How should spellchecking look in Helix?

[the-mikedavis]

This is intended to be a more open-ended offshoot of #11660.

Spellbook works and it's pretty quick. I've been running a very hacky change in my daily-driver branch for a while now. That patch spots typos in the code within the viewport at rendering time and draws squiggly lines under them. Not very fancy but it's obvious when there's an error.

What I'm wondering is how the "proper" version of the feature should look. Mainly I'm concerned that it will be obtrusive and annoying since code is often full of little abbreviations. When should spellchecking occur, what text should be checked and how should mistakes be presented? Other parts of the feature seem straightforward: if the cursor is over a spelling mistake, C-a (code action menu) should make suggestions which replace the typo when accepted.

In this discussion I'd like to document other editor's behaviors and field opinions.

[the-mikedavis]

The most direct inspiration will probably come from a language server and plugin for the Zed editor Codebook which uses Spellbook and Tree-sitter for spell checking. This should give good insight into how Tree-sitter queries should look and what needs to be configurable.

#3637 has a lot of prior discussion about using Tree-sitter to tag the ranges of text which should be checked. We shouldn't ever check keywords for example but we should maybe check function names. I know that Neovim mixes in captures for spellchecking with their highlight queries and typically only tags comments. Codebook has queries which tag different nodes with captures that correspond to different parsing schemes for running the spellchecker, like @string, @comment and @identifier.

[the-mikedavis]

I update a spellbook branch every now and then with integration commits. There's a fair amount of work in the core/view parts of the editor to enable diagnostics and code actions to originate from the editor rather than language servers - similar to the refactoring work in #2507.

What we will most likely need implementation-wise is the ability to create diagnostics and code actions belonging to the editor. Code actions are simpler: they can move to helix-view and be essentially Box<dyn Fn(&mut Editor)> with titles. Diagnostics take a larger refactor similar to completions.

[achenet]

There's a fair amount of work in the core/view parts of the editor to enable diagnostics and code actions to originate from the editor rather than language servers - similar to the refactoring work in #2507.

sorry for the noob question, but why not implement the spell check as a language server and just add it select file types - maybe txt and md by default, and then users can add it to various programming language files if they choose?

[the-mikedavis]

Also see #3637. There are already language servers that can be used for this. Language servers can be annoying to install and run though, and cause some duplication like holding another copy of the document in memory and another parse tree if they use tree-sitter. The goal with a core feature is that you can use spell checking out of the box without the extra setup for getting a language server running.

To that end we could also consider curating a list of dictionaries that can be downloaded, like Codebook does.

[the-mikedavis]

Kakoune has a spell.kak plugin shipped by default which shells out to aspell: https://github.com/mawww/kakoune/blob/master/rc/tools/spell.kak. I was never very good at Kakoune plugins and I don't quite understand if/how it is checking only some ranges of each buffer. Kakoune's plugin prefers typable commands like :spell to start spell-checking and :spell-clear to stop spell-checking and clear diagnostics. The spelling diagnostics are navigated separately from everything else with :spell-next/:spell-prev. It also uses a :spell-add typable command to add the current selection to the dictionary. (I'd like to use a code-action like menu and maybe prompt instead though.)

[zqianem]

I do like the idea of manually turning on and off a spell-checking mode, perhaps as an option in Helix's space menu by default? Ideally, I prefer writing my code or text in one go and then enabling/addressing diagnostics just before committing¹.

Vim has a similar concept with its :set spell and :set nospell commands, which I have been missing since I switched to Helix a few years back.

Footnotes

1. For that reason, it would be nice to have a way to toggle diagnostics without having to fully start and stop the language servers (not currently possible?), but that's a separate topic. ↩

[gyreas]

I think we can combine them both: in spell-check mode, 'n' (maybe bound to 'spellcheck_next') functions like normal but it goes to next questionable character sequence or word, then on it's selected in that fashion, we get a code action for relevant suggestions.

(EDIT: then there's no need for prompt since you have the offending selection(s) ready, you just edit it as need. nhe spellchecker will have to retokenize the buffer to pick up this new change tho)

To see all questionable words, we do something similar to 'workspace_diagnostics_picker', so you can preview these in context along with others.

What do you think?

[MiledMFN]

Hello. Spellchecker is the only thing that I miss from Helix. I use Codebook with Typst (that has no support for it), so the diagnostics target code, I also write several languages in a file. Typst has the feature of changing language across the source file, and the Typst App has spellchecker that actions in function of the language.

I think that asking for something like the Typst App is too much. What I want to suggest is something similar to Emacs' spellchecker: in addition to toggle spellchecker with a command, having the option of select some text and toggle it (obviously with the option of changing the dictionary) so that only that text gets analyzed. And I think that this behaviour matchs with the Helix's editing philosophy and style. In this regard, a multi language source file will be properly worked on it.

[wcampbell0x2a]

I've been using harper-ls.

The most annoying part is you need to add this as a language LSP for every language you want spelling enabled.

[zaggernut]

Do you have any solution for the lack of debounce, making harper trigger 20 times every time you write a line? I have tried setting editor.idle-timeout and hell even editor.completion-timeout to multiple seconds and it didn't do anything