Are formatters allowed to make "best effort" attempts to format invalid files?

[jfly]

This came up in NixOS/nixfmt#378: what should a formatter do if it sees an invalid file but wants to format it anyway? For example, tree-sitter has this concept of being "robust": that is, it will produce an AST with error nodes if possible. One can imagine formatters that behave similarly: they could still format syntactically invalid input.

IIUC, this is already possible: a formatter that sees a syntactically invalid file can just update the file on disk and print errors to stderr and exit nonzero. I believe this is spec compliant behavior, although it would probably be nice to call it out explicitly in the spec.

However, I believe treefmt --stdin would behave differently: we'd see the formatter exit nonzero and conclude that the file cannot be formatted. You could imagine that we'd instead print the formatted file to stdout, but still print errors to stderr and exit nonzero. Surrounding tooling could then decide how to handle that (either ignore it entirely, or accept the best-effort formatting). This also has implications if we ever add a stdin spec: #586