| 00:13 | <Hixie> | Philip`: so i dunno what happened but i now have 3346 e-mails in the database |
| 00:13 | <Hixie> | i'm thinking maybe the script died when it last ran |
| 00:14 | <Hixie> | do you want a dump again? |
| 00:14 | <Hixie> | (if so, what's your e-mail again?) |
| 00:16 | <Hixie> | in other news, who's in charge of the wiki again? |
| 00:16 | <Hixie> | i think we might want to do something about the spam, as suggested on http://wiki.whatwg.org/wiki/User_talk:Hixie, though i don't know what |
| 00:17 | <Philip`> | Hixie: When I download all the folder lists from you, that only gives 1515 messages |
| 00:17 | <Hixie> | it should have changed recently |
| 00:17 | <Hixie> | i reran the algorithm |
| 00:17 | <Hixie> | the script, rather |
| 00:17 | <Philip`> | Ah, okay |
| 00:17 | <Philip`> | The dump would be helpful so I don't have to download the rest, then |
| 00:18 | <Philip`> | (philip⊙zdcu) |
| 00:18 | <Hixie> | sent |
| 00:20 | <Philip`> | Received |
| 00:20 | <Philip`> | (Thanks!) |
| 00:20 | Philip` | tries to remember where he put his scripts |
| 00:24 | <Philip`> | Seems to be working |
| 00:26 | <Philip`> | The list of all emails is 7MB now |
| 00:26 | <Philip`> | and I see 3298 messages in it |
| 00:26 | <Philip`> | which is close enough to 3346 |
| 00:26 | <Philip`> | (I guess some messages are not in any folder?) |
| 00:31 | <Hixie> | you should have exactly 3346 in that csv dump |
| 00:31 | <Hixie> | the folders probably changed since yesterday |
| 00:31 | <Hixie> | if that matters |
| 00:33 | <Philip`> | I get 3346 from the CSV, but downloading the 'folders' list then each 'folder ...' list gives (as of ten minutes ago) 3298 |
| 00:33 | <Hixie> | odd |
| 00:34 | <Hixie> | i'd have expected the opposite |
| 00:34 | <Hixie> | that is, some of the e-mails are in more than one folder |
| 00:40 | <Philip`> | The CSV contains some message IDs like "<3019.217.124.88.212.1149790503.squirrel⊙wc" |
| 00:40 | <Hixie> | i bet the IDs are truncated to 64 bytes |
| 00:41 | <Hixie> | but that's 56 |
| 00:41 | <Hixie> | not 64 |
| 00:41 | <Hixie> | so nevermind |
| 00:41 | <Hixie> | i'm confused |
| 00:41 | <Philip`> | Of the messages that are in the CSV but not in the folders list, almost all are truncated like that |
| 00:42 | <Hixie> | weird |
| 00:42 | <Philip`> | except for <<<<f0a054a12911d5a153482846f135f669⊙1l>>>> |
| 00:42 | <Philip`> | Oh, I think I see that one in the online bit too |
| 00:43 | <Philip`> | *online folders list |
| 00:43 | <Philip`> | so ignore that |
| 00:43 | <Hixie> | it's in the database |
| 00:43 | <Hixie> | though why, i don't know |
| 00:44 | <Philip`> | I see 54 truncated names in the CSV, which is 3346 - 3298 |
| 00:44 | <Philip`> | Uh, not it isn't |
| 00:44 | <Philip`> | *no |
| 00:45 | <Hixie> | do you have a preferred username? |
| 00:45 | <Philip`> | I see 54 truncated names, plus 6 messages in two folders, and 3298 = 3346 - 54 + 6 |
| 00:45 | <Philip`> | Depends on the context :-) |
| 00:46 | <Philip`> | (Usually something like "philip" works alright) |
| 00:46 | <Philip`> | (except when someone else has stolen that name, and I have to add random punctuation characters) |
| 00:47 | <Philip`> | ((Actually, 5 messages in two folders and 1 message in three folders)) |
| 00:53 | <Hixie> | ok so how did these get truncated |
| 00:53 | <Hixie> | wtf |
| 00:54 | <Hixie> | btw the way i update the tables each night is to lock them, drop them, then recreate them entirely |
| 00:54 | <Hixie> | which is why the number can change dramatically (if the script dies while it's updating) |
| 00:54 | <Philip`> | When you measured 56 bytes, did you happen to do it with something like perl -e'print length("<3019...⊙w.")' ? |
| 00:55 | <Hixie> | yes, did i screw that up? :-) |
| 00:56 | <Philip`> | Try the -w flag :-) |
| 00:56 | <Hixie> | d'oh |
| 00:56 | <Hixie> | i suck |
| 00:56 | Philip` | made precisely the same mistake |
| 00:56 | <Hixie> | ok so i need to make it longer |
| 00:56 | <Hixie> | that means shrinking the other index column |
| 00:58 | <Philip`> | Why does it affect indexes? |
| 00:58 | <Hixie> | cos i use this as port of one of my primary indexes |
| 00:58 | <Hixie> | in the 'folders' table iirc |
| 01:01 | Philip` | doesn't see why expanding one would require shrinking another |
| 01:05 | <Hixie> | because i'm at the limit of the index key size |
| 01:06 | <Philip`> | The index can just be on a short prefix of the column and it should work exactly the same, I thought |
| 01:06 | <Philip`> | But maybe that doesn't work so well when it's meant to be a unique index |
| 01:13 | <Hixie> | ok regenning the db |
| 01:14 | <Hixie> | the reason they were cropped in the dump and not on the site is that the site was using the folders table instead of the emails table and the two tables had different lengths for the ids! |
| 01:14 | <Hixie> | <- moron |
| 01:17 | <Philip`> | Okay - once it's done, I can just download the 54 extra emails through the API |
| 01:18 | <Hixie> | done |
| 01:18 | <Hixie> | (you can also dump it straight from the database) |
