 11:34 | <othermaciej> | hsivonen: JS arrays are arrays |
| 11:34 | <gsnedders> | I guess the main reason why it was so much slower was there were an extra couple million functions calls… |
| 11:34 | <Philip`> | thomaslee_: (Don't know how widely that's implemented, but metrics.height isn't implemented at all either :-) ) |
| 11:34 | <hsivonen> | othermaciej: OK. I thought JS arrays had some whacko hashtable dualism, too |
| 11:35 | <othermaciej> | they are also hashtables, but foo[n] where n is a smallish integer will be at an array access level of speed |
| 11:35 | <Philip`> | thomaslee_: http://www.whatwg.org/specs/web-apps/current-work/multipage/the-canvas-element.html#dom-context-2d-textbaseline |
| 11:35 | <othermaciej> | at least in decent implementations |
| 11:35 | <othermaciej> | but you can also set an arbitrary named property on an array |
| 11:36 | <gsnedders> | hsivonen: SPLFixedArray which is in a disablable extension, and new in PHP 5.3, only allows integer keys and is a really array is actually slower from a reading POV, but quicker writing and a lot lower memory usage |
| 11:36 | <thomaslee_> | Philip`: ah, so I can calculate the vertical centre and render the text using textBaseline middle |
| 11:36 | <hsivonen> | othermaciej: the GWT-compiled version turns DOMStrings into UTF-16 code unit arrays and runs the tokenizer on those |
| 11:36 | <Philip`> | thomaslee_: In theory, yes |
| 11:36 | <thomaslee_> | let's see how it goes :) |
| 11:36 | <othermaciej> | hsivonen: that I'm confused about |
| 11:36 | <hsivonen> | othermaciej: so if you put a very long string into the live DOM input field, you get a very long array internally |
| 11:37 | <othermaciej> | it's not in general a good idea to use JS arrays of code point values instead of JS strings |
| 11:37 | <othermaciej> | especially for stuff you got out of the DOM |
| 11:37 | <jgraham> | hsivonen: That seems like it could be bad. |
| 11:37 | <othermaciej> | strings will almost certainly have better perf |
| 11:37 | <thomaslee_> | Philip`: you've saved the day again :P |
| 11:38 | <othermaciej> | I wonder if Google actually uses GWT for any of their products |
| 11:38 | <jgraham> | I seem to remember that V8 does some magic for array indicies > about 1000 |
| 11:38 | <hsivonen> | ideally, the JS version would use strings after I've made the input buffer read-only |
| 11:38 | <othermaciej> | an array of code point values will be bigger and slower than a JS string |
| 11:38 | <jgraham> | othermaciej: Isn't wave supposed to be written entirely with GWT? |
| 11:38 | <othermaciej> | in near any implementation |
| 11:38 | <hsivonen> | but that would entail some kind of Java-to-Java transformation before showing the code to GWT |
| 11:38 | <othermaciej> | jgraham: dunno, I have not studied its internals |
| 11:39 | <hsivonen> | using arrays makes sense here for both Java and C++ and the JS version is a side show currently |
| 11:39 | <othermaciej> | hsivonen: I would think the sensible model is to use String on the Java side and strings on the JavaScript side |
| 11:39 | <othermaciej> | instead of char arrays in either place |
| 11:39 | <othermaciej> | oh |
| 11:39 | <thomaslee_> | Philip`: out of curiosity, how exactly are you involved in HTML 5? |
| 11:40 | <othermaciej> | if you actually need something mutable then yeah you can't use a JS string |
| 11:40 | <hsivonen> | othermaciej: given how java.io works and how SAX works, I want to have char[] in Java |
| 11:40 | <hsivonen> | othermaciej: also, the XML coercion magic will still write into the input buffer |
| 11:40 | <hsivonen> | but the XML coercion stuff is unnecessary in Gecko |
| 11:40 | <hsivonen> | where speculation will be necessary |
| 11:41 | <gsnedders> | hmm, DOMNode::appendChild is by far the most expensive method/function |
| 11:42 | <Philip`> | thomaslee_: I just hang out here and annoy people, and sometimes send feedback on the spec and write http://philip.html5.org/tests/canvas/suite/tests/ and stuff |
| 11:42 | <hsivonen> | fwiw, in Gecko, DOM appends show up on profiles independently of the actual CSS box creation for those appends |
| 11:43 | <gsnedders> | 46% of the time taken to parse the spec (which appears to be closer to 1m46s) is appendChild |
| 11:43 | <othermaciej> | Gecko's DOM is pretty slow all around |
| 11:43 | <othermaciej> | <http://www.hixie.ch/tests/adhoc/perf/dom/artificial/core/001.html>; |
| 11:43 | <othermaciej> | (as far as core DOM operations at least) |
| 11:44 | <gsnedders> | (This is PHP's DOM, though) |
