Use of the new `Fallback` interface

[brogdonm]

I have a scenario where I am loading a FontSystem with exactly one font and I don't want any fallback lookups. In other words, I want to gaurantee that the font is used when rendering. I then implemented my own NoFallback structure, which implements the new Fallback trait and every function returns just a &[], as I don't want it to fallback. When doing this, I would have expected all of the logs about "Failed to find preset fallback" would have been skipped over since I didn't provide any fallbacks.

Am I going about this the wrong way? I know I can go in and configure the logs to filter the messages, but I just want to use the most optimum way with loading this font and using it.

[tigregalis]

#369 The implementation of Fallback was pretty simplistic: What used to be a set of static platform-specific fallback lists, is now provided via a trait instead. Any other existing behaviour was exactly preserved:

    pub fn check_missing(&mut self, word: &str) {
        if self.end {
            missing_warn!(
                "Failed to find any fallback for {:?} locale '{}': '{}'",
                self.scripts,
                self.font_system.locale(),
                word
            );
        } else if self.other_i > 0 {
            missing_warn!(
                "Failed to find preset fallback for {:?} locale '{}', used '{}': '{}'",
                self.scripts,
                self.font_system.locale(),
                self.face_name(self.font_match_keys[self.other_i - 1].id),
                word
            );
        } else if !self.scripts.is_empty() && self.common_i > 0 {
            let family = self.font_system.fallbacks.common_fallback()[self.common_i - 1];
            missing_warn!(
                "Failed to find script fallback for {:?} locale '{}', used '{}': '{}'",
                self.scripts,
                self.font_system.locale(),
                family,
                word
            );
        }
    }

This function is called during shape_run:

    // Debug missing font fallbacks
    font_iter.check_missing(&line[start_run..end_run]);

[brogdonm]

Hey @tigregalis I understand this implementation, thanks for pointing it out. I have since just filtered the logs for this crate.

But I started wondering, could the Fallback trait really change to a Option<> instead and then an implementor could state None, effectively disabling fallback all-together?

[tigregalis]

I'd probably have the Option<Fallback> on FontSystem and in the new_with_locale_and_db_and_fallback function, and check before constructing the font fallback iterator, but I'm not yet sure how to untangle the fallback logic from the rest of the work. It gets constructed on both Basic and Advanced shaping.

[tigregalis]

@brogdonm I've looked into it and maybe you want Shaping::Basic at the Buffer level. The FontFallbackIter is always constructed, but by default it takes the font family from the Attrs, when Shaping::Basic, it uses the shape_skip path which only calls FontFallbackIter::next() exactly once, effectively just taking the font family from the Attrs.

EDIT: Actually, I just tested and Shaping::Basic disables ligatures too, so it's probably not what you want.

It may be worthwhile then to separate font fallback ability from ligature ability, but these are currently tangled up as part of shaping.

[brogdonm]

@brogdonm I've looked into it and maybe you want Shaping::Basic at the Buffer level. The FontFallbackIter is always constructed, but by default it takes the font family from the Attrs, when Shaping::Basic, it uses the shape_skip path which only calls FontFallbackIter::next() exactly once, effectively just taking the font family from the Attrs.

EDIT: Actually, I just tested and Shaping::Basic disables ligatures too, so it's probably not what you want.

It may be worthwhile then to separate font fallback ability from ligature ability, but these are currently tangled up as part of shaping.

Thanks @tigregalis for looking into this.