| 01:23 | <Philip`> | Hmm, it looks like I got the same folders list as before |
| 01:24 | <Philip`> | (unless I messed up somewhere) |
| 01:24 | <Philip`> | with only the 3298 messages |
| 01:29 | <Hixie> | there's something really weird going on |
| 01:32 | <Hixie> | there are 3,353 entries in the 'emails' table |
| 01:32 | <Hixie> | there are 3,359 in the 'folders' table (which includes dupes) |
| 01:33 | <Philip`> | There are 3,353 distinct 'message' values in the 'folders' table, and they all correspond to an 'id' in 'emails' |
| 01:33 | <Hixie> | right |
| 01:33 | <Philip`> | so that all seems to be correct |
| 01:33 | <Hixie> | (i just checked) |
| 01:33 | <Hixie> | so why do you only get 3298 if you use the api |
| 01:33 | <Philip`> | Is there any caching somewhere? |
| 01:33 | <Hixie> | oh, maybe, yeah |
| 01:33 | <Philip`> | (I deleted my own cache before downloading the folders list again) |
| 01:34 | <Hixie> | i just killed my server, try again |
| 01:35 | Philip` | tries |
| 01:38 | <Philip`> | 3359 |
| 01:38 | <Philip`> | Perfect :-) |
| 01:38 | <Hixie> | phew |
| 01:38 | <Hixie> | good call on the caching |
| 01:38 | <Hixie> | i've added a scary line to my regen script that kills the server |
| 01:40 | <Philip`> | Is there not a less drastic way to clear the cache? |
| 01:41 | Philip` | 's usual approach to caching is "nobody uses my stuff anyway, so it doesn't matter if it's slow and resource-intensive" |
| 01:41 | <Hixie> | Philip`: that was my approach, and then you started fetching all this data... ;-) |
| 01:41 | <Hixie> | Philip`: there's probably a better way, but *shrug* |
| 01:42 | <Hixie> | the server knows how to handle it and the CGI scripts that talk to it know how to handle the server script going away |
| 01:42 | <Hixie> | (the server i'm talking about isn't the database, it's a cgi script that talks a little tcp protocol to my other cgi scrits) |
| 01:42 | <Hixie> | scripts |
| 01:43 | <Philip`> | Hmm, good point - I'll make sure I don't offer any APIs and tempt people to use them :-) |
| 01:44 | <Philip`> | (Ah, okay, that's less drastic than killing MySQL and Apache and whatever) |
| 01:44 | <Hixie> | yeah |
| 01:44 | <Hixie> | i really wish we had TCPConnection |
| 01:44 | <Hixie> | it would be so much better than having this mess |
| 01:44 | <Hixie> | and then i could do away with everything except the server script |
| 01:50 | <Hixie> | hey anyone recall the uris of the studies similar to the webstats one i did? |
| 01:50 | <Hixie> | there was one that had a flower on the front page iirc |
| 01:51 | <Philip`> | http://triin.net/2006/06/12/Coding_practices_of_web_pages ? |
| 01:51 | <Hixie> | yes! |
| 01:51 | <Hixie> | thanks! |
| 02:01 | <Philip`> | http://triin.net/archive/kool/webstat/figure-33.png - you can tell the data set had a very high proportion of CNN pages |
| 02:02 | <Hixie> | you should see the number of <nyt_copyright> elements in the studies i do :-P |
| 02:02 | <Hixie> | not enough to come in the top 100 or anything, but still significant |
| 02:02 | <Philip`> | cnn.com was 5% of the URLs on dmoz.org when I last counted |
| 02:03 | <Hixie> | yeah |
| 02:03 | <Hixie> | cnn.com has a lot of pages |
| 02:03 | <Philip`> | I've no idea who keeps submitting them all to dmoz.org |
| 02:03 | <Hixie> | cnn, maybe |
| 02:03 | <Philip`> | Oh, that would make sense |
| 02:04 | <Hixie> | comparing http://triin.net/archive/kool/webstat/figure-12.png to http://code.google.com/webstats/2005-12/charts/unique-elements-per-page.svg is interesting |
| 02:05 | <Hixie> | it's clearly not a perfect normal distribution |
| 02:05 | <Hixie> | there's a very clear bump around 5-10 in both graphs |
| 02:05 | <Hixie> | actually i guess i'd be approximately a poisson, not a normal, distribution. |
| 02:07 | <Philip`> | It seems everyone has <html>, <head>, <title>, <body>, and then I guess you don't need more than a few of the next ones (<meta>, <a>, <img>, <br>, <table>, etc) to make up the rest of your page |
| 02:07 | <Lachy> | Hixie, I'm unable to login to the wiki server using SFTP. It responds with unable to authenticate. |
| 02:07 | <Hixie> | oh it was probably one of the accounts that dreamhost changed the passwd on |
| 02:07 | <Lachy> | same with the blog |
| 02:07 | <Hixie> | let me fix it |
| 02:08 | <Lachy> | whatwikiuser and lhunt were the usernames |
| 02:08 | <Philip`> | The spike at 6 on triin.net looks odd, since I can't imagine people doing much with just two extra tags |
| 02:08 | <Hixie> | Philip`: the spike is a bit later on the google one but yeah |
| 02:08 | <Hixie> | there's a prike |
| 02:08 | <Hixie> | spike |
| 02:09 | Philip` | wonders if he can get that graph from his own data |
| 02:09 | <Hixie> | what's the uri to yours again? |
| 02:09 | <Philip`> | http://canvex.lazyilluminati.com/survey/2007-07-17/analyse.cgi/index |
| 02:14 | <Philip`> | http://canvex.lazyilluminati.com/survey/2007-07-17/tagcount.html |
| 02:15 | Lachy | installs the captcha extension |
| 02:16 | <Lachy> | can someone with an ordinary user account on the wiki attempt to make an edit to see if it works? (sysops won't see it) |
