[WIP] Calling external formatters

[mkatychev]

Hey all, what are your thoughts on exposing a way to provide a formatter for the injection work, something like to allow external formatters for injections:

{
  languages = {
    go = {
      extensions | default = ["go"],
      injection_formatter = "gofmt",
      # this grammar is only for injections
      # grammar = null,
    },
    # ...
}

Motivation

In the same way that foreign languages are treated as a black box by tree-sitter grammars (a SQL string inside of a Rust AST/CST is just a string), foreign formatters should be treated in the same way std::fmt::Formatter obfuscates the underlying dyn Writeer.

So whether the formatter is "local" or a CLI call for a foreign binary, they produce the same effect

• Topiary having to compete with native formatters like rustfmt or prettier is a battle that's hard to sell.
• injections and off-the-shelf tree-sitter grammar usage further reinforces my view of topiary as the polyglot "glue" platform.
• Tree-sitter allows topiary to offer grammatical integrity (with things like topiary check-grammar) that contemporaries such as treefmt can't really compete on because there is no lexical scope

[Xophmeister]

Would it also make sense for an external formatter to also target host languages?

For example, as you say, we're not going to compete with rustfmt, but we can offer macro formatting with Topiary, which solves a genuine problem. Something like this:

{
  languages = {
    rust = {
      extensions = [ "rs " ],
      query_dir = "path/to/wherever",  # say, only `injections.scm` exists
      formatter = {
        binary = "rustfmt",
        extra_args = [],
      },
    },

    json = {
      # As currently defined
    },
  }
}

That is, when Topiary sees a Rust source, it knows to use rustfmt because a binary is defined but, because injections.scm also exists, it also knows to do injection formatting for any queries that match.

[Xophmeister]

(The same logic applies when Rust source is embedded in something else. i.e., Injections aren't special-cased, but rather all languages are first-class citizens regardless of whether Topiary formats them or calls to an external tool.)