Automattic/harper

My PR #4211 passes all cargo test for rust but keeps failing in the Obsidian integration. I'm not doing to well trying to understand the error messages so I'm not positive but I'm tending toward thinking that when the OneOfMany config type was introduced that support for it might not have been added as needed to the Obsidian plugin. Both CoPilot here on GitHub and the free AI agent in my editor (Devin) seem not to have been able to figure out why there are failures in the Obsidian integration. Because the Obsidian plugin uses harper.js the problem could be there or in the WASM?

I was planning to add more linters that handle verbs with multiple past participles like "hanged" vs "hung", "dived" vs "dove", "pleaded" vs "pled", and started looking at how I did the "sneaked" vs "snuck" one. This led me to discover that both "sneaked" and "snuck" are now flagged: Digging further in #4209 I discovered that this is due to the change in linter configuration from lint_group/mod.rs to default_configuration.json and the fact that there is a OneOfMany type of linter preference that was introduced. I've looked manually, Googled in both site:github.com/Automattic/harper and site:writewithharper.com, and asked the built-in coding agent in the Devin fork of VS Code if it can find documentation for OneOfMany and we can't find any. We could still be missing something but it seems to be undocumented and so far not used for any linter. I have not dug into whether anything needs to be documented downstream for harper.js or any of the integrations.

This is even more important than it already was due to the rise of AI-generated rules, which are still terrible at including real-world examples rather than contrived examples made on assumptions that are too short-sighted to avoid false positives. (Humans also do this, but humans that become Harper contributors can learn how to overcome this. So far AIs can't.) An ideal tool would use existing corpora and XML/database dumps, feeds, and APIs of coding and tech-related sites. I believe Stack Overflow still publishes database dumps, or at least that older ones are still available. Neither Slashdot nor Hacker News publish dumps but that do seem to have real-time feeds. For GitHub we might be able to leverage their search API. Such a tool would probably not belong in the Harper repo, but as one or more separate projects. But ideally may be installed on a server that authorized Harper devs/contributors would be able to access for reasons of storage space, API limits, etc.

We do have a linter that flags "that" when it should be "than" but I just spotted this exact typo by me right here in a comment and it's not being flagged. way less common that mp3

I'm not a browser extension expert but I've stumbled through making a couple by following howtos or guidance of an AI. But getting harper.js to work inside one seems to be next level complicated. I'm only using Google Search AI because I don't have tokens for a coding AI but every time it says we're done there's something wrong that's outside my areas of expertise. Maybe it's obvious to a webdev or someone who's made lots of Chrome Extensions. I was going to make a little tool similar to harper-glasses but instead of linting the selected text, called the tokenization API and annotates the tokens in the selection with POS tags. But a tiny example either as part of the Harper docs that just opened a popup/modal and displays the harper-core version it's built with would be nice and minimal perhaps. I also find the harper.js API Reference very opaque. It seems to list only objects and not functions. For instance how to I find the function to get the harper-core version? I'm sure it exists.