| 02:16 | Philip` | still likes the "[x] Are you a spambot?" approach to blocking spam |
| 02:16 | <Lachy> | wiki.whatwg.org |
| 02:16 | <Philip`> | (Doesn't work on human spammers, but computers never answer the question right) |
| 02:16 | <Hixie> | Philip`: see! see! your data has the same thing! |
| 02:17 | <Philip`> | My data comes from exactly the same source as the triin.net stuff, so that's not surprising :-) |
| 02:17 | <Philip`> | with the same 5% cnn.com, and quite a lot of weather.com |
| 02:17 | <Hixie> | heh |
| 02:17 | <Hixie> | pity :-P |
| 02:17 | <Philip`> | (except a much smaller sample) |
| 02:18 | <Philip`> | Lachy: It let me save a change without asking any hard questions |
| 02:19 | <Lachy> | oops, It hadn't uploaded LocalSettings.php |
| 02:19 | <Lachy> | try again |
| 02:19 | <Lachy> | aargh! fatal error. |
| 02:21 | <Hixie> | i assume triin.net is Pene Saarsoo's site |
| 02:21 | <Hixie> | Rene |
| 02:21 | Philip` | looks at the spikey bits in more detail |
| 02:22 | <Lachy> | Philip`, fixed the error, can you try again? |
| 02:26 | <Philip`> | Lachy: Is it meant to show on the editing page, or only after saving? |
| 02:26 | <Lachy> | I believe it's supposed to be after attempting to save |
| 02:27 | <Lachy> | these are the instructions http://www.mediawiki.org/wiki/Extension:ConfirmEdit |
| 02:27 | <Lachy> | it doesn't tell me much |
| 02:27 | <Philip`> | It let me save again without anything |
| 02:28 | <Philip`> | Hixie: Of 114 the pages with 6 tag names, 61 have <frameset> and <frame> |
| 02:28 | <Philip`> | (and most have <meta>, <title>, <html>, <head>) |
| 02:28 | <Philip`> | which explains the spike at 6 |
| 02:28 | <Philip`> | (since they're all just frameset pages) |
| 02:29 | <Hixie> | oooh |
| 02:29 | <Hixie> | interesting |
| 02:29 | <Philip`> | Of the 208 pages with 8, 137 have <noframes>, 149 have <frameset> and <frame>, and most have meta/body/title/head/html |
| 02:30 | <Philip`> | so it seems they're all just frameset pages again |
| 02:30 | <Hixie> | aah |
| 02:30 | <Hixie> | interesting stuff! |
| 02:30 | <Philip`> | Same for 9 tags (244 pages, 168 have frames, 159 have noframes, 141 have p) |
| 02:31 | <Lachy> | Philip`, it will if you attempt to add a URL or attempt to create an account |
| 02:31 | <Lachy> | it asks the user a simple math questions like: 44 - 5 = ? |
| 02:32 | <Philip`> | Lachy: Aha, that works |
| 02:32 | <Philip`> | "69 - 5 =" |
| 02:32 | Philip` | thinks a bit |
| 02:32 | <Philip`> | Oh, it wasn't 61 |
| 02:33 | <Philip`> | The "Create account" page says "82 + 6 =" |
| 02:33 | <Philip`> | so that seems to be working |
| 02:46 | <Philip`> | Hixie: http://canvex.lazyilluminati.com/survey/2007-07-17/tagcount.html |
| 02:46 | <Philip`> | now with details about what tags the various groups of pages use |
| 02:46 | <Hixie> | neat |
| 02:47 | <Philip`> | Hmm, the page using 41 tags looks quite legitimate |
| 02:48 | <Hixie> | heh |
| 02:48 | <Philip`> | (http://community.webshots.com/user/JulioUU) |
| 02:49 | <Hixie> | woot! you have some nyt_copyright elements too! |
| 02:49 | <Philip`> | http://canvex.lazyilluminati.com/survey/2007-07-17/analyse.cgi/tag/nyt_copyright |
| 02:49 | <Philip`> | Only one :-( |
| 02:50 | <Hixie> | i got over a million of those in my sample last year |
| 02:50 | <Hixie> | (don't recall what the recent numbers are) |
| 02:50 | <Hixie> | er, not last year. year before. |
| 02:50 | <Hixie> | whenever my first sample was. |
| 02:53 | <Philip`> | 11 pages with nothing but <meta> - looks like those are redirects |
| 02:53 | <Hixie> | yeah |
| 02:59 | <Philip`> | (It's nice having data about eight thousand pages rather than eight billion, because I can write hopelessly inefficient SQL queries and still get information back in a few seconds ;-) ) |
| 03:32 | <Lachy> | why do people keep overreacting and bringing up the headers issue all the time?! http://lists.w3.org/Archives/Public/public-html/2007Aug/0926.html |
| 04:07 | <Hixie> | "headers= are allready counted by our editor as insignificant" |
| 04:07 | <Hixie> | they are? i thought i'd not yet looked at them. |
| 04:14 | <Lachy> | well, apparently Leif knows something about you that you don't :-) |
| 04:15 | <Hixie> | good to know |
| 04:15 | <Hixie> | i better update my opinions to think that headers="" is insignificant then |
| 04:31 | <Lachy> | I'm trying to think of a way to explain that the significance of a feature isn't dependent upon stats alone and give some examples of other factors that could be considered. any suggestions? |
| 04:32 | <othermaciej> | there's a lot of things to consider |
| 04:33 | <othermaciej> | - does the feature fulfill a needed use case? |
| 04:33 | <othermaciej> | - is it possible to do the same things as well without the feature? |
| 04:33 | <othermaciej> | - which existing user agents, if any, implement the feature? |
| 04:33 | <othermaciej> | - is the feature used widely in existing content? |
| 04:34 | <Hixie> | - have existing user agents invented similar proprietary features to address the use case |
| 04:34 | <Hixie> | - have libraries (e.g. dojo) implemented work arounds for the lack of the feature? |