| 11:45 | <Philip`> | gsnedders: Maybe you should parse not into a DOM, but into an XML string, and then use the presumably-fast built-in XML parser to construct the DOM |
| 11:45 | <othermaciej> | (in retrospect though, that test can't distinguish slowness of the bindings glue from slowness of the core DOM op itself) |
| 11:45 | <hsivonen> | gsnedders: the DOM is C, right? |
| 11:45 | <gsnedders> | Philip`: That limits us to streamable output |
| 11:45 | <gsnedders> | hsivonen: yes |
| 11:45 | <hsivonen> | othermaciej: Gecko's DOM Core bindings glue is getting thinner |
| 11:45 | <Philip`> | gsnedders: Just do that for well-formable fragments, not for entire documents |
| 11:46 | <jgraham> | gsnedders: You could have a custom tree implementation that could serialize to XML |
| 11:46 | <othermaciej> | (WebKit nightly is nearly 7x as fast as recent Minefield on that test) |
| 11:47 | <gsnedders> | jgraham: That'll be more expensive as that means function calls. |
| 11:47 | <hsivonen> | aargh. I need to *think* about character reference tokenization again |
| 11:47 | <gsnedders> | The function call overhead _has_ to be avoided. |
| 11:47 | <gsnedders> | (Mainly by avoiding functions!) |
| 11:48 | <zcorpan_> | i wonder why accessibility people put so much weight on a small set of features (alt, summary, etc) while completely ignoring other features (drag and drop, details, etc) |
| 11:48 | <hsivonen> | zcorpan_: a no-regression policy, I think |
| 11:49 | <Philip`> | zcorpan_: Maybe the small set of features is all that there's any chance of getting implemented in the next decade |
| 11:49 | <thomaslee_> | Philip`: I see. Seems like you keep yourself busy :) |
| 11:50 | <zcorpan_> | Philip`: some features have already been implemented and not received accessibility expert review |
| 11:51 | <othermaciej> | it's not unreasonable to be more concerned about regressions, or what is perceived as such |
| 11:51 | <gsnedders> | jgraham: Function call overhead is big enough for it actually be to worthwhile to have an if statement for A–Z and only call strtolower on characters we know it'll have an effect on |
| 11:52 | <zcorpan_> | othermaciej: agreed, but i'm seeing something close to all-focus-on-regressions-and-completely-ignore-everything-else |
| 11:52 | <Philip`> | gsnedders: Is that function call overhead, or strtolower overhead? |
| 11:52 | <gsnedders> | Philip`: function call overhead more than strtolower overhead |
| 11:53 | <jgraham> | gsnedders: I don't see how it would add more function calls. It would just mean replacing calls to the DOM API to calls to your own code |
| 11:53 | <jgraham> | The big problem is that your own code might be even slower than the DOM API |
| 11:53 | <othermaciej> | zcorpan_: their approach may seem reasonable to them |
| 11:53 | <Philip`> | gsnedders: What if you say "if (char == 'A') char = 'a'; else if (char == 'B') ..."? :-) |
| 11:53 | <gsnedders> | jgraham: Well, instead of appendChild in C we have more function calls within our implementation of it |
| 11:53 | <Philip`> | zcorpan_: Some people talked about accessibility of <canvas> too, which is a new thing and not a regression |
| 11:54 | <gsnedders> | jgraham: Hence in total we end up with more function calls |
| 11:54 | <zcorpan_> | Philip`: yes that's basically the exception to the rule |
| 11:54 | <zcorpan_> | though i haven't seen that much discussion about <canvas> |
| 11:55 | <gsnedders> | Philip`: I doubt that's quicker, haven't tried though |
| 11:56 | <Philip`> | zcorpan_: I think most of the discussion has been saying "<canvas> must be accessible", and the only concrete proposal is to make the spec say "authors must provide an accessible alternative to any <canvas> content" (which the spec already requires) |
| 11:57 | gsnedders | installs kdesdk4 on his old Power Mac G5 as that has more RAM than this laptop, hence runs KCachegrind with large profiles a lot better. |
| 11:57 | <zcorpan_> | Philip`: indeed |
| 11:57 | <Dashiva> | Philip`: And the proposal to make not doing so illegal :) |
| 11:58 | <Philip`> | zcorpan_: so there hasn't been any interesting technical discussion on accessibility there |
| 11:58 | <annevk42> | zcorpan_, there's been discussion about <video> and <audio> too |
| 11:58 | <zcorpan_> | annevk2: true |
| 11:58 | <hsivonen> | should mv_embed be considered ready for production use for enabling Ogg <video> in IE using Cortado or VLC plugin? |
| 11:58 | <Dashiva> | There was some talk about making <canvas> contain a tree of elements that could be used to extract structure info |
