09:09
<keithamus>
Is there something we can do about the pr-preview[bot] using stale data to overwrite issue comments. It makes it quite error prone to edit PR descriptions
09:10
<keithamus>
In the image above you can see i added a webkit bug for this PR, but the pr-previewbot updated the preview but in doing so used state comment data and undid the change, removing the webkit bug.
09:17
<annevk>
We have instructions for how to deal with this in the PR template as I mentioned. The alternative would be attempting to patch https://github.com/tobie/pr-preview I suppose. I'm not entirely sure how that would work as I think the comment has to be submitted as a whole to GitHub so there would always be a bit of a race.
09:18
<annevk>
(And you could imagine alternatives such as PR Preview always creating its own comment, but that adds noise for everyone that's not really desirable.)
09:45
<annevk>
Ideally GitHub would have some way for a bot to annotate issues/PRs with Markdown in a way that doesn't result in notifications, but we lost our resident GitHub person. 😀
09:47
<keithamus>
I think typically bots add a comment to the timeline. There's also the "deployments" feature which was theoretically designed to solve this kind of issue - it adds a new timeline entry. Both of these have quite large trade-offs compared to updating OP though.
09:47
<Ms2ger>
If "minimal notifications" is also acceptable, we could have a single comment that the bot updates as needed
09:49
<annevk>
WPT used to have a single comment and that was a huge source of frustration. At least for Domenic and I.
09:50
<keithamus>
https://github.com/tobie/pr-preview/pull/163 hopefully this goes some of the way to minimising the issue at least.
12:08
<farre>
Noam Rosenthal: hey, I'm trying to figure out a buggy navigation api test case. I'm looking at navigate-event/navigate-navigation-navigate.html, and tracing it, I can only find that this should fire a navigate event with navigationType == "push". It does so in gecko, but chromium disagrees
12:09
<farre>
(and I got extra eyes from smaug, so I don't think I'm missing anything)
12:11
<smaug>
location.href setter doesn't behave like navigate() when it comes to 'completely loaded' checks.
12:45
<Noam Rosenthal>
Taking a look.
12:46
<annevk>
zcorpan: how do you want to handle that reformat thing? Should I rubber stamp it? Will a follow-up Meta commit ensure blame is not impacted?
12:47
<annevk>
Also not sure if jmdyck wants to take a look before it happens.
12:57
<Noam Rosenthal>
farre: yes it seems like chromium follows the location.href behavior of setting the history handling to replace when not completely loaded. However that's not in the spec.
12:57
<farre>
ok, cool. In an upcoming patch I'm updating the test to reflect spec
12:59
<jmdyck>
I'm still looking at the reformatting PR, I'm just delayed because the merge of the algorithm PR caused some re-org of my code
13:44
<zcorpan>
annevk: I think I'd like to add the tool to CI so that we can enforce it for PRs. Maybe integrate with html-build
13:47
<zcorpan>
annevk: For blame, one needs to specify to skip that commit manually, right?
13:48
<annevk>
zcorpan: https://docs.github.com/en/repositories/working-with-files/using-files/viewing-and-understanding-files#ignore-commits-in-the-blame-view
13:49
<zcorpan>
Ah, yeah can do that as a followup
13:49
<annevk>
I'm hoping that Searchfox plays ball, maybe asuth can confirm.
14:34
<jmdyck>
Is there interest in having a more comprehensive formatter for the HTML spec, that enforces more style guidelines (think rustmt or gofmt)?
14:41
<annevk>
jmdyck: yes, definitely! That would make life easier for reviewers and contributors.
22:17
<mfreed>
Hi all, just a friendly reminder to post any discussion topics for this Thursday's joint CSSWG/WHATWG/OpenUI task force meeting to the meeting agenda issue: https://github.com/whatwg/html/issues/11670