09:25
<annevk>
benjamingr__: are you around? Do you want your name in the Acknowledgments section?
11:29
<jgraham>
annevk: What's the COOP scenario, exactly? (my initial reaction is "you get a contextDestroyed event for the old context and a contextCreated event for the new one, even if they're the same 'tab' from the point of view of the user")
11:30
<annevk>
jgraham: that is correct; I guess I'm mainly wondering if you know they relate to the same tab
11:31
<annevk>
jgraham: HTML still needs to describe something called browsing session, which will provide some coherence for this
11:31
<annevk>
(a browsing session is a tab, I guess)
11:32
<jgraham>
Yeah, I guess that's an interesting point. For WebDriver/HTTP we have window handles that are supposed to be for the user-percieved windows
11:32
<jgraham>
And these things are supposed to be compatible
11:33
<annevk>
jgraham: and we might not discard in the future, but freeze the browsing context somehow so you can do fastback
11:35
<jgraham>
Yeah, I assume that navigation can not discard nested contexts even in the no-COOP case, and it's not super clear if users will be confused by not getting events in that case
11:39
<andreubotella>
I was under the impression that a top-level browsing context was a tab
11:40
<annevk>
andreubotella: that was the case until we changed that with COOP
11:40
<annevk>
andreubotella: now it's N:1
11:41
<andreubotella>
I see
11:49
<andreubotella>
annevk: btw, playing around with file inputs and urlencoded I found more newline normalization bugs
11:50
<annevk>
Cool, did you file bugs?
11:51
<andreubotella>
it's actually something all browsers share, even if imo it should count as a bug in the browsers, so I was gonna file a bug on the url standard when I had the tests on wpt
11:52
<andreubotella>
but I got slightly tripped by the fact that webkit somehow doesn't percent-encode \x00 on urlencode over POST, leading to the same nonsense with the page being detected as a download
11:52
<andreubotella>
I'll file a bug for that one
12:08
<annevk>
That sounds like a fun, thanks for testing 🙂
12:23
<annevk>
andreubotella: that fileInput creation technique (with Object.assign) is rather clever
12:24
<andreubotella>
I saw bsittler use it in the original multipart/form-data PR that got closed downstream
12:25
<annevk>
Heh, must have missed that cause I don't remember seeing this before
12:28
<annevk>
andreubotella: let me know if you want it to be boilerplate-free and I'll wait a bit with merging
12:28
<andreubotella>
sure, I'll make it window.js
12:44
<annevk>
andreubotella: so I realized something since you filed that issue
12:44
<annevk>
andreubotella: this is testing form submission, and that is doing the newline normalization, and only after that is it testing urlencoding
12:45
<andreubotella>
annevk: in multipart/form-data, newlines in filenames aren't touched
12:45
<annevk>
andreubotella: I see that you are suggesting that newline normalization should only happen for urlencoding and not multipart, but I'm not sure if that's correct as it also happens when creating a FormData from a <form> (or perhaps only in Chrome)
12:47
<annevk>
I guess I should look up when we defined "append an entry" as I'm pretty sure we considered at least some of this
12:49
<andreubotella>
when I tested the newline normalization when creating a FormData from a form, I forgot to test filenames
12:50
<andreubotella>
I'll open a wpt PR for that later today
12:51
<annevk>
https://github.com/whatwg/html/commit/a1e8b1a9725dc716326f564b91a29aca19daf2f2 does seem to preserve intent in that it only happens for value if it's not a File, but maybe it could be read some other way that name should also not be affected?
12:51
annevk
will go further back in blame
12:55
<annevk>
I found something
12:55
<annevk>
https://github.com/whatwg/html/commit/d9fa54d643a7cd041ce31766f1ba5848cd76d18f
12:55
<annevk>
https://www.w3.org/Bugs/Public/show_bug.cgi?id=17460
12:55
<annevk>
zcorpan_ is to blame
12:56
<zcorpan_>
wat
12:56
<andreubotella>
I'll look at it later, g2g
12:56
<annevk>
And as was tradition back then there are no tests, links to implementation bugs, or anything 🙂
13:01
<zcorpan_>
ah I remember that issue. I was going to say it predated testharness.js, but that wasn't the case. https://github.com/w3c/testharness.js/commit/1780607fc4582b65b7fcdded3e39e79e95c8d915
13:04
<zcorpan_>
Has anyone written about the history of web-platform-tests?
13:04
<annevk>
I don't think jgraham has a blog or it must be very well hidden
13:30
<jgraham>
If I do I also hid it from myself
13:53
<jkt_>
:')
14:02
<andreubotella>
so apparently filenames just fell through the cracks?