| 04:34 | <Lachy> | - benefits to users and authors |
| 04:34 | <othermaciej> | - is most existing use (if there is any) such that it would be beneficial or hamful to the goal of the feature to support it? |
| 04:35 | <othermaciej> | for example if a feature is not implemented by mainstream browsers but is widely used, it becomes likely that at least some content using it will depend on it being ignored |
| 04:36 | <othermaciej> | there's all sorts of things to think about |
| 04:36 | <Lachy> | thanks, that's a good list |
| 04:37 | <othermaciej> | and another thing to consider is that different criteria may apply to whether a feature is required for implementations, and to whether it is valid for content |
| 04:37 | <othermaciej> | a feature that is considered to have no valid use cases and to violate various important principles may nontheless be required for implementations |
| 06:10 | <Hixie> | so i'm examining some of my data collected over teh past few months with an eye towards what can be published |
| 06:10 | <Hixie> | looking at the longdesc="" data closer, it seems the situation is even more dire than i thought |
| 06:11 | <Hixie> | only about 0.6% of <img> elements with a longdesc="" attribute have a useful value, as far as i can tell |
| 06:11 | <Hixie> | which make it about 0.0007% of <img> elements |
| 06:11 | <othermaciej> | what are some of the kinds of values that are non-useful? |
| 06:12 | <Hixie> | all the wikipedia <img> elements have bogus values that point to non-descriptions |
| 06:12 | <Hixie> | there are also many <img> elements that have longdesc="" attributes with values that are redundant with an ancestor <a> element's href="" attribute |
| 06:12 | <Hixie> | or that point to the root of another domain |
| 06:13 | <othermaciej> | wow |
| 06:14 | <Hixie> | hm, i didn't check for identity with the alt="" attribute |
| 06:14 | <Hixie> | i should have |
| 06:16 | <Lachy> | Hixie, there's a bug in your issues list http://www.whatwg.org/issues/top - see the top issue that says 3 votes, yet there's only 1 |
| 06:17 | <Hixie> | woah, clicking it changes the result |
| 06:17 | <Hixie> | freaky |
| 06:17 | <Hixie> | noted |
| 06:17 | <Hixie> | thanks |
| 06:18 | <Lachy> | what evidence was presented in support of longdesc? |
| 06:18 | <Hixie> | dunno, haven't studied it beyond looking at this data so far |
| 06:19 | Lachy | checks the wiki |
| 06:19 | <Lachy> | hmm. some people want longdesc on iframe too |
| 06:20 | <othermaciej> | that sounds amazingly redundant |
| 06:20 | <othermaciej> | the iframe itself would in general contain markup |
| 06:20 | <othermaciej> | why would you need a markup alternative to it? |
| 06:21 | <Hixie> | where else can spammers put their links? we've blocked off most of the other places |
| 06:25 | <Lachy> | Hixie, can you publish a list of URIs for sites that look like they're using longdesc legitimately? |
| 06:27 | <Hixie> | http://junkyard.damowmow.com/292 |
| 06:27 | <Lachy> | is that all of them, or just a sample? |
| 06:27 | <Hixie> | sample of 100 |
| 06:28 | <Hixie> | there were millions that didn't hit any of my heuristics |
| 06:28 | <Hixie> | (about 6% of <img> elements with a longdesc="") |
| 06:28 | <Lachy> | This is not legitmate: <a href="http://www.google.co.jp/"><img src="http://blog2.fc2.com/2/20century/file/Logo_20s.gif"; alt="Google" height="75" width="143" longdesc="http://www.google.co.jp/logos.html"; /></a> |
| 06:28 | <Hixie> | indeed not |
| 06:29 | <Hixie> | there are many that are not |
| 06:29 | <Hixie> | about 90% by the sample i looked at |
| 06:29 | <Lachy> | I'll let you know if I find one that is |
| 06:32 | <Lachy> | it's interesting how the entire list of use cases given in the wiki, doesn't actually include any use cases. Just a list of different disabilities that might benefit from it |
| 06:39 | <Hixie> | k well the score doesn't change when you open one anymore |
| 06:39 | <Hixie> | but i don't see why the score of the top one is wrong |
| 06:40 | <Lachy> | maybe it thinks I'm so important that my vote is worth 2? :-) |
| 06:41 | <Hixie> | seems unlikely! ;-) |
| 06:47 | <Hixie> | either i don't understand GROUP BY / COUNT() or there's a bug in mysql |
| 06:56 | <Hixie> | oh! |
| 06:56 | <Hixie> | the message is in two folders! |
| 07:01 | <Hixie> | ok fixed |
| 07:01 | Hixie | changed COUNT(...) to COUNT(DISTINCT ...) |
| 07:58 | <Lachy> | Hixie, one attempted legitimate use of longdesc, but it's still questionable http://www.tcfp.state.tx.us/standards/standards_manual/standards_manual.asp?chapter=431429421 (see the state of texas logo at the bottom) |
| 07:58 | <Lachy> | links to http://www.tcfp.state.tx.us/image_description.txt |
| 08:20 | <jruderman> | .txt! |
| 08:21 | <othermaciej> | that could totally be alt text |
| 08:21 | <othermaciej> | the best part is that the plaintext contains a URL |
| 08:34 | <Lachy> | I updated the wiki to add some *real* use cases to replace the list of purported beneficiaries that was there http://esw.w3.org/topic/HTML/LongdescRetention |
