hunspell stumbles over copyright symbol

[drahnr]

Describe the bug

Encountered error Utf8Error { valid_up_to: 2, error_len: Some(1) } returned from Hunspell_suggest(handle, ["\"\\xc2\\xa9\""]): 0: "\xc2\x80\x93"
[2022-09-16T10:03:50Z DEBUG hunspell] © --{suggest}--> []

To Reproduce

Steps to reproduce the behaviour:

1. A file containing ©
2. Run cargo spellcheck file.rs
3. ...

Expected behavior

Handle or ignore, currently hunspell-rs is hacked to print an error.

Screenshots

Please complete the following information:

• System: Fedora
• Obtained: cargo + git
• Version: 0.12.2 / git

[drahnr]

CC @lopopolo that was the issue at hand, the suggestion should accept 0xC2 0xA9 as valid since it is valid itself, but returns garbage suggestions instead. It's still present in 0.12.2 but will only be a verbose message that will be handled in the next release.

[lopopolo]

Oh awesome. That's in the generated headers of the Unicode files. Great catch @drahnr and thanks for debugging!

[drahnr]

@lopopolo I realized you have custom dict, with a - - removing the single char items from the list resolves the issue. It seems that trips the parser and makes it's way into the lut inside hunspell and then surfaces with some byte sequences.

[lopopolo]

I just tried this workaround and that seemed to work! I get no warning messages in cargo-spellcheck 0.12.1.

[drahnr]

The core issue is due to the fact that Hunspell uses the encoding used in the affix file, for the dictionaries as well. For en_us.aff (both builtin and Fedora 36) this was latin-1 encoding rather than utf-8. Hunspell then treats all inputs of encoding equiv to the affix file, implicitly by the used prefix tree.

Solution would be to use i.e. encoding_rs to re-encode the dictionaries to UTF-8 and only afterwards feed them to Hunspell or reject all encodings set besides utf-8.