bug#81624: New command: xref-find-by-kind

On 18/08/2026 14:06, Sean Whitton wrote:

> It's difficult to think about this in the absence of more detailed user
> documentation.  Could you write up the manual entries, perhaps including
> some examples of what sort of kinds might be available for a language?

I've added a manual entry in commit ceb0173179b, hope it helps. It 
rehashes what is already written in docstrings (not sure what to add, 
please ask), but maybe a summary like that would give a new impression.

> The main entry point user command's docstring should say what is prompts
> for.  Presumably you have to be asked for the kind and then the symbol.
> Which one comes first?

In the order they're mentioned: first the kind, then the identifier.

> Are they both minibuffer prompts or do you get
> an *xref* buffer at some point?
Both of them are minibuffer prompts, you get an *xref* buffer at the 
end. The first prompt (mandatory) uses 'read-multiple-choice', asking 
the user to type the character that corresponds to the desired kind. 
Which makes it function like a dynamic keymap (possibly avoiding the 
need for separate commands, discussed previously).

Then, if there is no identifier at point or if called with C-u, it reads 
the identifier, with completion. In Elisp mode, since today, it also 
filters the identifiers to only show matching the specified kind.

So for example you can type "C-u M-' c" and see the defined faces as 
completions. This is opt-in: a bit harder to do the same in Eglot.