| 03:54 | <sideshowbarker> | TabAtkins + all: I’ve added a CI cron job in the HTML checker repo that runs twice day, building Bikeshed output with https://www.w3.org/publications/spec-generator/ and then running against the checker to make sure it has no errors. |
| 03:56 | <sideshowbarker> | That way, I can hopefully get alerted to checker breakage before any spec editors do — or at least at around the same time any editors might start running into it — and start right away on fixing the checker before more editors are affected. |
| 03:57 | <sideshowbarker> | Context is that yesterday, both the WebExtensions and WebDriver spec builds broke because the Bikeshed output gained some CSS stuff that I hadn’t yet taught to the checker. |
| 03:59 | <sideshowbarker> | …And I only found out about it because of getting Cc‘ed on issues from those spec repos. |
| 08:23 | <annevk> | smaug: ping on the OpaqueRange cleanup PR. Also, I briefly looked at menu elements yesterday and it seems that at least menuitem also looks at its parent for behavior branching. In particular for submenu it will do act more like a button. Was curious what you made of that. |
| 09:16 | <smaug> | yeah, I'm still thinking if that is the right model for custom element case |
| 09:16 | <smaug> | where boundary points should be all internal |
| 09:17 | <smaug> | and exposed offsets could be anything |
| 09:27 | <smaug> | oh, I guess the pr is fine with that |
| 09:31 | <annevk> | Yes the PR allows for that |
| 09:32 | <annevk> | I'm not sure why Stephanie Zhang didn't spec the custom element variant immediately, but I did check OpaqueRangeController can be done on top of her (and my) work and afaict it fits. |
| 09:33 | <annevk> | zcorpan: very exciting to see progress on ORB again. AI found a couple of things. Would be nice to finally get it merged. |
| 13:59 | <zcorpan> | annevk: yep! farre and I are working through it. Will continue tomorrow. Thanks for the ai review |
| 14:01 | <zcorpan> | Looks like I'll need to open a new PR for the Participation check, but I'll do that later |
| 14:26 | <zcorpan> | Done, comments are editorial |
| 14:31 | <zcorpan> | Very nice. I need to link to this in my book |
| 14:32 | <zcorpan> | The illustration in https://htmlparser.info/parser/ looks almost the same :) |
| 14:34 | <Tim van der Lippe> | As a community member who contributes to web standards because of the community feeling: it makes me sad to see human interaction fade away with more and more of my PRs to web standards reviewed only by AI. It removes the human aspect of the "web community" for me and makes it less enjoyable to contribute and more of a chore. I don't get paid to do this, so my own motivation keeps me going. AI reviews don't help with that. No need to start a whole debate about this. Only wanted to share my current feeling and needed to write it in a place where I know humans read the actual stuff. Thanks for reading humans, would love to continue reading your not-so-perfect sentences. Please continue your day as you were planning to. |
| 14:40 | <zcorpan> | Does it work for the tree builder? |
| 14:44 | <zcorpan> | asuth: we need to reimplement the spec PR preview thing to have something like the searchfox view. Could support stacked PRs maybe! And ability to leave review comments from that view |
| 14:51 | <annevk> | Tim van der Lippe: I wish there was a way that would both catch the various nuances and still make it feel like a collaboration. |
| 14:52 | <annevk> | (Your contributions haven't gone unnoticed to me, FWIW. They're very valuable.) |
| 14:55 | <Noam Rosenthal> | Aye. Perhaps it's wishful thinking but I hope that this is a transition period that is going to converge on machines catching more nuances and humans spending a larger percentage of the time discussing design choices |
| 16:45 | <dbaron> | I don't know exactly what Tim van der Lippe is experiencing in terms of AI reviews... but maybe one approach that might make it AI reviews feel less awkward is if the infrastructure for them is automated and shared so that they're more like a lint that runs automatically on people's PRs at a stage that's separate from human review? |
| 16:52 | <Noam Rosenthal> | This was discussed at TPAC and we are actively investigating how to fund/orchestrate something like this. |
| 17:05 | <zcorpan> | Tim van der Lippe: thanks for raising it. For https://github.com/whatwg/html/pull/13057 the remaining feedback is editorial; I can push fixes for those. I'll think about this and see if I can adjust to avoid the dry AI review friction |
| 20:02 | <zcorpan> | Maybe we can make the review skill first focus on behavior and normative changes, and present options for fixes (if appropriate), and just note editorial review and nits is for later. We can then discuss those issues and the merits of each option. When there are no more design decisions or normative issues, it switches to an editorial only review and fix loop, with verification that it didn't change the normative behavior. (Editorial cleanup first could be on-demand.) If the editorial fixup finds new non-editorial issues, it raises them as new open questions. I think this could work both locally (for contributors and editors) and on shared infra. |
| 20:27 | <Noam Rosenthal> | ship it |
| 21:22 | <zcorpan> | I wrote a long prompt |
| 21:24 | <zcorpan> | wpt review should probably live in the wpt repo |
| 22:22 | <zcorpan> | https://github.com/whatwg/spec-ai/issues/8 |
| 23:52 | <zcorpan> | https://github.com/whatwg/spec-ai/pull/9 , with the skill used to review itself. Many found defects/questions means a wall of text, but contributors can run it locally and address defects before opening the PR. And yes I realize this is still AI review, but I think this flow allows for better collaboration. |