# [00:39] <Hixie> there's a prefix used without even a declaration
# [00:39] <Philip`> Hixie: I do get a download dialog if the file is served as application/octet-stream or foo/bar etc, but it's just shown as text if it's text/plain
# [00:43] <Hixie> it's deployed, as i understand it
# [00:45] <Philip`> How does the deployed implementation work? Given that they've seemingly got a properly namespaced schema, it might just be the spec writer didn't understand what was going on
# [01:14] <jruderman> Hixie: thanks, now we'll lose fewer moco employees to pedestrian fatalities on the 101 freeway
# [01:15] <Philip`> Why does Google Maps choose a certain route when there's an alternative that's two minutes shorter, and that it's happy to pick when I drag the route around a bit?
# [01:15] * Philip` isn't sure why it'd be optimising for anything except distance
# [01:17] <gavin> it takes into account speed limits and such, doesn't it?
# [01:17] <gavin> favors highways/major streets too, I think
# [01:17] <Hixie> jruderman: well the pin can be edited by users (and has been. a lot. maybe other people in your office are moving it to the wrong location. :-) )
# [01:20] * Quits: KevinMarks (n=KevinMar@nat/google/x-3fa2a77709c664ad) ("The computer fell asleep")
# [01:20] * gsnedders passes jruderman a future product or service
# [01:20] <Philip`> Hixie: I suppose it could be, but the shorter route is only two extra steps, and I'm too lazy to travel an extra two minutes just for that
# [02:26] <Hixie> roc_: why not? (as far as i can tell, the pin location caused maps to decide the nearest street was 101.)
# [02:27] <roc_> because even if the nearest street *is* 101 (and presumably there are valid locations for which that's true), it should know you can't stop there.
# [03:27] <Philip`> roc_: It ought to be easy for the compiler just to do pattern-matching to detect a Fibonacci function and replace it with an efficient library call
# [03:27] <roc_> Philip`: yeah, but that's cheating
# [03:27] <Philip`> (like what MSVC does with memcpy loops)
# [03:53] <BenMillard> Hixie, there's a similar problem with ARIA examples for static web content: existing semantics in HTML4 are often adequate, or the ARIA becomes unnecessary when the feature is changed to follow a usable design pattern.
# [03:53] <BenMillard> (Re: your comment about workers examples)
# [04:33] <BenMillard> the blog entry it links to has numerical analysis of the tables cited as supporting evidence versus corrected and redesigned versions I've done
# [08:29] <hsivonen> Hixie: I guess in the case of Java the problem scenario is that someone has serialization code buried at the end of a pipeline somewhere
# [08:30] <hsivonen> Hixie: and then they start sending HTML5 stuff through the pipeline
# [08:30] <hsivonen> Hixie: and the output is wrong
# [08:30] <hsivonen> Hixie: and ripping out 4 or 5 lines of code and replacing them with 2 lines of code referring to the serializer I wrote is onerous
# [08:31] <hsivonen> as they'd need to discover and add Yet Another Jar
# [08:31] <hsivonen> and perhaps convince their lawyers that code under the MIT license and copyright Mozilla Foundation is OK to use
# [08:32] <hsivonen> I get an impression that the HTML5 content would end up in the pipeline in a semi-uncoordinated way
# [08:33] <hsivonen> because if it were coordinated, one would think the serializer swapping could be coordinated, too
# [11:26] <Lachy> LOL! I like how John Foliot claims my proposed study of longdesc is flawed based on the fact that it will reveal the exact problems I designed to reveal. :-)
# [11:27] <Dashiva> Seems quite in line with most commercial studies :)
# [11:32] * Quits: Lachy (n=Lachlan@85.196.122.246) ("This computer has gone to sleep")
# [11:33] <hsivonen> one can read quite a bit of browser history in the UA string of Chrome
# [11:34] <hsivonen> I wonder if we ever come up with a way to stop the growth of the UA string
# [11:35] <hsivonen> does a typical HTTP GET request still fit on one IP packet?
# [11:35] <Dashiva> Yeah, one day we'll have RealUserAgent instead
# [11:35] <Dashiva> Since everything in the original UA string will be back-compat and talismans
# [11:39] <hsivonen> is Netscape 6 the only browser in history that has swam against the UA string cruft stream?
# [11:39] <hsivonen> Opera did recently, but it can still vary it per site, right?
# [12:52] <hsivonen> How many Windows ports of WebKit are there with different graphics layers? Safari, Arora, Chrome, AIR. Did the company that ported WebKit to Windows Mobile have a 5th graphics layer?