| 09:06 | <Dashiva> | Lachy: Sorry if I double-posted you yesterday, had some trouble finding the right account for w3c posting |
| 09:06 | Dashiva | dreams about a mailing list with useful reply-to |
| 09:06 | <Lachy> | double posted me? which message? |
| 09:07 | <Lachy> | which name do you use in emails? |
| 09:07 | <Dashiva> | magnusrk+something@ |
| 09:07 | <Lachy> | BTW, munging reply-to headers is considered bad practice and I'm glad the W3C doesn't do it |
| 09:09 | <Dashiva> | I've yet to see an argument against it that wasn't based on happening 20 years ago |
| 09:09 | <Lachy> | ah, right. I did get multiple copies of that one, but it doesn't bother me since I just deleted it along with the other dupes |
| 09:10 | <Lachy> | I've yet to see a convincing argument for it that isn't based on a users inability to use their mail client properly |
| 09:10 | <takkaria> | given that HTML5 is based in part around users' inability to write valid markup, it seems like that might be a pretty good reason. :) |
| 09:11 | <Lachy> | just press Reply All (or equivalent). It's usually right next to the Reply button! |
| 09:11 | <Dashiva> | That's the problem |
| 09:11 | <takkaria> | Lachy: oh, I agree, I just enjoy the mild irony. :) |
| 09:11 | <Dashiva> | I got a mail from Robert Burns today that didn't contain a single word I've said because i've been passed along by reply all for who knows how long |
| 09:12 | <Lachy> | people need to learn to trim the recipient list too! |
| 09:12 | <Dashiva> | User education doesn't work, you should know that ;) |
| 09:12 | <takkaria> | Lachy: btw, rob moved your use cases to "authoring cases" and moved what were "use cases" before back there |
| 09:13 | <Lachy> | WTF??? |
| 09:13 | <Lachy> | They're not use cases! |
| 09:14 | <takkaria> | I tried to clean up the <img>fallback</img> section a while ago to remove duplicates, and he just reverted it, so I gave up on it |
| 09:16 | <zcorpan> | amusing |
| 09:17 | Lachy | reverted rob's change |
| 09:17 | <Dashiva> | revert war, I choose you |
| 09:18 | zcorpan | escapes the expected bashing |
| 09:21 | takkaria | wonders if the Design Principles aren't better changed to "How the HTML WG should work" |
| 09:25 | <takkaria> | Lachy: btw, are you going to look through all of Hixie's longdesc urls? |
| 09:25 | <Lachy> | not all of them. I looked through about a dozen randomly selected |
| 09:27 | <zcorpan> | random selection, eh? in the statistical sense? not biased in some way? ;) |
| 09:28 | <Lachy> | well, not entirely random. I just picked a couple from the top, scrolled down a bit and picked a few others, then repeated. |
| 09:28 | <Lachy> | it was sort of biased by ignoring the wikipedia entires that I knew were bogus |
| 09:29 | <takkaria> | I assume someone's already argued with the wikimedia people about how they're misuing longdesc? |
| 09:30 | <Lachy> | don't know, probably. |
| 09:32 | takkaria | adds that to his todolist |
| 09:43 | <Lachy> | Rob isn't listening again, I give up |
| 09:45 | <takkaria> | Lachy: apologies; I've sent you two duplicate replies to your longdesc post, both off-list |
| 09:45 | <takkaria> | I confuse myself by having too many email accounts and never hitting reply-to-all |
| 09:47 | <Lachy> | takkaria, just resend it to the list, it doesn't matter if I get dupes |
| 09:47 | takkaria | has done so |
| 10:29 | <zcorpan> | um |
| 10:29 | takkaria | sighs at Rob for another contentless post |
| 10:29 | <zcorpan> | what should happen if you serve a gif as image/png and use it in an <object>? |
| 10:30 | <zcorpan> | browsers just decode it as a gif |
| 10:52 | <Lachy> | zcorpan, browsers don't care what image MIME is use, they just invoke the appropriate image library by sniffing the first few bytes of the file, looking for the signature |
| 10:56 | <Whiskey_M> | 'lo |
| 11:29 | <zcorpan> | Lachy: right... so image/png, image/gif, etc, all go through "Content-Type sniffing: image" |
| 12:03 | <Lachy> | zcorpan, yeah, something like that |
| 13:42 | <zcorpan> | hmm, wonder if browsers look at the file extension in their sniffing code |
| 13:44 | <Whiskey_M> | if it helps I have noted that for sites that serve content (normally avi's ) with incorrect headers IE will normally serve it with the application based upon extension, rather than FF which tends to serve based on the header |
| 13:46 | <zcorpan> | interesting |
| 13:48 | <Philip`> | Is that for content that the browser displays itself, or for content where it either saves to disk or asks the OS to find some external program to open it? |
| 16:26 | <Lachy> | http://blog.whatwg.org/omit-alt |
| 16:29 | <Lachy> | jgraham, that's a good explanation of the flag example you posted. Thanks, now I don't have to respond. :-) |
| 18:56 | <gsnedders> | who actually owns html5.org? |
| 18:57 | <Lachy> | gsnedders, anne |
| 18:57 | <gsnedders> | ah |
| 22:18 | <Hixie> | btw Lachy if you can suggest some page-only heuristics (i.e. not involving the network) for detecting bogus longdesc=""s that would have caught them in the URLs i mentioned, it would be useful |
| 22:18 | <Hixie> | i'd be able to rerun the study excluding those |
