Add machine-readable compile error output for tooling (LSP) use

[m00qek]

Hello there, I'm building a language server for ucode on top of my tree-sitter grammar. The LSP can lean on the grammar for parsing, scopes, symbols, etc., but for semantic compile-time diagnostics it could invoke ucode -c and surface the results in the
editor.

The current human-readable stderr is parseable with a regex but brittle, as the format mixes the message, an optional "Expecting X" hint after a \n, the optional filename clause (only present in multi-source mode), and a ^-- Near here block. Any minor formatting change in uc_compiler_syntax_error would silently break the LSP.

A machine-readable output mode would make this robust and is a small change given the current code structure.

My proposal is to add a flag (something like -J) that switches compile error reporting from the current human-readable text to JSON Lines, one diagnostic per line:

{"file":"x.uc","severity":"error","line":5,"col":12,"message":"Unexpected token","hint":"Expecting ')'"}

Do you think that the idea is valid? If so, I can work on a PR for it :)

[jow-]

An alternative would be evaluating the code at runtime (loadstring(), loadfile(), ...) within try { ... } catch (e) { ... } and examining the structured stack trace.

Btw - could you check whether you could base your work onto https://github.com/jow-/tree-sitter-ucode and/or merge your grammar with the one above?

[m00qek]

An alternative would be evaluating the code at runtime (loadstring(), loadfile(), ...) within try { ... } catch (e) { ... } and examining the structured stack trace.

Can't eval the code cause undesirable side effects? I'd need to sandbox it somehow

Btw - could you check whether you could base your work onto https://github.com/jow-/tree-sitter-ucode and/or merge your grammar with the one above?

Oh, I missed your impl! I think the biggest diff is that I wrote two grammars (one for template, one for "code")... It seemed easier to reason about and I could base the "code" one on the JS grammar

[jow-]

The load*() functions will compile the code into a function but not execute it so it should be safe.

$ ucode -e '
  try {
    loadstring("print(`hello missing quote)");
  } catch (e) {
    printf("Syntax error in code: %.J\n", {...e});
  }
'
Syntax error in code: {
	"type": "Runtime error",
	"message": "Unable to compile source string:\n\n  | Syntax error: Unterminated string\n  | In line 1, byte 7:\n  |\n  |  `print(`hello missing quote)`\n  |         ^-- Near here\n",
	"stacktrace": [
		{
			"filename": "[-e argument]",
			"line": 3,
			"byte": 45,
			"context": "In [-e argument], line 3, byte 45:\n\n `    loadstring(\"print(`hello missing quote)\");`\n  Near here ----------------------------------^\n"
		}
	]
}