| 09:19 | <annevk> | Hixie: hmm, can investigate tomorrow |
| 17:03 | <gsnedders> | Philip` reached a point of eventually autogenerating code for the tree constructor as well as the tokenizer, right? |
| 17:03 | <gsnedders> | Because some of the stuff (around doctype tokens esp.) looks pretty hidious to parse the text. |
| 17:09 | <jgraham> | Pretty sure that at some point Philip`'s code became self aware |
| 17:10 | <jgraham> | But the government keeps it secret |
| 17:10 | <jgraham> | That's why you never hear from him these days |
| 17:11 | <gsnedders> | Well that would explain that. |
| 17:28 | <Hixie> | i had big plans to parse the spec and turn it into code when i started this parser i'm doing |
| 17:28 | <Hixie> | but then i ended up making it output things a bit different than the spec does |
| 17:28 | <Hixie> | so that went out of the window |
| 17:29 | <Hixie> | (e.g. the way i handle character tokens that are in the original input is to just return pointers to the original string to avoid string copies) |
| 17:49 | <gsnedders> | Hixie: I may write such a thing over the summer |
| 17:49 | <gsnedders> | Hixie: Where I'm really not interesting in performance or memory usage, provided it's "usable". |
| 17:51 | <Ms2ger> | A natural language parser for en-US-x-hixie? |
| 17:53 | <gsnedders> | The parser's spec text is deliberately regular so you can quite easily parse it. |
| 17:54 | <gsnedders> | I'm not interested in general en-US-x-hixie |
| 17:54 | <Hixie> | heh |
| 17:54 | <Hixie> | gsnedders: yeah. my goal is to make a really fast one since my use case is "anolis is slow" |
| 17:55 | <Hixie> | (and that's with anolis using libxml2 sometimes, iirc) |
| 17:56 | <gsnedders> | Essentially, I want to be able to: a) check html5lib tests for correctness v. current spec; b) generate more tests for paths not covered (so both tokenizer and tree constructor tests are complete); c) formal assertions of properties of the spec, as then merely the code to generate the model need be checked (instead of checking the whole model v. the spec). |
| 17:57 | <gsnedders> | Hixie: honestly, I don't think the parser is *that* slow. the problem is the serializer and the passes --- you want to be able to merge the passes to operate over fewer loops over the document. |
| 18:07 | <gsnedders> | Really what you want is to use something with good loop fusion optimization |
| 18:21 | <IZh> | Is there WHATWG's version of CSS standard? Or do you agree with W3C about it? |
| 18:23 | <Ms2ger> | We don't have the manpower to deal with it |
| 18:29 | <IZh> | Ms2ger: But do you have your own vision of CSS3 or W3C's version is suitable for you? |