| 22:33 | <othermaciej> | I looked at a few of those and they generally didn't look detectable from the page context alone |
| 22:36 | <Hixie> | yeah that was my conclusion too |
| 22:37 | <Hixie> | so i haven't yet looked at the headers="" stuff in detail |
| 22:37 | <Hixie> | but i wonder |
| 22:38 | <Hixie> | if i look at it, and find that imho we shouldn't include headers="" (which could be the case, just like it could be the case that we _should_ include it) |
| 22:38 | <Hixie> | will the situation become worse or better? |
| 22:38 | <Hixie> | in other words, is it better for the working group for me to address the issue at its proper time, or should i prioritise it given that the risk of that is that the attribute does in fact not get kept? |
| 22:42 | <othermaciej> | whether it has a positive effect on attitudes depends on the outcome |
| 22:42 | <jgraham> | Hixie: The problem is that we are unlikely to ever reach consensus on a spec that does not include headers irrespective of its technical merit |
| 22:42 | <othermaciej> | doing enough work to predict the outcome but not actually make any changes would not really be acting in good faith though, I think |
| 22:43 | <Hixie> | jgraham: we'll never reach consensus on the spec anyway |
| 22:43 | <jgraham> | Hixie: Fair point. |
| 22:43 | <Hixie> | othermaciej: yeah i wouldn't do that |
| 22:43 | <othermaciej> | I think there is a sense from some quarters that any feature that is nominally "for accessibility" is absolutely needed, regardless of whether it does anything to help accessibility in practice |
| 22:43 | <Hixie> | right |
| 22:44 | <othermaciej> | and that even trying to study the question of whether it helps accessibility in practice is somehow illegitimate |
| 22:45 | <othermaciej> | notwithstanding what I think of specific features, it is hard to make a good spec when there are sacred cows that may not be examined critically, so I don't know how much we can accomodate that point of view |
| 22:45 | <Hixie> | i'm not considering any cows sacred |
| 22:45 | <billmason> | I would suggest that if headers got some kind of priority and a determination made, even if the determination was "keep it", there would just be a move to argue the next accessibilty issue in dispute immediately. I would not make it a special priority just to give those evangelizing for it a more immediate answer. |
| 22:46 | <Hixie> | i'm not really concerned about appeasing people, i want to make the spec the best html spec possible, including addressing the goal of making documents that use the language universally accessible. |
| 22:46 | <othermaciej> | I know you don't - I'm just saying it's hard to collaborate with someone who does have sacred cows, even if you happen to agree with them on one point |
| 22:46 | <jgraham> | But it makes everyone's life miserable when some people consider cows to be sacred and they are taken to slaughter (so to speak) |
| 22:46 | <Hixie> | othermaciej: yeah |
| 22:47 | <Hixie> | billmason: yeah. so that argues for just treating it like any other issue, and going with the FIFO principle, right? |
| 22:47 | <billmason> | That would be my take on it, yes. |
| 22:47 | <Hixie> | i think that's probably the best course of action |
| 22:47 | <billmason> | And for parsing logs later to write emails, I say that as a person with a general accessibility interest/focus/etc. |
| 22:47 | <Hixie> | i expect Dan to start telling me to prioritise public-html wg feedback over existing whatwg feedback though. |
| 22:48 | <Hixie> | he seems to have interpreted my "a few months" as "3 or so months" not "20 or so months" which is probably closer to realistic. |
| 22:48 | <othermaciej> | I do think Chaals made a good point that accessibility features might deserve special status even if only a subset of especially accessible sites use them (properly), since those might be the only sites that someone with a given disability can use at all |
| 22:48 | <Hixie> | not sure what to do about that if he does ask me to do so |
| 22:48 | <othermaciej> | but I don't know if the evidence bears that out as a justification for headers="" or the like |
| 22:49 | <jgraham> | Hixie: Two stacks, one issue from the top of each in turn? |
| 22:49 | <othermaciej> | it would require examples of actual particularly accessible sites that consistently use it in a way that is beneficial |
| 22:49 | <othermaciej> | not just a theory that there might be some |
| 22:49 | <Hixie> | othermaciej: well, there are two things, right, there's the features that are rarely used but when used make the page better, and there are the features that are very widely misused to the point where they actively hurt the user's experience on most sites |
| 22:50 | <Hixie> | longdesc clearly falls into the latter category according to all the studies i've done or seen |
| 22:50 | <Hixie> | as in, exposing longdesc will mostly expose the user to spam or useless content, even on supposedly accessibility-aware pages |
| 22:51 | zcorpan | would think that the way tabindex is implemented, it actively hurts the user experience too. treating all positive numbers as 0 for tabindex would, i think, have better user experience |