| 11:58 | <Dashiva> | Apparently some OSes do it like that |
| 11:59 | <Dashiva> | But as I recall, that just reduced to authors who cared doing it, and those who didn't care still wouldn't do anything |
| 11:59 | <jgraham> | I think in generaly the a12y people have benn concered with things that fit cleanly into familar WCAG categories |
| 12:00 | zcorpan_ | wonders what a12y stands for |
| 12:00 | <zcorpan_> | accessibillity? |
| 12:01 | <jgraham> | Can I not spell/count/something |
| 12:01 | <jgraham> | What a surprise |
| 12:01 | <Dashiva> | Just remember that a11y is a name |
| 12:01 | <Dashiva> | And l10n is an animal :P |
| 12:02 | <jgraham> | What about l18n? |
| 12:02 | <jgraham> | or whatever it is |
| 12:02 | <Dashiva> | i18n? |
| 12:02 | <zcorpan_> | i18n |
| 12:02 | <Dashiva> | That's just weird |
| 12:03 | <jgraham> | Oh yeah, I. That makes more sense |
| 12:04 | jgraham | can only remember simple facts like "it has 18 in it and is something about making software for funny foreign types" |
| 12:05 | <zcorpan_> | like swedes? |
| 12:05 | <Dashiva> | And then there's that m12n or whatnot which I never remember what is |
| 12:05 | beowulf | wonders what the real title of the film se7en is |
| 12:06 | <annevk42> | Dashiva, l12n? |
| 12:06 | <Dashiva> | No, mNumberLetter |
| 12:06 | <zcorpan_> | m12n is modularization |
| 12:06 | <zcorpan_> | xhtml2 wg term :) |
| 12:06 | <Dashiva> | Aha |
| 12:07 | <Philip`> | beowulf: Semicitizen? |
| 12:07 | <Philip`> | beowulf: or Sempergreen? |
| 12:07 | <Dashiva> | 7 would be the total number of letters, wouldn't it? |
| 12:07 | zcorpan_ | wonders whether Philip` has an English dictionary which he can query with regexps |
| 12:07 | <zcorpan_> | Dashiva: no |
| 12:08 | <zcorpan_> | Dashiva: the numbers of letters that are replaced with the number |
| 12:08 | <Philip`> | zcorpan_: /usr/share/dict/words |
| 12:08 | <Philip`> | (on any sensible OS) |
| 12:09 | <jgraham> | http://www.morewords.com/ does what you want (probably) |
| 12:09 | <jgraham> | (although it didn't find Philip`'s matches so it obviously sucks) |
| 12:11 | Philip` | first heard the word "l10n" when Ultima Online had a "Public L10n Test" server, but it was drawn in a handwritingy font so it looked very much like "Lion" |
| 12:12 | <Philip`> | (UO's concept of "l10n" was to remove all the nice grammar in English, like adjectives in front of nouns for the names of magical weapons, and replace everything with boring-but-easy-to-localise phrases instead) |
| 12:13 | <annevk42> | btw, for future reference: http://en.wikipedia.org/wiki/Numeronym |
| 12:32 | <hsivonen> | yay for ecj: |
| 12:32 | <hsivonen> | 9392: goto 0 |
| 12:32 | <hsivonen> | 9395: goto 9277 |
| 12:32 | <hsivonen> | 9398: goto 0 |
| 12:33 | <hsivonen> | seems like a trivial optimization not to generate gotos like that |
| 12:35 | <Dashiva> | zcorpan_: That doesn't seem to match most of the ones listed here? |
| 12:52 | <zcorpan_> | Dashiva: it does afaict |
| 12:52 | <zcorpan_> | l10n - "ocalizatio".length == 10 |
| 12:53 | <Dashiva> | Hmm, indeed. I must be sleepy still. |
| 13:09 | <Dashiva> | I wish lastweek would link some good music for once |
| 13:25 | <Philip`> | Dashiva: Maybe he'll take requests if you have something specific to recommend |
| 13:37 | <hsivonen> | I wonder if anyone makes good old-fashioned Java bytecode optimizers anymore |
| 13:37 | <hsivonen> | that'd flatten multilevel gotos and stuff |
| 13:43 | <roc> | probably |
| 13:43 | <roc> | I doubt they matter |
| 13:43 | <roc> | the JIT takes all that apart anyway |
| 13:44 | <hsivonen> | roc: the thing is that the JIT doesn't kick in at all if a method grows over 8000 byte codes in size |
| 13:45 | <roc> | well, that's dumb |
| 13:45 | <hsivonen> | roc: Or I could document that if anyone uses this library, they have to configure their JIT in a special way |
| 13:45 | <roc> | I guess you should avoid making such methods :-) |
| 13:45 | <hsivonen> | roc: but most likely people would just ignore that and say that HTML5 is slow |
| 13:46 | <hsivonen> | roc: as long as the JIT kicks in, one huge method of gotos JITs to faster code than a bunch of methods |
| 13:49 | <jgraham> | hsivonen: Break it into to so something? |
| 13:50 | <jgraham> | s/to/two/ |
| 13:50 | <jgraham> | s/so/or/ |
| 13:50 | <hsivonen> | jgraham: then I have to transfer local variables over |
| 13:50 | <hsivonen> | jgraham: and using fields instead of local variable is very bad if you are trying to minimize the # of bytecodes |
