10:00
<andreubotella>
annevk: w.r.t https://github.com/whatwg/html/pull/6287#issuecomment-762646151, see https://github.com/whatwg/html/issues/6247#issuecomment-753997698
10:00
<andreubotella>
I might have done a poor job explaining, but although "string/File-sourced entries" include programmatic inputs (such as some FACE inputs), "FormData-sourced entries" are all programmatic
10:02
<andreubotella>
so my initial idea was to keep things as they were in the spec, other than normalizing FormData-sourced entries for urlencoded and text/plain since Firefox and Chrome already do that
10:06
<annevk>
andreubotella: I might not have been clear enough either, I think I meant to not normalize them at all
10:07
<annevk>
andreubotella: it's kinda moot without implementer input though and I suspect it might be hard to get a quick take from someone given all the intricacies
10:07
<andreubotella>
annevk: true
10:09
<annevk>
andreubotella: I think you had a summary comment somewhere about the current state of things in implementations, perhaps that can be updated with the proposed state and then we get feedback on that?
10:09
<annevk>
andreubotella: pretty sure I can reach someone from Chrome; I think for Mozilla it doesn't matter much as long as we end up with interop
10:11
<andreubotella>
annevk: turns out that for Firefox, doing it some other way than the current state of https://github.com/whatwg/html/pull/6287 would need fixing https://bugzilla.mozilla.org/show_bug.cgi?id=1651887 first
10:12
<andreubotella>
but sure, I'll write that up
10:13
<annevk>
andreubotella: that bug is a good indicator that folks don't want their FormData "magically" changed
10:14
<annevk>
(though more likely is that it's anecdata that reinforces my biases 🙂)
10:17
<andreubotella>
annevk: Firefox and Chrome agree on normalizing FormData-sourced entries on urlencoded and text/plain, though
10:18
<andreubotella>
and Safari only doesn't because it can only create a payload from FormData-sourced entries with Request/Response.body
10:20
<andreubotella>
so if we don't want to normalize FormData entries on Request.body, the shortest path to interoperability would probably going with Chrome for multipart/form-data
10:23
<annevk>
andreubotella: I'm sorry, but I no longer know what that means
10:23
<annevk>
Interop matters, but if everyone has to do a little more and we end up with something a lot simpler that's valuable too, esp. long term
10:55
<annevk>
andreubotella: what does (names and filenames N/A) mean?
10:56
<andreubotella>
annevk: Firefox turns newlines into spaces in the MIME headers in multipart/form-data
10:56
<andreubotella>
with https://github.com/whatwg/html/pull/6282, that's no longer conforming to the spec
10:56
<annevk>
andreubotella: that might be worth noting
10:57
<andreubotella>
right
10:57
<annevk>
andreubotella: I hadn't really considered that names would become headers; can you even reliably parse that format if those contain newlines?
10:58
<andreubotella>
https://github.com/whatwg/html/pull/6282 makes them percent-escaped
10:58
<annevk>
I see
10:58
<annevk>
That should work
10:59
<andreubotella>
That's why I waited to file both PRs together in the same Firefox bug
11:01
<annevk>
andreubotella: I still like the idea of normalizing for String/File and not normalizing for FormData
11:03
<andreubotella>
annevk: I'd prefer normalizing everything for urlencoded and text/plain and going the Chrome way on multipart/form-data
11:03
<andreubotella>
but let's wait to hear from the implementers
11:04
<annevk>
andreubotella: why is that your preference?
11:06
<andreubotella>
annevk: I'm not sure; the other options are definitely simpler
11:07
<andreubotella>
I don't care too much which way we end up going for multipart/form-data
11:51
<andreubotella>
annevk: I was thinking that for urlencoded and text/plain, it might not be simpler to not normalize for FormData, spec-wise and for implementations like Chrome
11:52
<andreubotella>
I'll try to put my words in order and write a post about it
13:52
<annevk>
JakeA: so how do you trick a media element into making multiple range requests?
14:03
<JakeA>
annevk: give it back less than it wants. If it asks for 0-, give it 0-100 or whatever
14:03
<JakeA>
it'll make another request for the rest
14:03
<JakeA>
either that, or seek the media
15:47
<annevk>
JakeA: so both Chrome and Firefox seem to give up for cross-origin audio, even without redirects, not sure what to make of this yet (perhaps DevTools is broken)
15:48
<JakeA>
annevk: give up, as in don't make a second range request?
15:48
<annevk>
JakeA: yeah, and Firefox doesn't even show the first request
15:48
<annevk>
JakeA: get a single progress event in Chrome, two in Fx
15:50
<JakeA>
annevk: https://github.com/jakearchibald/range-request-test might be useful, as this logs the requests the server receives
15:51
<annevk>
JakeA: yeah, looking at WPT stashes now
15:52
<JakeA>
annevk: https://github.com/web-platform-tests/wpt/blob/master/fetch/range/general.window.js here's what I did for the range stuff in wpt
15:52
<annevk>
JakeA: yeah I'm using some of that 🙂
15:52
<annevk>
JakeA: if I get somewhere you might even get a review request
15:52
<annevk>
but prolly not today at this rate
15:53
<JakeA>
😀 happy to
16:20
<annevk>
JakeA: it's really weird, if I append a uuid it doesn't happen anymore; maybe it hits some cache somewhere
16:55
<JakeA>
annevk: hmm, it's possible. Does no-store produce the same effect?
17:01
<annevk>
JakeA: does not seem to matter; I've also "disable cache" enabled
17:01
<JakeA>
huh, that is odd
17:01
<annevk>
JakeA: anyway, not investigating that too much, looking more at redirects and such now
17:02
<annevk>
JakeA: it seems that Chrome might stop at the first redirect, Firefox at the first cross-to-same-origin redirect, Safari keeps trucking, I guess I need to test cross-to-cross origin
17:03
<JakeA>
annevk: I seem to remember that Chrome would 'remember' the redirect, and consult the redirected-to URL for subsequent parts, whereas Firefox would go back to the original URL
17:03
<JakeA>
Or the other way around
17:06
<annevk>
JakeA: I can reproduce Firefox continuing with the original URL, but Chrome stops after it gets the 206 response from the redirect
17:07
<annevk>
Which is admittedly weird and maybe a sign I'm doing something wrong as it would make more sense for Chrome to just abort upon seeing a redirect after the initial fetch
17:08
<annevk>
JakeA: ah, if the initial request redirected I can reproduce that behavior in Chrome
17:08
<annevk>
JakeA: but not if the second request redirects
17:09
annevk
basically branches of a request counter to try to make browsers do stupid stuff
17:09
<JakeA>
ohhhh! I think that behaviour is new to me
17:10
<annevk>
JakeA: maybe it still follows redirects in the hope that the final URL is identical and it will continue in that case? That does kinda match what mkwst described
17:12
<JakeA>
annevk: As in redirects to another origin and back again? That would surprise me, but everything about this stuff surprises me
17:41
<annevk>
JakeA: as in, if the first request ends up at /x and the second request thus goes to /x but then is redirected to various places including https://y/test but ends up at /x, it's okay (not verified though, will likely do so tomorrow)
21:10
<smaug____>
Are some directories in wpt magical, so that files in them aren't run?
21:11
<smaug____>
Or do all the .html files with <script src="/resources/testharness.js"></script> or <script src="/resources/testharnessreport.js"></script> or something get run as a test
21:25
<Domenic>
The latter