ignoring rules per workspace in nvim...?

[cprn]

I installed harper_ls in Neovim via Mason. It works, but I don't understand how I'm supposed to control it, as docs don't explain it well.

I have a file notes.md containing e.g.:

TODO: something in other.md
TODO: another thing in whatever.md

I move my cursor over the line and diagnostic says:

Diagnostics:
1. Did you mean to spell `md` this way? [SpellCheck]
2. Hyphenate `to-do`. [ToDoHyphen]

I activate Code Actions (CAs) and see:

╭─────────────────────── Code actions ───────────────────────╮
│>                                                      9 / 9│
├────────────────────────────────────────────────────────────┤
│> 1: Replace with: “TO-DO”                   harper_ls      │
│  2: Ignore Harper error.                    harper_ls      │
│  3: Replace with: “Too”                     harper_ls      │
│  4: Replace with: “Todo”                    harper_ls      │
│  5: Replace with: “To-do”                   harper_ls      │
│  6: Ignore Harper error.                    harper_ls      │
│  7: Add "TODO" to the user dictionary.      harper_ls      │
│  8: Add "TODO" to the workspace dictionary. harper_ls      │
│  9: Add "TODO" to the file dictionary.      harper_ls      │
╰────────────────────────────────────────────────────────────╯

Selecting 8 (add TODO to the workspace dict) doesn't solve the issue. There's no CA to ignore rule in this workspace. It's the project's convention, so how do I handle that?

Furthermore when I add md to the workspace dictionary, diagnostics change to:

Diagnostics:
1. This word's canonical spelling is all-caps. [OrthographicConsistency]
2. Hyphenate `to-do`. [ToDoHyphen]

I come from ltex_plus where there were no such issues.

It also shows things like:

Diagnostics:
1. 'utilised': Possible spelling mistake. 'utilised' is British English. [MORFOLOGIK_RULE_EN_US]

despite:

vim.lsp.config('harper_ls', {
    filetypes = {'markdown', 'text', 'tex', 'typst'},
    settings = {
        ['harper-ls'] = {
            dialect = 'British',
        },
    },
})

(which is copied from documentation).

[hippietrail]

This has come up before and it's still never been tackled: #997 #1057 #1430 #3076 are all related

Basically, the user dictionary only comes into play for the spellcheck linter.

But a word can trigger multiple linters.

But when a word triggers multiple linters only one linter gets to suggest corrections.

But there are one or two specific linting rules added way back early in Harper's lift to address TODO vs Todo vs to-do vs todo.

There are different opinions on which variants are preferred and linters can only be turned off or on, they can't have settings.

There's also the problem that some variants are preferred in code comments, such as TODO, but in English prose the preferred form is "to-do". I'm not sure but I don't think the linters can know that kind of context.

It's quite a mess of overlapping tiny problems that shows up rather prominently.

[cprn]

That is very unfortunate.... Thank you, I'll follow the tickets.

[cprn]

BTW I'm pretty sure editors like Neovim pass a context (semantic token), e.g. right now Neovim sends a didOpen event when opening a buffer and it contains file type (e.g. markdown, plaintext, or code language, like lua or python), and on every request informs the LangServ provider whether a given piece of text is a comment, a function, and so on (as detected by AST generating add-ons, e.g. Treesitter). It's part of the LSP protocol. So, at least I think that's true, such different rules for different semantic tokens could be applied.

[hippietrail]

BTW I'm pretty sure editors like Neovim pass a context (semantic token), e.g. right now Neovim sends a didOpen event when opening a buffer and it contains file type (e.g. markdown, plaintext, or code language, like lua or python), and on every request informs the LangServ provider whether a given piece of text is a comment, a function, and so on (as detected by AST generating add-ons, e.g. Treesitter). It's part of the LSP protocol. So, at least I think that's true, such different rules for different semantic tokens could be applied.

That would be very handy. I'm not very familiar with the LSP stuff but on the linter side I'm not aware of anything getting passed. I think this needs some feedback from @elijah-potter or somebody interested to dig deep into the code...

[cprn]

Closing as the issue is clear now, but feel free to reopen for discussion or whatever. I can help with testing nvim if it's ever needed.