| 00:58 | <asuth> | zcorpan: I got the searchfox wubspecs diagramming working with the HTML parser algorithm-ish things: for comment-less-than-sign-bang-dash-dash-state with depth cranked up, screenshot of part of it above. |
| 01:00 | <asuth> | somewhat amusingly, the literate programming algorithm view for that algorithm that's supposed to pick out firefox source code that references the algorithm picks up on an html parser lib inside webpack that's in the tree. screenshot above. |
| 10:02 | <Sam Sneddon [:gsnedders]> | https://github.com/mozsearch/mozsearch/blob/hyperblame/tools/src/spec_analyze/wattsi.rs is what's driving this? |
| 18:32 | <asuth> | Sam Sneddon [:gsnedders]: Yes, all the firefox-disco stuff is from the hyperblame branch there, which is a very tall stack of claude byproducts built on top of a bunch of my human prior searchfox work and my personal searchfox to-dos/design notes/investigations/etc,. The exception is that for bikeshed specs the semantic indexing comes from the bikeshed parser and specifically from https://github.com/mozsearch/bikeshed/blob/mozsearch-index/bikeshed/index/__init__.py which is based on my prior initial human efforts/research there (the enhancements to the bikeshed parser were very transformative -- to be clear, I mean the upstream enhancements to the AST, nothing I did or claude roboted), and in theory that branch adds a nice "index" command that will let anyone get nice JSON output out of bikeshed maybe? |
| 18:40 | <asuth> | Another hopefully useful thing (at least for me if no one else), is that searchfox's wubspecs repo can now sorta sync the list of outstanding spec PRs and then sorta ingest those to try and help make it easier to understand what's going on with PRs or different PRs. So the workflow is sorta, someone goes to https://searchfox.org/wubspecs/spec-prs and hits the sync button there which then pulls down the PR metadata cheat-sheet that https://github.com/mozsearch/spec-pr-state crated via an hourly github action, which then lets the sync stuff just do a bunch of git operations. Then I click on one of the specs there and end up on a spec PR listing page like https://searchfox.org/wubspecs/spec-prs/fetch for fetch and then if all the PRs aren't already imported and there are less than 50 of them, there is an "import" button and clicking it should import them all which then tries to ~semantically index them, which includes running the above-mentioned bikeshed parser for its semantics and magic. If I then click on one of the "with pull requests" links, I can end up (with anchor for example) at https://searchfox.org/wubspecs/rendered-prs/fetch/fetch.bs#destination-type which by default tries to show all of the effects of all of the imported PRs in a second column next to the current trunk spec with rendered blame (hover on the stuff on the left to see blame). There's actually a faceting UI at the top of the page which probably should move into the fancy bar on the right like it is elsewhere in searchfox. But so you could turn off already approved PRs or stuff. Screenshot above. |
| 18:41 | <asuth> | And here's a service-workers example which I know is not WHATWG but I like service-workers: https://searchfox.org/wubspecs/rendered-prs/service-workers/index.bs#installevent (screenshot above) |