| 22:52 | <Hixie> | that's possible too, i haven't even looked at tabindex yet other than adding the negative thing |
| 22:53 | <zcorpan> | yep. :) i just suddenly came to think of tabindex when reading the above |
| 22:53 | <Hixie> | one thing i really don't know how to fix is accesskey="" |
| 22:54 | <zcorpan> | dunno either |
| 22:54 | <jgraham> | That isn't going to be fun. |
| 22:54 | <zcorpan> | although opera's implementation is somewhat useful, or at least not harming the user experience |
| 22:55 | <Hixie> | opera's implementation is unintuitive |
| 22:55 | <jgraham> | From what I remember Mike Smith saying it sounds like different solutions are appropriate on different devices |
| 22:55 | <Hixie> | and doesn't actually solve the problem of how to make it device independent |
| 22:55 | <Hixie> | yeah |
| 22:55 | <Hixie> | maybe it really is a stylistic thing |
| 22:55 | <Hixie> | we do have key-equivalent in CSS iirc |
| 22:56 | <Philip`> | Hixie: I would assume users wouldn't see the spam or useless content since they wouldn't be explicitly asking their tool to give them the longdesc, except in the cases where they have a good expectation that it's going to be useful, so misuse wouldn't hurt the user much |
| 22:56 | <Hixie> | Philip`: really? how would they know to ask? |
| 22:57 | <Hixie> | for longdesc it really seems to me that the only cases i've seen where the longdesc was actually useful and wasn't something you could have just stuffed into alt="", it was actually useful to sighted users too |
| 22:57 | <Hixie> | and could have just been included on the page or in a link from the page |
| 22:57 | <Philip`> | I'd assume the tool would indicate in some quick way that there is a longdesc attached to the image, similarly to how it must quickly indicate wherever there is a link |
| 22:58 | <Philip`> | but I have precisely no experience of how relevant tools work in practice |
| 22:58 | <zcorpan> | jaws says "press enter for long description" after reading the image alt, and if you press enter it will open the url in a new window |
| 22:58 | <zcorpan> | iirc |
| 22:58 | <Hixie> | right but how do you know it's appropriate? |
| 22:59 | <Philip`> | (At least from what I've heard about table headers, they're not read out by default - you press some key when you've got the right cell selected, and if you see one cell has rubbish headers then you won't bother checking every single other cell - so the harm caused by misuse is similarly minimised) |
| 23:01 | <Philip`> | Hixie: By considering the context, like whether other images on the page have useful longdescs - e.g. you'd know it's worthwhile reading all the longdescs in the CSS spec after you've seen the first few |
| 23:02 | <Philip`> | (and on sites which misuse it, you'd quickly realise you should just ignore all the rest) |
| 23:02 | <Hixie> | that doesn't seem like the optimal user experience |
| 23:02 | <Hixie> | it's like reminding the user continually that the page wasn't designed for them but there's this secondary set of content they can access |
| 23:03 | <Hixie> | seems like it would make one bitter |
| 23:04 | Hixie | tries to build up a list of requirements for offline web apps |
| 23:06 | <Philip`> | It seems a less significant negative point than with e.g. <input usemap> (where if your browser supports it, there are features of some sites that just don't work at all) - it might waste a bit of time to read out useless longdescs, but it's not preventing the user from using the site |
| 23:08 | <Philip`> | A closer-to-optimal user experience would be good, though I don't have any ideas for that :-) |
| 23:08 | <Hixie> | yeah the main argument for me against longdesc is that it's not useful at all, except in rare cases where frankly sighted users would benefit too, and therefore you're better off putting the content on the page itself |
| 23:08 | <othermaciej> | Hixie: one thing that might work is to limit the number of access keys and make it a set that UAs can map in a natural way on each platform to something w/ no conflicts |
| 23:08 | <Hixie> | it's hard to say since i've seen so few useful uses of it |
| 23:08 | <Hixie> | othermaciej: yeah that's been suggested |
| 23:08 | <Hixie> | othermaciej: but then the UI becomes unclear |
| 23:09 | <Hixie> | othermaciej: i.e. discoverability drops through the floor |
| 23:12 | <kingryan> | Hixie: having CSS-useable hooks could help with styleability |
| 23:12 | <Hixie> | yeah |
| 23:12 | <kingryan> | s/styleability/discoverability/ |
| 23:13 | <Hixie> | like ...:after { content: ' (' copy-key ')'; } |
| 23:13 | <Hixie> | ... { key-equivalent: copy-key; } |
| 23:13 | <kingryan> | yeah |
| 23:13 | <Hixie> | or ... { key-equivalent: copy-key; } ...::after { content: ' (' key-equivalent ')'; } |
| 23:13 | <Hixie> | ...to have resilience against the cascade |
| 23:13 | <kingryan> | ....:access-key('n'):after { content: 'meta-key N'} |
| 23:14 | <kingryan> | yeah |
| 23:14 | <Hixie> | i was thinking of also having othermaciej's idea since that solves the device problem too |
