Tracking issue: Typst support

[alexpovel]

TODO:

• richer Typst support, WIP at https://github.com/alexpovel/srgn/tree/typst

• make it useful with type checking (aspell), for which the text node type seems most useful

• which requires piping into aspell, for which srgn currently doesn't have good support. Probably needs refactoring the crappy apply function:

  srgn/src/main.rs

  Line 568 in 9238fc1

  fn apply(

  so its output is something like (command run on commit 47be5f0)

  [filename if not stdin];linenum;startcol;endcol: text
  

  e.g. (numbers made up)

  $ cat tests/langs/typst/base.typ | c run -- -j --typ text --typ strings
  [...snip...]
  62;12;13: In case you missed it, there's a recording on
  62;58;62: "http://purview.edu/lts/2023-lee"

  i.e., it doesn't print non-text nodes such as #link(, as it currently does:

  

cc @ju6ge

[alexpovel]

Kinda blocked, as I'd prefer not vendoring and only using grammars part of the tree-sitter org. Otherwise, it can get pretty painful with updates and keeping things in sync (see HCL support in srgn).

[lysogeny]

For spell checking I was wondering if it might be sufficient to strip out the markup, similar to what the TeX tool detex does, only making this print the selected nodes in some mode. The resulting file could then be piped into a user's favourite spellchecker and should, for that task, probably be good enough?

[ju6ge]

For me the interesting use case would be CI Pipelines. In that case it would be interesting be able to map the spelling error to the line and column of the file in which it occurred. But for that srgn would need to print the filename and line numbers, so that downstream tools could keep track of it correctly. (It does this currently but not in a from that is easily pipeable…) On the other hand I have yet to find a spell checker that will actually give me the desired output to make this work correctly, they seems to either only print a list of misspelled words or use a custom stream format to not just show the misspelled word, but also suggest fixes.

[alexpovel]

Thanks for your thoughts!

I had considered skipping the whole part about "if srgn is piped into something that is not a terminal, use machine-friendly/parseable output" and instead just offer JSON output. That's much more flexible, and (probably) easier to implement at the same time. It avoids work on my side (coming up with a parse-friendly format), but also allows full flexibility on the reading side of the pipe. It could look like srgn --whatever | jq '.whatever' | spellcheck. You would "only" need to massage the jq query, which is infinitely more powerful than what srgn could ever offer (and an LLM could probably write it very well for you). Let me know if you have any thoughts on that.

(work on srgn is currently slow, sorry!)

[ju6ge]

I am not opposed to json. Might very well be more flexible when transformed with jq. The blocker on my side ist that I have not found a spell checking tool gives me simple enough output that can be further processed easily. So no need to apologise for slow development.

[alexpovel]

There's now parseable/pipeable output formatting, done in #219 and released as part of #212 🙂

Next step is for a tree-sitter Typst crate to become blessed and available: https://crates.io/search?q=tree-sitter-typst