| 13:51 | <hsivonen> | But I may end up splitting some NCR stuff out |
| 13:51 | <hsivonen> | Of course, if Hixie adds a new state for --!>, I'll have to make room for some more bytecodes... |
| 13:52 | <jgraham> | Yeah, we really need at least --!> and preferably -- > to work |
| 13:52 | <jgraham> | So I would start clearing room :) |
| 13:54 | <hsivonen> | hmm. I guess at some point there's no point in counting byte codes and I should just document that HotSpot has a lame magic limit and how to disable the limit |
| 13:55 | <hsivonen> | limits like that feel really applet-ish |
| 13:55 | <hsivonen> | who'd want the JIT not to do its thing on a server? |
| 14:00 | <Philip`> | Maybe someone who doesn't want their server to crash because it's trying to JIT a million-bytecode method all at once? |
| 14:04 | <hsivonen> | so many constrains: validation needs, Java library needs, speculative parsing in C++, JIT limits... |
| 14:06 | <Philip`> | I guess there's also the constraint of being a single codebase, rather than one for each different use |
| 14:10 | <hsivonen> | the 8000 bytecode limit is really the most annoying constraint |
| 14:10 | <hsivonen> | I'm very close to giving up and fixing it in documentation |
| 14:10 | <hsivonen> | "use this incantation to get a 10x speedup" |
| 14:11 | <roc> | JVM methods can't be more than 64k bytes of bytecodes |
| 14:11 | <hsivonen> | roc: the main tokenization loop is currently at 8701 bytecodes, so it's not too near that limit |
| 14:12 | <hsivonen> | oh, yeah, CRLF is hugely annoying, too. a lot of the extra bytecode fat is due to CRLF handling |
| 14:12 | <roc> | I once wrote a bytecode instrumentation tool that had the ability to take apart methods grew to over 64K bytes and automatically partition them into smaller methods |
| 14:12 | <roc> | without worrying too much about performance of course |
| 14:13 | <heycam> | roc, can you commit that to rhino's codebase please? :) |
| 14:13 | <heycam> | i run in to it from time to time |
| 14:14 | <heycam> | it'd be nice not to just tell it to use interpreted mode unconditionally in advance, in those situations |
| 14:14 | <heycam> | oops, please ignore the smiley, i'm trying to cut back |
| 14:14 | <hsivonen> | (it seems that major design decisions in Antlr have been made due to the 8000 bytecode limit) |
| 14:18 | <hsivonen> | I wonder if dalvik and whatever AppEngine uses have this kind of magic limits |
| 14:19 | <heycam> | while trying to optimise some of rhino's codegen once, i noticed it was generating a 4000 byte method that was a massive switch that just returned false from every branch |
| 14:19 | <heycam> | (true, that wasn't compilation of js to java, and just a generated method to support compiled js, but still... seemd a bit excessive) |
| 14:26 | <ezyang> | gsnedders: What was the next most expensive function call after appendChild()? |
| 14:27 | <gsnedders> | uh, dunno. |
| 14:27 | <ezyang> | Wait, you didn't profile it? |
| 14:27 | <gsnedders> | Yes, I did. |
| 14:27 | <ezyang> | Ok |
| 14:27 | <gsnedders> | I can't remember what the profile said :P |
| 14:28 | <ezyang> | You should post that up somewhere public |
| 14:28 | <gsnedders> | (It kinda crashes my laptop opening it in KCachegrind though as it eats so much memory.) |
| 14:29 | <gsnedders> | (So I'm installing KCachegrind on my old Power Mac G5 which has more memory.) |
| 14:29 | <ezyang> | Yeah, swapping can be pretty painful |
| 14:29 | <ezyang> | How big is the profile file? |
| 14:30 | <gsnedders> | It got to 600MB before I killed php stopping it becoming bigger. |
| 14:30 | <ezyang> | Hmm. |
| 14:30 | <ezyang> | My suggestion is to take a smaller HTML file and use that |
| 14:30 | <gsnedders> | (And KCachegrind tends to use double the size of the input when loading.) |
| 14:30 | <gsnedders> | That's boring! :P |
| 14:31 | <ezyang> | If html5lib is not pathological, it should still be representative |
| 14:31 | <ezyang> | Anyway, my current ordering of priorities is: 1. Clean up the APIs, 2. Kick out a release, 3. Optimize the hell out of it |
| 14:32 | <ezyang> | Also, I start working today, so I won't have as much time to work on html5lib |
| 14:33 | <gsnedders> | on my blog the profile is totally different |
| 14:33 | <ezyang> | Oh? |
| 14:33 | <gsnedders> | HTML5_Tokenizer::parse is slowest |
| 14:33 | <ezyang> | Is this total runtime or runtime within? |
| 14:33 | <gsnedders> | Just within it. |
| 14:34 | <gsnedders> | TreeConstructor::emitToken is next slowest |
| 14:34 | <ezyang> | Well, duh |