14:02
<zcorpan_>
andreubotella: yeah
14:03
<zcorpan_>
I'd guess there are a lot of old spec changes like this
15:30
<smaug____>
Now it escapes my mind... what is the name for the proposal to add some new/better widgets/controls to the web
15:32
<zcorpan_>
smaug____: open ui
15:33
<smaug____>
thanks
15:43
<MikeSmith>
https://lists.w3.org/Archives/Public/public-html/2007Apr/0564.html from Maciej is arguably where the history of web-platform-tests started (to trace it all the way back)
15:44
<MikeSmith>
or from the same thread, https://lists.w3.org/Archives/Public/public-html/2007Apr/0532.html from DanC
15:46
<jgraham>
Well like with all history there are a number of different threads you can pull to unravel different parts of the story
15:46
annevk
finds https://lists.w3.org/Archives/Public/public-test-infra/2013JanMar/0052.html
15:47
<annevk>
I'm curious about the Opera side of the story
15:48
<jgraham>
Right, I think there's an intesting part that happened before I was involved where Opera ended up having a QA department with Hixie writing tests for web standards
15:48
<MikeSmith>
https://krijnhoetmer.nl/irc-logs/html-wg/20090127
15:48
<MikeSmith>
https://wiki.whatwg.org/index.php?title=Testsuite
15:48
<MikeSmith>
https://lists.w3.org/Archives/Public/public-html/2009Jan/0471.html
15:49
<jgraham>
And the way that interacted with the Acid tests for CSS
15:49
<annevk>
jgraham: I think Hixie was the department (at least the automated part of it) and he got Kam hired to replace him at some point
15:50
<annevk>
But yeah, that's 2003-2005 or so
15:55
<jgraham>
Good thing we fixed all the problems we discussed in 2009 about how to get tests reviewed and how to link tests to the spec in a reliable way.
15:55
<jgraham>
Otherwise it might look like we hadn't achieved much in the last decade or so :p
15:58
<jgraham>
On the other hand I guess these days we don't bother with an email every time a browser vendor contributes 43 tests.
16:17
<gsnedders>
jgraham: also where Hixie wrote tests while a student ;P
16:18
<gsnedders>
MikeSmith: I forget the exact year when it was decided that CSS 3 should pretty much all be automated tests
16:19
<gsnedders>
annevk, jgraham: Hixie was an instrumental part of getting the tooling to run automated tests working within Opera, and it was that role which Kam was hired to take over, allowing Hixie to focus on the actual testing side of things
16:21
<MikeSmith>
gsnedders: was before 2007 maybe?
16:22
<MikeSmith>
at least I think the CSS test suite work started quite early
16:23
<MikeSmith>
Tantek would remember I guess
16:36
<MikeSmith>
annevk: in the general case where some server-side application code for CORS handling throws for some reason, can you think of an HTTP status code that can be used to convey it?
16:36
<MikeSmith>
context is https://stackoverflow.com/questions/42755131/enabling-cors-in-cloud-functions-for-firebase/65127649#65127649
16:37
<MikeSmith>
... where the person has the applicaion code intentionally written to send a 500 response
16:37
<annevk>
MikeSmith: 403 per https://fetch.spec.whatwg.org/#http-responses
16:37
MikeSmith
looks
16:37
<annevk>
(last paragraph)
16:38
<MikeSmith>
ok yeah but that says “In case a server does not wish to participate in the CORS protocol”
16:38
<MikeSmith>
which makes sense
16:40
<MikeSmith>
but I often see people having application code written to send 500 responses in code that’s intended to participate in the CORS protocol but that throws due to some unexpected thing
16:40
<MikeSmith>
to me, in general, it seems not optimal to send a 500 from application code — because that’s gonna make it harder for you to distinguish cases of your application code throwing an intended error from cases of the server/runtime code hitting some other kind of internal failure (maybe even before it ever executes your application code)
16:42
<MikeSmith>
(but maybe that’s just me, and sending a 500 in such cases is actually a normal best practice)
16:42
<annevk>
MikeSmith: we could clarify that line I suppose as it's a little vague
16:42
<MikeSmith>
OK
16:42
<annevk>
MikeSmith: what it really means is that if the server doesn't want to allow the particular request (or participate in the CORS protocol at all), do this
16:43
<annevk>
but it's still a suggestion since servers can do whatever
16:43
<MikeSmith>
sure
16:49
<Domenic>
MikeSmith: 500 makes sense for me in cases like that. E.g. in all my Node servers, if an exception gets thrown while processing a request/preparing to create a response, we turn it into a 500 response. That code snippet just looks like someone manually doing the try/catch, instead of letting it bubble to their server-level error handler, which is somewhat understandable since I'm not sure how well generic server
16:49
<Domenic>
request/response handling deals with preflight-type things.
16:50
<Domenic>
(But, the OP's code for that question makes no sense, where they're always sending a 500.)
16:57
<MikeSmith>
Domenic: ah yeah, good point about this case — about just letting the exception bubble rather than catching it
16:59
<MikeSmith>
but otherwise yeah I guess if there’s nothing more specific that the application code can determine, then 500 works
17:30
<gsnedders>
MikeSmith: pre-CSS3 it was all manual stuff though, that's the thing
17:30
<gsnedders>
well "manual", it was all screenshot based
17:37
<gsnedders>
MikeSmith: https://www.w3.org/blog/CSS/2009/06/23/resolutions_69/ seems like it was only June 2009 that reftests were officially adopted for CSS
17:53
<andreubotella>
is there an issue open anywhere for wpt.fyi trying to link to the nonexistent html source in github rather than to an .any.js?
17:55
<andreubotella>
nevermind, found it
17:55
<andreubotella>
https://github.com/web-platform-tests/wpt.fyi/issues/1476
20:04
<andreubotella>
annevk: I misinterpreted your earlier comment about FormData in Chrome
20:05
<andreubotella>
I wrote down the whole story about why this is a urlencoded-only thing in https://github.com/whatwg/url/issues/562#issuecomment-738276174