# [12:52] <hsivonen> does AIR use the same Abode vector graphics renderer as InDesign and Acrobat?
# [12:53] <hsivonen> does any of them actually delegate Bézier rasterization to GDI+ or WPF?
# [13:11] <annevk> zcorpan, if they depend on charset sniffing might be easier in HTML
# [13:12] <annevk> zcorpan, also, DOM Core already has various encoding features (all completely pointless of course, the DOM shouldn't deal with encoding)
# [13:12] <hsivonen> zcorpan: does query string treatment depend on charset in XML?
# [13:13] <annevk> eg, inputEncoding and xmlEncoding
# [13:16] <hsivonen> which reminds me that I should extend Jing to pass the document's original encoding around totally in violation of sane layering
# [13:47] <hsivonen> will IE8 mode break these custom namespace ActiveX controls?
# [13:47] * Philip` wonders if it's possible for a CGI script to print raw HTTP headers via lighttpd, instead of it being parsed and reserialised before transmission
# [13:49] <roc_> I'm still a bit nervous about returning fractional values from ClientRect. People see it and assume it's a bug. However, I still think it's the right thing to do and it doesn't seem to have caused any actual damage
# [13:56] <hsivonen> Philip`: do you know if it persists if the document contents are removed after that?
# [13:57] <hsivonen> does IE even support emptying the document and inserting a new root?
# [13:58] <Philip`> hsivonen: You can also do document.namespaces.add('v', 'http://www.w3.org blah blah it doesn\'t matter what you put here', '#default#VML');
# [14:14] <hsivonen> (without testing, I'm assuming this is one of those weird situations where the scripted view to the DOM and the engine understanding of the DOM are synced only upon spin through the event loop)
# [14:18] <Philip`> zcorpan: IE8b2 just freezes permanently
# [14:19] <Philip`> While trying to report the bug, something's broken Opera so it's using 100% CPU and making my fans very loud, but at least Opera is still being responsive
# [14:20] <Philip`> (To be more specific, IE8b2 freezes permanently and uses 100% CPU too)
# [14:21] <Philip`> (To be more accurate, IE8b2 freezes permanently and uses 100%-divided-by-number-of-CPU-cores CPU too)
# [16:32] <smedero> It feels like an edge use-case but is there a reason to allow <map> inside of <p> for something like a sparkline with an image map that had pointers to expanded info about different data points? (http://en.wikipedia.org/wiki/Sparkline)
# [16:32] <smedero> (i'm trying to bring up the whatwg list to see if that has been brought up before, but uh.. it is list.whatwg.org isn't responding for me...)
# [18:18] <Philip`> How lovely - BeautifulSoup is giving me a <div> element whose parent is a <div> element whose only child is an <a> element
# [18:20] <Philip`> Also, html5lib fails a zillion tests, so there's no way I can change anything without being quite likely to unknowingly introduce a regression
# [18:43] * Quits: nessy (n=nessy@124-171-27-224.dyn.iinet.net.au) ("This computer has gone to sleep")
# [18:46] * Quits: Lachy (n=Lachlan@pat-tdc.opera.com) ("This computer has gone to sleep")
# [18:50] <Dashiva> So earlier it was all "@headers is only used in generated content, so it doesn't matter that it's complex and easy to mess up" and today it's "but the USER wants to use @headers in handwriting, so we should let him"
# [19:05] <smedero> Dashiva: Well it is also funny because Gez said he modified his example to "simplify the problem"
# [19:06] <smedero> I don't think he redesigned it.... but it was modified in some form and I asked for a pointer the original so I can understand what was changed and I got zilch.
# [19:06] <smedero> (it wasn't just changes to anonymize the data... columns and such were removed to "make it easier to understand")
# [19:07] <Dashiva> Simplify the problem of getting people to accept it
# [20:46] <Lachy> I don't get why he can't comprehend that the proposed study is about seeing how effective longdesc is for users that currently have the ability to access it, nor why he doesn't understand why that would be useful
# [21:06] <webben> although in some recent versions, you've had to enable announcements for it.
# [21:06] <Hixie> Lachy: as far as i can tell he's not interested in improving accessibility, only in sounding like he's an expert and getting his ego stroked by getting his way
# [21:06] <webben> Dashiva: that behavior appears to have changed back to announcing by default in current JAWS 10 Beta.
# [21:07] <Lachy> Hixie, fair enough. But honestly, did you clearly understand what the purpose of my proposed study, and why it would be useful?
# [21:08] <Lachy> I just want to make sure its not me who's missing something
# [21:08] <Dashiva> webben: Wasn't it so that most JAWS users rarely, if ever, upgraded? So if it's gone in the latest version, it's not a big deal
# [21:09] <webben> Dashiva: "most". I'm not sure that's true. Certainly the lag time for upgrades is greater for commercial AT at hundreds of $ a shot than for free web browsers.
