Herestring not recognized _after_ stdout redirect.

[danfuzz]

It looks like the parser has trouble with herestrings that come after a redirect of stdout. For example, a parse of the following indicates an error on the here-string token <<<:

cat >x <<< 'x'

Parse tree:

(program [0, 0] - [1, 0]
  (redirected_statement [0, 0] - [0, 14]
    body: (command [0, 0] - [0, 3]
      name: (command_name [0, 0] - [0, 3]
        (word [0, 0] - [0, 3])))
    redirect: (file_redirect [0, 4] - [0, 6]
      destination: (word [0, 5] - [0, 6]))
    (ERROR [0, 7] - [0, 9])
    redirect: (file_redirect [0, 9] - [0, 14]
      destination: (raw_string [0, 11] - [0, 14]))))

Compare that to the following, which works:

cat <<< 'x'

Parse tree:

(program [0, 0] - [1, 0]
  (command [0, 0] - [0, 11]
    name: (command_name [0, 0] - [0, 3]
      (word [0, 0] - [0, 3]))
    redirect: (herestring_redirect [0, 4] - [0, 11]
      (raw_string [0, 8] - [0, 11]))))

[danfuzz]

I think the root cause here is that the grammar in this project has explicitly different rules for where herestrings are accepted vs. other redirects, whereas the actual shell grammar doesn't make that distinction (other than heredocs, which are of course super special).

[danfuzz]

My branch https://github.com/danfuzz/tree-sitter-bash/tree/danfuzz-redirects has a proposed fix, though (as I write this) it doesn't produce identical output for the Herestrings test. So, perhaps it's more of a "suggested direction" at this point.

Clarification (after looking more closely): I think my proposed change improves the parse output in most cases of the test, specifically wrapping herestring-redirected statements in (redirected_statement ...) nodes. However, it does fail to parse a case where redirections and arguments are mixed, e.g. (simplified) x <<<x x >x.

[danfuzz]

Update: I think the failure I noted above highlights a pre-existing bug in the grammar. In particular, it looks like (non-redirection) arguments after a file redirection are taken to be redirections. For example:

x <x a b c

Parse tree:

(program [0, 0] - [1, 0]
  (redirected_statement [0, 0] - [0, 10]
    body: (command [0, 0] - [0, 1]
      name: (command_name [0, 0] - [0, 1]
        (word [0, 0] - [0, 1])))
    redirect: (file_redirect [0, 2] - [0, 10]
      destination: (word [0, 3] - [0, 4])
      destination: (word [0, 5] - [0, 6])
      destination: (word [0, 7] - [0, 8])
      destination: (word [0, 9] - [0, 10]))))

[amaanq]

I've fixed the x <x a b c case but the first case in your issue is pretty difficult, throwing herestring_redirect under the 'redirect' field in the repeat1 body fixes it, but makes some parsing output in the tests worse, notably the same case I've fixed. But maybe that's desired, it'd be like this: