This is still potentially incorrect, due to trailing comments.

Instead, given that parsing is a relatively intense operation, I would prefer that we simply count the number of
lines in file.fileNode and the number of characters on the last line, and construct a Range using that. Don't bother with the spans in the AST, since they clearly don't reflect what we actually mean.

Also, include the following comment:

// XXX: The current compiler does not expose a span for the full file. Instead of
// potentially undershooting the correct span (which can cause comments at the
// start and end of the file to be duplicated), we instead manually count up the
// number of lines in the file. This is comparatively cheap, compared to sending the
// entire file over a domain socket.

Also, please file a bug against protocompile to include a way to retrieve a span of the whole file.

FWIW, it doesn't seem like the issue extends to trailing comments, maybe because NodeInfo.End() has special handling for EOF?

I would prefer that we simply count the number of lines in file.fileNode and the number of characters on the last line, and construct a Range using that.

I was looking around at the FileNode APIs and didn't find much exposed that would help with that; mind pointing me to how that would be done / make that change yourself? (Happy for you to take this over; my protocompile knowledge is weak 😅.)

I'm juggling a bunch of PRs already and am about to go on vacation, so I prefer you do it if only to make sure it gets done. Also, good excuse to grow your knowledge!

Anyways, the right way to do this:

var lastLine, lastLineStart int
for i := 0; i < len(file.text); i++ {
  // NOTE: we are iterating over bytes, not runes.
  if file.text[i] == '\n' {
    lastLine++
    lastLineStart = i + 1 // Skip the \n.
  }
}
lastChar := len(file.text[lastLineStart:]) - 1 // Bytes, not runes!

0057f0d, PTAL!

Also, please file a bug against protocompile to include a way to retrieve a span of the whole file.

bufbuild/protocompile#360!