| 23:14 | <Philip`> | Subtitles on TV shows would benefit sighted users too (e.g. when they get distracted and miss a couple of words, or can't understand someone's accent), but usually they aren't displayed along with the content since they're ugly and distracting and not sufficiently useful to be shown to all users; people may have similar reasons for not wanting to put image descriptions in the normal page content, and hiding it being [D] links or londesc |
| 23:15 | <kingryan> | Hixie: is om's idea of having a limited set? |
| 23:15 | <Philip`> | s/being/behind/ |
| 23:15 | <Hixie> | yeah but subtitles aren't only accessible to blind users or hidden behind long properties pages, they're one-button accessible |
| 23:15 | <Hixie> | like an <a> link would be |
| 23:15 | <Hixie> | kingryan: yeah see above |
| 23:16 | <Philip`> | That just seems like a browser UI issue |
| 23:16 | <kingryan> | yeah, me wishes browsers couldn't remap stuff that the browser already handles |
| 23:16 | <kingryan> | or allow me to re-remap it |
| 23:16 | <Hixie> | Philip`: i don't think we should hide longdesc behind a context menu or something. i'm saying the link should be right there in the content just like for everything else. |
| 23:19 | <Philip`> | That sounds like D-links - I think the only argument I've heard against them is they don't look very nice (but there are quite possibly other arguments I haven't heard) |
| 23:19 | <othermaciej> | Hixie: if the set was 0-9 for instance, it could be unmodified 0-9 on phones with a keypad, Cmd-0 - Cmd-9 on Macs, Ctrl-0 - Ctrl-9 on windows, etc |
| 23:19 | <othermaciej> | (assuming those are actually free) |
| 23:20 | <Hixie> | Philip`: yeah, though i wouldn't use [D]. |
| 23:20 | <othermaciej> | but I'm not sure how you address discoverability |
| 23:20 | <Philip`> | (http://www.w3.org/TR/WCAG10-HTML-TECHS/#long-descriptions says "Invisible d-links thus provide a (temporary) solution for designers who wish to avoid visible d-links for stylistic reasons.") |
| 23:20 | <othermaciej> | I think the page has to provide info about shortcuts |
| 23:20 | <Hixie> | othermaciej: i think kingryan's idea helps with that |
| 23:20 | <billmason> | I think there's some kind of study that says numeric shortcuts aren't really free, though. |
| 23:20 | <othermaciej> | that's how keyboard shortcuts work in native apps |
| 23:20 | <othermaciej> | they are listed in the menu bar |
| 23:21 | <othermaciej> | having a shortcut list with labels somewhere in the page or accessible from the chrome would be the natural analogy |
| 23:21 | <othermaciej> | I'm not sure if styling them to mention they key helps |
| 23:21 | <kingryan> | yeah, having it in the chrome seems reasonable |
| 23:21 | <othermaciej> | thee UA could do a better job if it is the only thing that knows the concrete key mapping, but you need a label to go next to the shortcut |
| 23:22 | <othermaciej> | *the UA |
| 23:22 | <othermaciej> | it could even be a menu or submenu in the menu bar |
| 23:22 | <Hixie> | hm yeah, for <command>s and other Command elements you could just have the UA create a menu somewhere with the key equivs |
| 23:22 | <othermaciej> | "Page Shorcuts" |
| 23:22 | <kingryan> | it'd be nice in interactive browsers if you could could style the access-key elements with the modifier key is pressed |
| 23:22 | <Hixie> | they could do that now with :active |
| 23:23 | <othermaciej> | kingryan: if the modifier is a commonly used one on the OS, that could be distracting |
| 23:23 | <kingryan> | othermaciej: indeed it could |
| 23:23 | <othermaciej> | something based on <command> seems like a good basic approach |
| 23:23 | <Hixie> | oh i misread what kingryan said |
| 23:23 | <othermaciej> | the idea is that a shorcut key actually activates a command, for which there might also be one or more UI elements |
| 23:23 | <kingryan> | there are some places, like dialogue windows in os x, where this is done already |
| 23:24 | <othermaciej> | it makes more sense to associate it with a command, which can then have an appropriate label |
| 23:24 | <othermaciej> | kingryan: example? |
| 23:24 | <othermaciej> | (I don't know of dialog windows that have keyboard shortcuts beyond the standard tab/enter/esc and such) |
| 23:25 | <kingryan> | I remember something adding "cmd-foo" in a button when cmd is pressed down |
| 23:25 | <kingryan> | I'm not sure where that is |
| 23:25 | <othermaciej> | I don't think that is standard |
| 23:25 | <kingryan> | it's not |
| 23:25 | <kingryan> | I'm just seeing that I've seen it done before and I helped with discoverability |
| 23:28 | <othermaciej> | usually in OS X all the keyboard shorcuts also have a menu item, so you can look in the menu system to see the shortcuts |
| 23:31 | <kingryan> | othermaciej: true |
| 23:33 | <othermaciej> | Windows is different since it uses the underline system sometimes, and that may apply both to menus and items in a dialog |
![http://blog2.fc2.com/2/20century/file/Logo_20s.gif"](http://blog2.fc2.com/2/20century/file/Logo_20s.gif"){kind=link}