Use OT rather than FT functions to query metrics from Harfbuzz

[alerque]

I don't know if this is a font version problem or a SILE problem, but tests/vertical.sil is being very inconsistent. Every system seems to have a different idea of how the metrics should go.

@simoncozens's system has produced one thing, yet my system is consistently producing another and the Travis test environment is producing a third set of values for the same test.

Comparing for example just the first line of the diff when tested on my system:

-My     117.74384
+My     117.74191

While Travis has this diff:

-My     117.74384
+My     117.73966

For reference on my system there are a whole bunch of weights for this font:

❯ fc-list | grep 'Sans CJK JP'
/usr/share/fonts/OTF/NotoSansCJKjp-Regular.otf: Noto Sans CJK JP,Noto Sans CJK JP Regular:style=Regular
/usr/share/fonts/OTF/NotoSansCJKjp-Light.otf: Noto Sans CJK JP,Noto Sans CJK JP Light:style=Light,Regular
/usr/share/fonts/OTF/NotoSansCJKjp-Thin.otf: Noto Sans CJK JP,Noto Sans CJK JP Thin:style=Thin,Regular
/usr/share/fonts/OTF/NotoSansCJKjp-Black.otf: Noto Sans CJK JP,Noto Sans CJK JP Black:style=Black,Regular
/usr/share/fonts/OTF/NotoSansCJKjp-Bold.otf: Noto Sans CJK JP,Noto Sans CJK JP Bold:style=Bold,Regular
/usr/share/fonts/OTF/NotoSansCJKjp-Medium.otf: Noto Sans CJK JP,Noto Sans CJK JP Medium:style=Medium,Regular
/usr/share/fonts/OTF/NotoSansCJKjp-DemiLight.otf: Noto Sans CJK JP,Noto Sans CJK JP DemiLight:style=DemiLight,Regular

And the default selection is actually one of the lighter weights:

❯ fc-match 'Noto Sans CJK JP'
NotoSansCJKjp-DemiLight.otf: "Noto Sans CJK JP" "DemiLight"

It's possible that the issue here is font weigt selection. I played with setting \font[weight=x] in Sile but none of the weights matched up with the metrics currently in the master branch.

[simoncozens]

And the default selection is actually one of the lighter weights:

❯ fc-match 'Noto Sans CJK JP'
NotoSansCJKjp-DemiLight.otf: "Noto Sans CJK JP" "DemiLight"

If only things were that simple:

$ fc-match 'Noto Sans CJK JP'
NotoSansCJKjp-Regular.otf: "Noto Sans CJK JP" "Regular"

[alerque]

Do you have the same set of available weights I do?

[simoncozens]

Ah, actually I just have regular installed. However, if SILE was selecting a different weight for you, then the "Set font" line in the test output would be different. I think. You could try manually forcing the weight to 200 and seeing what you get.

[alerque]

The "Set font" line in the output doesn't show a different font face no matter what weight I set (the weight value shows up, but the actual face picked isn't noted there). 200 seems to be the default and no, setting it to 200 manually doesn't change my output. Setting it to something more or less does that that, but none of the values I tried netted me the same metrics as the test.

[simoncozens]

OK, so I was using Noto Sans regular 1.003 whereas Travis was using Noto Sans Light 1.004. Now I am using the same version as Travis. This was discovered using our new handy font fingerprinting feature.

I also want to add support for groking the font OpenType tables and pulling out metadata. (And, in the future, other things...)

[simoncozens]

...but that hasn't helped. Now we have the same Harfbuzz version (1.0.1 on each), same font, but different metrics being returned. It's a Harfbuzz thing.

Here's Ubuntu 12:

$ md5sum  ~/.fonts/NotoSansCJKjp-Light.otf
f269c193b1ef0b739030d56386670c3f  /home/simon/.fonts/NotoSansCJKjp-Light.otf
$ hb-shape --direction=ttb ~/.fonts/NotoSansCJKjp-Light.otf '世界'
[gid9861=0@-500,-880+0,-1000|gid27506=1@-500,-880+0,-1000]

Build configuration:

Unicode callbacks (you want at least one):
        Glib:                   true
        ICU:                    true
        UCDN:                   false

Font callbacks (the more the better):
        FreeType:               true

Tools used for command-line utilities:
        Cairo:                  false
        Fontconfig:             true

Additional shapers (the more the better):
        Graphite2:              false

Here's OS X:

$ md5sum /Users/simon/Library/Fonts/NotoSansCJKjp-Light.otf
f269c193b1ef0b739030d56386670c3f  /Users/simon/Library/Fonts/NotoSansCJKjp-Light.otf
$ hb-shape --direction=ttb /Users/simon/Library/Fonts/NotoSansCJKjp-Light.otf '<00e4><00b8><0096><00e7><0095><008c>'
[gid9861=0@-500,-894+0,-1000|gid27506=1@-500,-894+0,-1000]

Build configuration:

Unicode callbacks (you want at least one):
    Glib:           true
    ICU:            true
    UCDN:           false

Font callbacks (the more the better):
    FreeType:       true

Tools used for command-line utilities:
    Cairo:          true
    Fontconfig:     true

Additional shapers (the more the better):
    Graphite2:      false

Platform shapers (not normally needed):
    CoreText:       false
    Uniscribe:      false

@behdad, help?

[behdad]

It's caused by some FreeType weirdness on OS X that I never spent the time to track down. Switching away from hb-ft to hb-ot for font funcs implementation (eg, --font-funcs=ot passed to hb-shape) should address this.

[simoncozens]

That works for hb-shape; excellent. But SILE can't move to the OT funcs until hb_ot_get_glyph_extents is implemented. So I guess we make the tests work on Travis. (and fail for me.)

[behdad]

SILE can't move to the OT funcs until hb_ot_get_glyph_extents is implemented.

That's on my todo for today.

[alerque]

@simoncozens I added an update to the expected output of this test to another test output related PR to bring the Travis test run that much closer to passing ever though this issue is outstanding.