| 14:34 | <gsnedders> | then consumeCharacterReference |
| 14:34 | <ezyang> | Those are our two big honking switch cases |
| 14:34 | <gsnedders> | Then Tokenizer::emitToken |
| 14:34 | <ezyang> | consumeCharacterReference isn't surprising either |
| 14:34 | <gsnedders> | then InputStream::charsUntil |
| 14:35 | <ezyang> | These are all fairly obvious, and non-trivial to fix, unfortunately |
| 14:35 | <ezyang> | I'll run the profiler myself (I tend to favor Wikipedia pages) and see what I can do, after (1) and (2) are done |
| 14:35 | gsnedders | wonders what a decent medium sized page is |
| 14:36 | gsnedders | does from a non-bias POV http://en.wikipedia.org/wiki/United_Kingdom |
| 14:38 | <gsnedders> | appendChild is 8th most expensive there |
| 14:39 | <gsnedders> | I think we could remove our own appendChild and just use DOMNode::normalize |
| 14:40 | <Philip`> | gsnedders: The mean page size is something like 20KB |
| 14:40 | <Philip`> | (in my samples) |
| 14:47 | <gsnedders> | ezyang: Where do you want the call for parsing to actually go? |
| 14:47 | <zcorpan> | hsivonen: Start: Go to "X-UA-Compatible <meta>?". - you've forgotten to curlify the quotes there |
| 14:48 | <hsivonen> | zcorpan: thanks |
| 14:49 | <hsivonen> | zcorpan: fixed |
| 14:52 | <gsnedders> | ezyang: What are these "godawful libxml bugs"? |
| 14:52 | <zcorpan> | hsivonen: IE=6 is the same as IE=5 |
| 14:52 | <zcorpan> | i.e. quirks mode |
| 14:52 | <hsivonen> | zcorpan: thanks. (that'll take a bit more time to update) |
| 14:53 | <annevk42> | really? that's somewhat unexpected |
| 14:54 | jgraham | wonders how long annevk42 will be known as "Anna" for in the html5lib wiki |
| 14:54 | <annevk42> | jgraham, considering how often that wiki is updated, pretty long |
| 14:54 | <jgraham> | That page was only created yesterday or so though |
| 14:55 | <zcorpan> | annevk2: well i just tested it. it wasn't what i expected, either |
| 14:55 | jgraham | is not sure why it is so important to know who the project is done by rather than ho to use it |
| 14:56 | <annevk42> | ah, i guess that must be from the person I added to the project |
| 14:56 | <annevk42> | he said he was going to work on documentation |
| 14:56 | <beowulf> | jgraham: babes might be interested |
| 14:59 | gsnedders | laughs, and shakes his head |
| 15:03 | <annevk42> | "formal W3C Note" hmm |
| 15:17 | <zcorpan> | re http://www.w3.org/mid/4A23E22C.2010502⊙gd - i thought there were more issues, like what to do with whitespace and lone % |
| 15:21 | <zcorpan> | hsivonen: forgetting a ">" of an end tag results in lots of messages |
| 15:21 | <zcorpan> | hsivonen: see e.g. http://www.webforum.nu/showthread.php?t=174764 |
| 15:21 | <zcorpan> | <p><a href="http://www.djurssommerland.dk/">Djurs sommarland</a är något man inte får missa om man ... |
| 15:22 | <zcorpan> | on a start tag it's even worse |
| 15:28 | <zcorpan> | hsivonen: maybe if a tag has an illegal attribute, silence all other illegal attribute messages and duplicate attributes messages for that tag |
| 15:28 | <zcorpan> | hsivonen: also, it seems pointless to emit any other messages than "attributes on end tag" for end tags |
| 15:32 | <zcorpan> | http://bugzilla.validator.nu/show_bug.cgi?id=579 |
| 15:37 | <hsivonen> | zcorpan: thanks |
| 15:53 | <hsivonen> | I wonder why Flickr slideshows are Flash |
| 15:53 | <hsivonen> | is it all because of the ability to go full-screen? |
| 16:05 | <jgraham> | Gah, the sanitizer in html5lib.filters.sanitizer seems to be broken |
| 16:22 | <jgraham> | hsivonen: It works with the video, of course |
| 16:23 | <jgraham> | But it would be a neat project to do a flickr slideshow viewer in html5 |
| 16:23 | <hsivonen> | jgraham: they had Flash slideshows long before video |
| 16:24 | <hsivonen> | but it would be cool if Flickr went Flashless |
| 16:25 | <jgraham> | hsivonen: I know. But at the moment it seems like a good reason not to use HTML |
| 16:26 | <jgraham> | The nice thing about a flashless slideshow is that I would get better colours in browsers that support embedded colour profiles |
| 16:26 | <zcorpan> | hmm, the <script src>//documentation</script> abnf only supports LF and multiline comments can't contain stars or slashes *at all* if i'm reading it correctly |
| 16:26 | <hsivonen> | support for color profiles doesn't necessarily mean better color :-/ |
| 16:27 | <hsivonen> | a lot of color management is GIGO, unfortunately |
| 16:27 | <jgraham> | hsivonen: For photography it generally means better colour |
