Today work has focused on finishing an initial “stable” release for this page (the one for daily reports,) and on further continuing work on the Typst compiler.
Yesterday I had almost finished work on the page associated with the repo where these reports (now) get published at. The only thing left was to have some messed up Just recipes fixed, so that the files actully got properly compiled.
This was an easy fix. I then added caching support for the Typst compiler version and ripgrep, so that I could speed up the CI runs. This was fairly straightforward to set up with the “built-in” GitHub Actions action.
Then I worked on brushing up some stuff that was left on the repo, mostly related to the README and LICENSE files. Then I tweaked some the justfile so as to wire up all dependencies between lazy expressions and recipes.
The site is now deployed, and I have rebased main to reflect only a single commit with all the changes up to the “stable” release. The prior commit history is kept in a separate branch for posterity on the timestamps of the first two daily reports.
Work on the Typst compiler has progressed at the expected pace for today’s task. I have been working on the let-binding analyzer, which should provide a query for Menhir semantic actions to call into and ensure basic (dumb) bindings that use some form of pattern are catched at parse time.
This initially consisted of a few nested recursive functions that interleaved their execution to sort out all expressions that I can interpret in-place (i.e. anything that doesn’t require a compilation context, like identifiers or callables.)
Then I tweaked that some to instead filter on certain “literal” expressions that I can merge in code blocks. This way, if at parse time I find any expresion on the right hand side of a let-binding that requires compilation context, I simply bail out early.
This is still going to require some more thought, though, and I am very much inclined to end up not using it. I don’t think it’s worth it to do a half-assed job at evaluation during parse time, if I can’t resolve all expressions then. Leaving that out for later is likely the best way out.
None. Today I missed a few hours of work in the morning, but I don’t think I lost them. I got the chance to try and explain to somebody my conclusions from studying basic universal constructions in category theory (initial and terminal objects, products and coproducts.)
The site is done. That means I can go back to working on rust-lang/libc, as I’ve received a bunch of notifications there from GitHub PRs and issues I’m involved in. That should start taking up the time I spent on the site for the last three days.
The current work on the compiler will be finished but will likely not be used. As mentioned in today’s summary, I’m fairly confident partial evaluation is not worth it here. But I want to finnish exploring that implementation to get a better feel once I start working on the evaluator (and once I have a proper compilation context available.)