GNU bug report logs - #61368
[PATCH] Extend go-ts-mode with support for pre-filling return statements

Previous Next

Full log

Message #11 received at submit <at> debbugs.gnu.org (full text, mbox):

From: Theodor Thornhill <theo <at> thornhill.no>
To: Evgeni Kolev <evgenysw <at> gmail.com>, bug-gnu-emacs <at> gnu.org
Cc: Randy Taylor <dev <at> rjt.dev>
Subject: Re: [PATCH] Extend go-ts-mode with support for pre-filling return
 statements
Date: Wed, 08 Feb 2023 20:20:27 +0100

Evgeni Kolev <evgenysw <at> gmail.com> writes:

> [CC Randy, author of go-ts-mode, Theo, who asked to keep the patches coming :)]
>

:-)

> This is a patch which adds support to go-ts-mode to insert
> context-aware return statements for the current function. The return
> statements are pre-filled with the Go zero values of the output
> parameters.
>
> For example, all these return statements are inserted by M-x
> go-ts-mode-insert-return:
>
> ```
> func f() (x, y, z int) {
>     return 0, 0, 0
> }
>
> func exotic() (bool, int, myStruct, *int, chan bool, error) {
>     return false, 0, myStruct{}, nil, nil, err
> }
>
> func closure(numbers []int) {
>     sort.Slice(numbers, func(i, j int) bool {
>         return false
>     })
> }
> ```

Cool!

>
> The command go-ts-mode-insert-return is experimentally bound to a key
> C-c C-r ("r" as return statement). It's a user error to run C-c C-r
> outside of a function body.
>
> Customization: when the output params contain "error" the pre-filled
> text is "err". A customization variable is added to allow the user to
> pick a different value, for example they might pick `fmt.Errorf("...:
> %w", err)` which will result in error wrapping:
> ```
> func f() (int, error) {
>     return 0, fmt.Errorf("...: %w", err)
> }
> ```
>
> Personally, I'll use yasnippet instead of C-c C-r. Something like this:
> ```
>  # -*- mode: snippet -*-
>  # name: ret
>  # key: ret
>  # type: command
>  # --
>
>  (let ((go-ts-mode-error-zero-value "${1:fmt.Errorf(\"${2:format}:
> %w\", err)}"))
>    (yas-expand-snippet (go-ts-mode-return)))
> ```
>
> I'm open to suggestions about how to best expose this functionality to
> the user. I think a snippet makes the most sense, but there's no
> standard way for major modes to expose snippets as far as I'm aware.
> It's possible to tweak C-c C-r to call (yas-expand-snippet) if
> available, otherwise call (insert). In general, I don't feel strong
> about the C-c C-r key binding, but I didn't have a better idea.
>

How about using tempo or skeleton as fallbacks when yasnippet isn't
installed? 

> Known bug: this example does not work, fails with error "Unknown Go
> type "[int]int"":
> ```
> func notOk() (int, map[int]int) {
> }
> ```
> The root cause is an issue with the tree-sitter-go parser. The bug has
> been reported here
> https://github.com/tree-sitter/tree-sitter-go/issues/107
>
> Known limitation: unfamiliar types are assumed to be structs. This
> assumption does not work for aliased types. For example:
> ```
> type MyInt = int
>
> func alias() MyInt {
>     return MyInt{}
> }
> ```
> Above the correct return statement is "return MyInt(0)". My assumption
> is that type aliases of primitive types are rare in Go codebases. In
> these rare cases, the user is expected to replace "MyInt{}" with
> "MyInt(0)", or not use C-c C-r at all.
>
> Inspired by: this emacs conf talk "08:21 Intelligent templates"
> https://emacsconf.org/2022/talks/treesitter/
>
> Feedback is more than welcome! In particular:
> - how to best expose the functionality - key binding, snippet (how
> exactly?), something else?
>

Personally I think this should be a contrib to gopls as a code action so
that others can benefit from it.  Is upstreaming it to gopls too hard?


Theo

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.