bug#60484: 29.0.60; c-ts-mode: short tokens are not identified as type_identifier

[Yuan Fu]

Eli Zaretskii <eliz@gnu.org> writes:

>> Date: Mon, 02 Jan 2023 18:13:34 +0530
>> From: Mohammed Sadiq <sadiq@sadiqpk.org>
>> Cc: 60484@debbugs.gnu.org
>> 
>> On 2023-01-02 17:45, Eli Zaretskii wrote:
>> >> Date: Mon, 02 Jan 2023 10:22:09 +0530
>> >> From: Mohammed Sadiq <sadiq@sadiqpk.org>
>> >> 
>> >> Short tokens are not identified as type_identifier in GNU Emacs
>> >> c-ts-mode, but does work fine with tree-sitter playground[0].
>> >> 
>> >> Say for example, 'a_type' in an empty buffer is identified as a
>> >> type_identifier in tree-sitter playground, but not in c-ts-mode,
>> >> while say, some longer tokens like 'window_type' is identified as
>> >> type_identifier.
>> > 
>> > Where is it written that FOO_type is a type identifier?  is this
>> > something new in some recent C Standard?  Or is it just a popular
>> > convention?
>> 
>> 'a_type' was just a made up example, it can be any valid token, say
>> 'g_file', or whatever.  I was pointing out a disparity in handling of
>> some token in c-ts-mode and tree-sitter: tree-sitter identifiers a 6
>> byte length token as an identifier, but c-ts-mode requires it to be
>> at least 11 byte sized for custom types.
>
> I'm not sure I see a problem here.  It sounds like different
> heuristics to me.  Nothing says that g_file is a type, only its
> parsing can tell.

The parse tree of a buffer with only a_type in it is this:

(translation_unit (ERROR (identifier)))

So tree-sitter-c parses it as a parse error instead of a type. I suppose
the difference is due to different version of tree-sitter-c used by
Emacs (the latest) and the tree-sitter playground website? Maybe the
playground is using an older version. The "cutoff" point for the
playground version seems to be 5 bytes: a_typ is considered an error but
a_type a type.

Yuan