| 16:28 | <hsivonen> | jgraham: if you actually have right profiles |
| 16:28 | <hsivonen> | throughout the workflow |
| 16:28 | <jgraham> | Since a large fraction of the people have reasonably well calibrated monitors and are using profile-aware applications like Lightroom |
| 16:28 | <jgraham> | (and photoshop) |
| 16:32 | <jgraham> | e.g. compare http://www.flickr.com/photos/jgraham/3567837536/sizes/l/ in various browsers |
| 16:32 | gsnedders | wonders what 15 minute walk is |
| 16:33 | <jgraham> | (although I actually think that I made some mistakes with the processing of that in terms of aesthetic judgement, but that's not really the point) |
| 16:34 | <gsnedders> | jgraham: if Bibbi said 15 minutes walk, what sorta distance do you think she'd mean? |
| 16:35 | <jgraham> | gsnedders: I guess the sort of distance that you could walk in 15 minutes |
| 16:35 | <jgraham> | gsnedders: Do you have an address or anything? |
| 16:35 | <gsnedders> | jgraham: No |
| 16:35 | <gsnedders> | Just, "Close to the office (about 15 minutes walk)" |
| 16:35 | <jgraham> | Well that could be anywhere but I guess within 1km or so |
| 16:36 | <zcorpan> | i have about 15 minutes walk to the office |
| 16:36 | <jgraham> | me too |
| 16:36 | <gsnedders> | 15 minutes at what speed for a start :P |
| 16:36 | <zcorpan> | walking speed of course |
| 16:36 | <jgraham> | but we live at pi/2 radians from each other |
| 16:37 | <jgraham> | gsnedders: I guess normal walking speed is about 2.5 miles/hour |
| 16:37 | <jgraham> | so 15minutes walk is about 5/8 of a mile. Or 1km |
| 16:37 | <jgraham> | Which is what I said before |
| 16:39 | gsnedders | works out basically where 1km away from the centre is |
| 16:39 | <gsnedders> | (Or rather, from where the office is) |
| 16:40 | gsnedders | notes Google Maps puts the office right in the middle of Centrum Ring, which I kinda doubt :P |
| 16:40 | <zcorpan> | if (window.widget.harharhar) |
| 16:42 | hsivonen | notes that Chrome doesn't show the video thumbnails in the YouTube demo |
| 16:45 | zcorpan | notes that the youtube demo uses <div><h1> instead of <details><legend> |
| 16:46 | <zcorpan> | the document outline will be shorter if they migrate to <details> |
| 16:46 | <jgraham> | gsnedders: According to http://maps.forum.nu/gm_sensitive_circle2.html I live about 1.5km from the office. But I guess it takes a little more than 15 minutes for me to get in |
| 16:52 | gsnedders | needs to work out when he'll leave Linköping |
| 16:52 | <jgraham> | gsnedders: It is like Hotel California |
| 16:52 | <gsnedders> | jgraham: ? |
| 16:53 | <gsnedders> | I think not the day after I finish, as the only flight that day is a 6:30am |
| 16:53 | <jgraham> | gsnedders: Just a throwaway pop-culture reference |
| 16:53 | <gsnedders> | jgraham: Ah. Pop-culture. That's the problem :P |
| 16:55 | <gsnedders> | Wait, what… |
| 16:55 | <gsnedders> | Flights from Stockholm (Skavasta) to Berlin (Schönefeld) for 55SEK!? |
| 16:56 | <jgraham> | gsnedders: Ryanair? |
| 16:56 | <gsnedders> | jgraham: Yeah |
| 16:56 | <jgraham> | gsnedders: Check the "Taxes" and "Fees" |
| 16:56 | <gsnedders> | Taxes and Fees it claims to be 0. |
| 16:56 | <gsnedders> | That's what is surprising me so much |
| 16:57 | <jgraham> | gsnedders: Oh you found one of those flights. In that case you will only have to pay for luggage, booking fee, etc. |
| 16:57 | <gsnedders> | Every day apart from Friday and Sunday |
| 16:57 | <gsnedders> | (On the Sunday it is 314.78 SEK) |
| 16:58 | gsnedders | probably won't get a plane ticket from Berlin to Britain yet, as he doesn't know whether he'll be going to London or to home… |
| 17:04 | <gsnedders> | WTF? |
| 17:04 | <gsnedders> | Flights for £0.00 (excluding taxes/fees) |
| 17:05 | <gsnedders> | At 6am, though :P |
| 17:10 | <jgraham> | gsnedders: 0 excluding taxes/fees == about 50GBP all told, usually |
| 17:12 | <gsnedders> | It was more than that |
| 17:12 | <gsnedders> | But I don't care. |
| 17:12 | gsnedders | is going for a flight at a more reasonable time of day :P |
| 17:25 | jgraham | has paid an extrodinary amount of money for the last few sets of Ryanair flights he has booked |
| 17:26 | <jgraham> | Because I don't tend to book early enough and because they are not nearly as cheap as they make out |
| 20:27 | <annevk42> | how does http://software.hixie.ch/utilities/js/live-dom-viewer/saved/124 demonstrate margin collapsing? |
| 20:27 | <annevk42> | oh, zcorpan is not here |
| 20:35 | <Hixie> | aaah, the epitomy of process over progress: http://www.w3.org/mid/4A2428A2.6080205⊙gd |
