Add support for old-style function definitions

[dehorsley]

I'm looking to refactor some very old code and it would be useful to parse old-style (K&R) function definitions. For example:

int main(argc, argv)
int argc;
const char* argv[];
{
return 0;
}

(I see even the Github syntax highlighting doesn't handle it)

Adding repeat($.declaration) in the function definition seems natural.

Unfortunately, the attributes cause a little problem. For example in

void foo(bar) __attribute__((noreturn)) __attribute__((format(printf,1,2)))
int bar;
{
}

it isn't clear if the __attributes__ are attached to the function of the variable (or one each way!). This causes the generator to complain about a conflict.

By changing the precedence and adding a conflict rule, I was at least able to get sensible parses in most cases, but I'm not entirely happy with it. Perhaps someone more familiar with tree-sitter parsing can take a look.

I suspect the number of C files using both K&R defs and attributes is vanishingly small, but it would be nice to be complete.

For reference Clang allows the above and attaches the attributes to the function, but throws an error with:

void foo(bar) __attribute__((noreturn))
 __attribute__((format(printf,1,2))) int bar;
{
}

(Just a single newline! Whoever said C was whitespace independent!)

I don't have GCC on hand to check this right now.

[XVilka]

As C23 deprecated such syntax, I wonder if it worth the effort to support now. I understand that it will take quite some time for the old code to migrate, but these days it is usually happens faster than before, especially for projects who active and on GitHub.

[amaanq]

Added now