Re: Better c-ts-mode indentation when there are macros in the way

On 7/26/26 7:32 AM, [email protected] wrote:
> On 7/25/26 5:59 AM, Yuan Fu <[email protected]> wrote:
> > I read through it and I like it! Especially since we’re limiting 
> > ourselves to just indentation and further just the parent-is rule, we 
> > only need to handle the case where the directive is the first line in 
> > the if body, and that makes the heuristic very simple. Very clever! 
> > I’ll incorporate this into c-ts-mode with some changes.
> >
> > BTW, since you’ve been using this for a while, do you see any edge 
> > cases that this doesn’t handle? You mentioned that it’s not as brittle 
> > as you imagined, but did you encounter any concrete cases where it 
> > didn’t work?
> >
> > Yuan
> >
> 
> Thanks for having mercy on my code! :)
> 
> Maybe adding new matchers instead of changing how parent-is, n-p-gp,
> and friends behave is safer since it could break existing setups?
> Ideally, I'd like a "matcher" that says:
> 
>     (look-upwards-til-anchor anchor-bol c-ts-mode-indent-offset)
> 
> And the "anchor" is any node that modifies indentation: if_statement,
> compound_statement, argument_list, etc., but not preproc_* or
> expression_* nodes. Like how a C programmer actually reasons about
> indentation.
> 
> Brittle is relative. :) Walking the tree upwards is easy with
> tree-sitter and does not cause lag. Surprisingly robust. But
> tree-sitter-c often fails to parse preprocessor heavy code:
> 
>     printf(
> #ifndef BRUH
>      "tttt\n"
> #endif
>     );
> 
> or
> 
>     if (blah)
>         return;
>     else
> #define EEH
>      printf("wwah");
> 
> It is unfortunate because such code is not extremely rare. The grammar
> needs an overhaul to better incorporate macros and to make it easier
> for code editing. But that doesn't seem to be in the cards.
> 
> Here is my new proof-of-concept:
> 
> https://gist.github.com/bjourne/63ffea0c420bf84104c3f98ca1623ea6
> 
> I have this matcher as the last rule:
> 
>     (my/c-parent-is-container my/c-parent-bol c-ts-mode-indent-offset)
> 
> 
> -- 
> mvh/best regards Björn Lindqvist

Hello again!

Here is a complete c-ts-mode indentation setup, in case anyone want to
test it:

https://gist.github.com/bjourne/31f53043c72728cb4d8b90feedd0625b

In addition to correcting some tree-sitter-c preprocessor macro
misparses, it also handles cases where you need to arbitrarily
walk up the tree. For example, in

    return c1 || c2 ||
	c3 || c4 || c5;

I want c3 to "dangle" because it continues the return statement. But in

    return (c1 || c2 ||
	    c3 || c4 || c5);

I want c3 to line up because the opening paranthesis isn't
alone. Since binary_expression nodes (and similar node types) can be
arbitrarily nested, you need to walk past them when looking for the
ancestor you should anchor identation to.


--
mvh/best regards Björn Lindqvist