# [21:10] <Hixie> Lachy: though next time i would phrase your hypothesis to sound like you agree with the side that isn't yours
# [21:11] <Hixie> Lachy: since the hypothesis doesn't matter (we get the same results either way) and the people who are clearly unable to understand how to do research would at least not get a knee jerk reaction against you
# [21:11] <Philip`> Now all we need is infinite resources so we can carry out reliable user studies without worrying about their cost
# [21:12] <Dashiva> Hixie: They'd probably knee-jerk just from the from address
# [21:12] <webben> Dashiva: No. It's the other way round, I think. Default announcement has come _back_.
# [21:12] <Philip`> Hixie: But then they'll say the hypothesis is self-evident and there's no point doing new studies on it
# [21:12] <Dashiva> So it was added, removed and then re-added? Quirky
# [21:13] <Lachy> Hixie, ok. I'll keep that in mind. I'm rewriting it as clearly as I possibly can so that there can be absolutely no confusion about what I mean
# [21:13] <webben> Lachy: FWIW I agree that your proposed study would be a very good addition to information collected so far, though I don't think studies of this (human-tech interactions) sort tend to produce "irrefutable" results.
# [21:14] <Hixie> studies never provide irrefutable results
# [21:14] <Lachy> webben, my point was that the results would take us from having only observation and hypothesis closer to having a theory
# [21:14] <Hixie> they can always be refuted by more studies that contradict them
# [21:14] <webben> Hixie: Absolutely. It's not my word.
# [21:15] <Philip`> Hixie: They should object to resources being spent on that study instead of on something more interesting and informative and insightful
# [21:22] <Hixie> hsivonen: i take it your validator instance wouldn't be able to handle the load of a google blog pointing at it?
# [21:22] <hallvord> hsivonen: AFAIK Element.prototype and friends is coming in the very latest beta of IE8
# [21:22] <hallvord> ..well, should have used past tense: came .. thing is I haven't tested it yet :)
# [21:23] <webben> Lachy: e.g. this might uncover whether (for example) people don't use longdesc because they don't expect it to work because of bad UA implementation (as with the guy in Joshue's video) or whether (for example) they miss the description link because of their navigation strategy.
# [21:23] <webben> Hixie: Couldn't Google host an instance of it (if that were okay with hsivonen)?
# [21:24] <hsivonen> it would be interesting to see how badly it would melt down, though :-)
# [21:25] <Hixie> webben: hsivonen might be able to port it to appengine, i don't know
# [21:25] <hsivonen> Hixie: if you had appengine for Java...
# [21:25] <hsivonen> which I suspect Google already has internally for deploying its own Java-based apps
# [21:26] <Hixie> i don't think we've deployed any java-based apps on appengine
# [21:26] <Hixie> i don't know what the plan is for that team, either, so i don't even know if java is on the cards :-)
# [21:26] <hsivonen> Validator.nu is well-suited for scaling on an infrastructure that replicates the app automatically
# [21:27] <hsivonen> after all, it doesn't need write storage except for logs
# [21:27] <hsivonen> and it doesn't need a login mechanism
# [21:28] <hsivonen> so it could be put on a cloud computing platform without the usual BigTable or Amazon SimpleDB lock-in issues
# [21:28] <hsivonen> it does have the issue that it wants to do *a lot* of initialization up front to build read-only data structures that are shared among threads
# [21:29] <hsivonen> I figured that Validator.nu on a usual day requires less CPU and RAM than one EC2 compute unit provides
# [21:30] <hsivonen> so currently, it doesn't make sense to invest in automatic scaling on EC2
# [21:32] <hsivonen> I'm trying to come up with a way to route setAttributeNS calls to something else in IE without imposing a per-call cost on other browsers
# [21:33] <hsivonen> Hixie: I've been speculating about what's next for App Engine
# [21:34] <hsivonen> Hixie: and since Google has a small set of supported languages, Java is the obvious runner up after Python
# [21:34] <Hixie> that does seem like a good guess
# [21:34] <Hixie> (i really don't know what that team is doing)
# [21:34] <hsivonen> Hixie: but I figured providing App Engine for Java would need new restrictions on the VM
# [21:34] <hsivonen> and probably a class loading model that isn't transparently compatible with a normal JVM
# [21:35] <smedero> mostly unrelated but there's a 20% project going to get Perl into App Engine... so there's a chance that Java could be a 20% as well.
