Customize the hyphenation bounds?

[baskerville]

The hyphenation bounds are currently hardcoded. Would it be possible to load the bounds from a file instead? For example, a file named en-us.bnd.txt with content 2 3 would define the bounds for the EnglishUS language.

[tapeinosyne]

Although default values are indeed hard-coded, the hyphenation boundaries are a public field of the dictionary type and can be changed freely. I will add a note in the documentation to this effect.

[baskerville]

It doesn't seem to be working:

    let word = "hyphenation";
    let mut en_us = Standard::from_path(Language::EnglishUS,
                                        "en-us.standard.bincode").unwrap();
    println!("{:?}", en_us.opportunities(word));
    en_us.minima.0 = 3;
    println!("{:?}", en_us.opportunities(word));

The output I'm getting is:

[2, 6, 7]
[2, 6, 7]

[baskerville]

I can get close to the expected result ([6, 7]), with:

en_us.opportunities_within(word, en_us.boundaries(word).unwrap())

which returns [6].

[baskerville]

In fact it seems to occur only when a word has a known hyphenation:

    fn opportunities(&'h self, lowercase_word : &str) -> Vec<Self::Opportunity> {
        match self.boundaries(lowercase_word) {
            None => vec![],
            Some(mins) => {
                match self.exact(lowercase_word) {
                    None => self.opportunities_within(lowercase_word, mins),
                    Some(known) => known
                }
            }
        }
    }

You can't just return known here, you have to take mins into account.

[baskerville]

The reason opportunities_within returns [6] instead of [6, 7], in the above example, is because it doesn't rely on self.exact.

[tapeinosyne]

I have a vague recollection of deliberately skipping boundary checks for exceptions, but you are right: the current behavior is counter-intuitive at best and plainly broken at worst. I'll see if I can release an updated version over the weekend.

[baskerville]

the current behavior is counter-intuitive at best and plainly broken at worst

I concur.

[tapeinosyne]

Fixed in b5f526f.

Exceptions now respect hyphenation boundaries. The old behavior can be replicated by manually calling the internal method exception_within with (0, word.len()) as custom boundaries.