hb_set_intersect appears to be broken in 1.6.2

[jfkthame]

I'm seeing unexpected results from hb_set_intersect (often, but not invariably, returning an empty result, when the two sets did have some intersection).

This is breaking behavior in Firefox, where it is wanting to determine whether a given feature supports all the glyphs in a text run.

E.g.

inputGlyphs: 886 887 888 889 890 891 892 893 894 895 1024 1152
glyphsInRun: 889 1024
intersection: (empty)

But:

inputGlyphs: 886 887 888 889 890 891 892 893 894 895 1024 1152
glyphsInRun: 887 888 889 1121
intersection: 887 888 889 1121

For a given pair of sets, the result is consistent, but it's not obvious why the one case fails while the other succeeds.

[jfkthame]

(See https://bugzilla.mozilla.org/show_bug.cgi?id=1411005.)

[jfkthame]

For a given pair of sets, the result is consistent, but it's not obvious why the one case fails while the other succeeds.

On second glance, neither result there looks correct, actually: in the second example, the "intersection" includes a GID that was not present in the inputGlyphs set. Oops!

[behdad]

That's why I hated implementing my own set... I'll take a look soon.

[behdad]

Separately, I'm interested to learn more about how you use this stuff in Firefox. I know about the word cache usage, but this sounds like something different.

[behdad]

Humm. I added a test for the samples you provide, but it's passing.

[behdad]

@jfkthame any chance you can debug this a bit?

[jfkthame]

See PR#582 for a test that fails (at least for me). Haven't tried to debug yet.

[jfkthame]

With the second commit in #582, the added test now passes for me.

[jfkthame]

Separately, I'm interested to learn more about how you use this stuff in Firefox. I know about the word cache usage, but this sounds like something different.

We use this when determining whether a feature such as superscript covers all the glyphs that are present in an element styled with font-variant-position: super. If so, we can render the font-variant style using the OpenType feature, but if not, we'll use a synthetic fallback for the whole run (because a mix of "real" superscripts and fake ones within a single element could look pretty bad).

[behdad]

We use this when determining whether a feature such as superscript covers all the glyphs that are present in an element styled with font-variant-position: super. If so, we can render the font-variant style using the OpenType feature, but if not, we'll use a synthetic fallback for the whole run (because a mix of "real" superscripts and fake ones within a single element could look pretty bad).

Cool. Thanks. FYI @drott