| 20:38 | <Dashiva> | But what if about:blank totally changes meaning during the process? ;) |
| 20:41 | <gsnedders> | Hixie: s/y/e/ |
| 20:42 | <Hixie> | so it seems |
| 20:42 | <Hixie> | thanks |
| 20:42 | <Hixie> | Dashiva: exactly |
| 20:43 | gsnedders | fixes a test case in php html5lib by fixing the test handler to actually follow the test format :P |
| 20:43 | <annevk42> | Hixie, yeah :/ |
| 20:43 | <annevk42> | Hixie, though if they think it really matters maybe they should leave the value space up to some other RFC... |
| 20:44 | annevk42 | wonders who'll write all of those |
| 20:44 | <Hixie> | i have faith that adam will do the right thing |
| 20:44 | <annevk42> | the right thing was just made harder |
| 20:45 | annevk42 | kind of like about:noreferrer |
| 20:45 | <annevk42> | likes* |
| 21:08 | jgraham | wonders who the people are that go "oh no that's just a draft standard I can't possibly implement the well defined thing that I need for interoperability" |
| 21:11 | <tantek> | jgraham, Perhaps those that then must maintain back/bug-ward compatibility with their implementation of the draft standard for 3+ years for enterprise users and others that don't upgrade implementations very often. |
| 21:14 | <jgraham> | tantek: In the case where the draft is really unstable that seems reasonable. But in the case of e.g. about: it's fixed by legacy anyway |
| 21:15 | <Hixie> | i like larry's observation that we don't understand the ietf process given how he doesn't understand the whatwg process :-) |
| 21:15 | <tantek> | jgraham, agreed. when a draft standard is simply documenting existing (hopefully widely known) interoperable behavior, the excuse of "draft" does not hold much weight. |
| 21:21 | <annevk42> | afaict Larry is wrong too |
| 21:21 | <annevk42> | tools.ietf.org is generally considered to be the space of official IETF documents |
| 21:21 | <annevk42> | moreover, I believe there's no documented guarantee that the links there won't break either |
| 21:21 | <annevk42> | is not generally* duh |
| 21:37 | <expensivelesbian> | Hello all, just a random question - if one wanted to write some sort of "visualiser" (VU meters, graphical EQ bars and the like), taking the output of an <audio> element, would this be possible currently, with HTML5 + JavaScript? If this question is off topic, someone please say nicely and I'll scurry off elsewhere |
| 21:38 | <annevk42> | http://lists.w3.org/Archives/Public/public-html/2009Jun/0036.html really, what's the obsession with introducing versioning? is it some kind of nirvana? |
| 21:40 | <annevk42> | expensivelesbian, it's not off-topic, but I don't think it's possible either currently |
| 21:40 | <roc> | it's not possible |
| 21:40 | <annevk42> | expensivelesbian, unless you skip the <audio> element and parse the stream directly in ECMAScript, but besides that being hard you might run into some limitations as well (e.g. no real byte stream representation) |
| 21:43 | <expensivelesbian> | ok, thanks for the answers. I'll have a ponder on this |
| 21:45 | <annevk42> | I guess due to <canvas> you can do a lot more with <video> |
| 21:45 | <annevk42> | but then audio manipulation is prolly much harder |
| 21:46 | <expensivelesbian> | well, it was that bounding boxes demo got me thinking, "there must be something _like_ that for poor old neglected <audio>" |
| 21:50 | <annevk42> | expensivelesbian, sorry |
| 21:50 | <annevk42> | expensivelesbian, there have been discussions, but nothing concrete yet |
| 21:59 | <annevk42> | quotes in http://edward.oconnor.cx/2009/BarCamp-San-Diego-5/ are nice -- navigation within that presentation is a bitch though, what's back? |
| 22:26 | <hober> | I don't think I got around to implementing back... |
| 22:27 | <hober> | I should really make a print stylesheet too.. |
| 23:22 | <ezyang> | gsnedders: (a very late response) Put it in HTMLParser. That should be the "public interface" |
| 23:23 | <ezyang> | gsnedders: Godawful libxml bug == "Namespace Error" when you try to set a namespaced attribute. |
| 23:24 | <jwalden> | so, looking at the websocket protocol, what constraints are enforced on the server regarding processing of the input? is it an error if tab-delimited "header" lines are received? what if "multi-line headers" are received? basically, websockets looks like HTTP, but at what point does accepting HTTP input that's more lax than the websocket client input constitute a violation of the protocol... |
| 23:24 | <jwalden> | ...requirements? |
| 23:26 | <Hixie> | what do you mean by "enforced"? |