# [21:35] <hsivonen> but I wouldn't be too surprised if Google hacked up yet another environment that uses the Java language without the full JVM runtime semantics
# [21:41] <Hixie> "comments are HTMLCommentElement interfaces and inherit from Element" lol
# [21:42] <Hixie> "Early on, we had to decide if we wanted to implement a "fake" layer for the constructor/prototype objects that looked identical to the W3C model, or to expose what we had, warts and all."
# [21:42] <Hixie> ...and they picked the warts... because of their commitment to standards?
# [21:42] * Quits: mpt (n=mpt@canonical/launchpad/mpt) (Remote closed the connection)
# [21:43] <hsivonen> "Object doesn't support this property or method"
# [23:03] <webben> Lachy: I think the distinction between a user agent requiring users to change their configuration to be aware of long descriptions (e.g. earlier versions of JAWS) and announcing them by default (e.g. current JAWS 10 Beta on reading, iCab on hover) is significant.
# [23:04] <webben> although I realize John only says "natively support" in the quote.
# [23:05] <Lachy> webben, I wasn't aware of that. What should I say about it?
# [23:07] <webben> Lachy: I guess a prelude to this study would need to be some sort of agreement about what a decent implementation of longdesc might do and whether any current implementations do it.
# [23:07] <webben> If none do, you'd need to introduce an test implementation that does (e.g. by hacking NVDA perhaps).
# [23:07] <Lachy> I was just going to have the test run with each user's normal software and setup, to make it as realistic as possible
# [23:11] <webben> Lachy: I agree that it introduces other complications. But I don't think a test of bad implementations would tell one much about the actual point at issue (at least from my understanding of the point at issue).
# [23:12] <webben> Lachy: So I guess I'm saying you _might_ have to take software users are familiar with and modify it somewhat.
# [23:12] <webben> I'm not saying: make a brand new screen reader or browser or something ;)
# [23:14] <Lachy> I don't like that idea, because that would require informing the user of what changes have been made, which would bias the results due to having additional information about the test that they would not otherwise have
# [23:15] <webben> Lachy: Hmm. Couldn't you tell them that you've modified it but not tell them how?
# [23:15] <Lachy> in what way would you want to modify it?
# [23:15] <webben> for that matter, actually, what's wrong with them knowing about the feature?
# [23:16] <webben> otherwise you're not just testing whether the feature can do a given task, but also the initial intuitiveness of the feature's presentation, which could bias the study in both directions
# [23:17] <Lachy> because if the users didn't know about the longdesc before hand, and are supposedly representative of the general population, then informing them about the feature would influence the result
# [23:17] <webben> e.g. people might think, hmm, what's this long description thing, must try that. or people might think, hmm, dunno what that is, better not try that.
# [23:18] <webben> yes, but you're not trying to test people's knowledge of features, as far as I can tell.
# [23:18] <Lachy> I'm trying to test how useful a feature is in practice given the current, near the top of the range ATs
# [23:19] <webben> okay, but that's not the actual point being debated.
# [23:20] <webben> Lachy: again, that's basically testing implementations not the concept.
# [23:22] <Lachy> hmm. ok. So then we would have to set a baseline for the ATs that can be used.
# [23:22] <webben> Lachy: yeah, exactly, it very much depends on working out what a "good" implementation would do.
# [23:23] <webben> which isn't necessarily trivial.
# [23:23] <Lachy> fair enough. I suppose that's fair given how we would test the proposed headers algorithm
# [23:23] <Lachy> which would need modified UAs/ATs to do it
# [23:23] <webben> that is to say, it's easy to distinguish very bad from better implementations, but hard to distinguish very good from better implementations.
# [23:24] <webben> e.g. Firefox native longdesc implementation is pants. the longdesc firefox add-on clearly improves it.
# [23:24] <webben> iCab is an improvement again (because of the on-hover indicator).
# [23:25] <webben> I'm pretty sure that defaulting to announcing longdesc is "better" than defaulting to not announcing it.
# [23:27] <Lachy> I added a paragraph saying "We need to set a set of baseline feature requirements for the assistive technologies that can be used. I'll leave this for those who are more familiar with them,"
# [23:44] <annevk> 'Jonas Jacobi, Kaazing’s CEO, will be delivering a three day course through Kaazing’s partnership with SkillsMatter in London on October 6th, 2008. The course is entitled “Comet Evolved: HTML 5 Web Sockts & Server-Sent Events.”' -- http://thepeninsulasedge.com/blog/?p=156
# [23:44] <annevk> for features with no browser implementations, that's pretty impressive
# [23:49] * Quits: zcorpan (n=zcorpan@c-cb21e353.1451-1-64736c12.cust.bredbandsbolaget.se) (Remote closed the connection)