hunspell_stem error: basic_string::_M_construct null not valid

[zeuner]

I've been facing an issue where hunspell_stem will crash with a C++ string constructor error:

basic_string::_M_construct null not valid

After that, the stemmed output is unusable. When using it in a tokenizer, I just get the words from the error message as tokens.

I have yet to figure out a minimum failing example, which will require me to first investigate how exactly hunspell_stem was called. This is because it is currently being called from CreateDtm (textmineR package) on large text corpora. It happened to me when trying to create a document term matrix with stemming from an Italian-language text corpus, using code which already worked reliably on different text corpora in other languages. If you have any suggestions for tracking down the issue, I'd be happy to hear about them.

For now, I solved it using a small change to your library which you can find at https://github.com/zeuner/hunspell/tree/string_error . I'm checking for the NULL pointer that causes the invalid constructor call and return early. I didn't open a pull request because I don't know the library well enough to decide whether this is the correct way to solve the issue, but it works for me. If you think it is, feel free to pull it in. If you want to do so, you might want to check whether it's permissible to just return a default-constructed Rcpp::CharacterVector. This might be the case because R_hunspell_stem seems to even allow out[i] not to be set at all.

[jeroen]

Can you please include example code so that I can reproduce the crash?

[zeuner]

I was able to reduce the offending stemming operation to this fairly minimal example code that crashes for me:

library(hunspell)

stem <- function (x) {
    hunspell_stem(x, "it_IT")
}

stem(c("altresı"))

I'm not sure whether the stemmed word is correct Italian. But I think hunspell should at least avoid crashing with a cryptic error message even if it encounters low-quality input.

[jeroen]

Thanks. I don't have an italian dictionary myself, can you please tell which OS you run and where you got the dictionary, so that I can test it?

[zeuner]

I encountered the error on both Debian Buster and Ubuntu Focal Fossa, using the hunspell-it dictionary packages provided by the OS. In particular, this was hunspell-it 6.4.3 ( https://packages.ubuntu.com/focal/hunspell-it ) and 6.2.0 ( https://packages.debian.org/buster/hunspell-it ).