| 23:26 | <Hixie> | the protocol spec says what you're expected to suport |
| 23:26 | <Hixie> | support |
| 23:26 | <Hixie> | as a server-side author |
| 23:27 | <jwalden> | s/enforced/required/, then |
| 23:27 | <jwalden> | the client is required to follow the exact steps given in the draft |
| 23:27 | <Hixie> | http://tools.ietf.org/html/draft-hixie-thewebsocketprotocol-12#page-15 |
| 23:28 | <jwalden> | the server, it seems, has latitude to accept more than that |
| 23:28 | <jwalden> | is the server required to accept input from a client that doesn't actually follow the strictures of the client-processing section? |
| 23:28 | <jwalden> | yes, I've been looking at that |
| 23:29 | <ezyang> | gsnedders: My idea is to implement SimpleTree and watch our performance problems melt away ^_^ |
| 23:29 | <Hixie> | jwalden: the server doesn't have to do any error handling, if that's what you mean |
| 23:29 | <jwalden> | okay |
| 23:30 | <jwalden> | that seems...dangerous |
| 23:30 | <jwalden> | not in any tangible sense |
| 23:30 | <Hixie> | the ability to control authors is basically nil |
| 23:30 | <Hixie> | so getting them to implement particular error handling behavior... forget it |
| 23:30 | <jwalden> | but in the sense of I-can-plausibly-believe-this-will-bite-us-at-some-point |
| 23:30 | <jwalden> | no, this is error handling server-side |
| 23:31 | <Hixie> | right |
| 23:31 | <Hixie> | authors write the server side |
| 23:31 | <jwalden> | oh, there are authors on both sides |
| 23:31 | <jwalden> | hum, this might work then |
| 23:32 | <Hixie> | i could write a new section |
| 23:32 | <Hixie> | which is "full requirements for server-side implementors that actually follow specs" |
| 23:32 | <Hixie> | if you like |
| 23:32 | <Hixie> | might not be a bad idea |
| 23:32 | <jwalden> | dunno, yet |
| 23:32 | <Hixie> | if you do think that would be useful, send a mail to whatwg, and i'll add it |
| 23:33 | <jwalden> | I'm just looking at this from the POV of the Mozilla bug, in which our HTTP server is being lightly hijacked to support all this, worried about being more generous in accepting input than it should be |
| 23:33 | <Hixie> | is this for testing or for production? |
| 23:34 | <Hixie> | for regression testing, it doesn't matter what the server does, it just has to check that the client sends exactly the right bytes |
| 23:34 | <jwalden> | I try not to differentiate the two when I write servers (or rather, when I write the one server I've written so far) |
| 23:34 | <Hixie> | a regression test suite server and a production server are completely different |
| 23:34 | <Hixie> | you can't, e.g., use apache to do serious regression testing of client-side http behaviour |
| 23:34 | <jwalden> | yes, but that doesn't mean one shouldn't make an effort not to make it safe-by-default |
| 23:35 | <Hixie> | i don't really see how a regression-testing server framework can be unsafe -- it has to accept any input whatsoever |
| 23:36 | <jwalden> | hm, I think we're talking about different meanings; "production" to me is use in which inputs are unconstrained, "regression-testing" in the sense of inputs are constrained but may intentionally be random junk |
| 23:36 | <jwalden> | constrained by whatever's in the test suite using it |
| 23:40 | <jwalden> | on an unrelated note, reading RFC-style formatting rather than whatwg-style formatting is no fun :-( |
| 23:40 | <jwalden> | especially since things like "find" don't work due to ASCII text-decorations |
| 23:42 | <Hixie> | jwalden: an "in production" server is one serving end-user traffic |
| 23:42 | <jwalden> | we agree on that |
| 23:42 | <Hixie> | jwalden: a regression testing server is one used to test another tool's implementation |
| 23:43 | <jwalden> | sure, just say that I like my regression testing servers to have as rigorous an implementation as a production server must have |
| 23:43 | <Hixie> | so an in-production server can do things like disconnect if the input is bogus |
| 23:43 | <Hixie> | but a regression testing server needs to report pass/fail for all inputs |
| 23:55 | <gsnedders> | http://larry.masinter.net/versioning-html.html |
| 23:55 | <gsnedders> | (from www-tag) |
| 23:56 | <ezyang> | 'lo gsnedders |
| 23:56 | <ezyang> | Did you see my late late late msgs? |
| 23:56 | <gsnedders> | ezyang: yes |
| 23:57 | <gsnedders> | ezyang: I've fixed another test case (actually, I've fixed the test runner). |
| 23:57 | <ezyang> | Makes sense |
| 23:57 | <ezyang> | The strDom() function is hack on hack on hack |