Incorrect x_advance in English result using Jellee Roman font since release 2.1.2 (easily reproducible)

[chanon]

I was integrating harfbuzz to my game and thought I did something wrong at first.

Thanks to the binary win32 releases I was able to use hb-view to verify that it seems to be a bug in harfbuzz since 2.1.2

2.1.2 and 2.1.3 produce the same results.
2.1.1 and before behave correctly.

To reproduce, get the Jellee Roman font. Eg. from here:
https://smartfonts.com/jellee-roman.font

With version 2.1.1:

hb-view.exe "Jellee Roman Bold.ttf" abcdefghijklmnop > result1.png

gives:

And

hb-view.exe "Jellee Roman Bold.ttf" ckckck > result2.png

gives:

With version 2.1.2 (and 2.1.3):

hb-view.exe "Jellee Roman Bold.ttf" abcdefghijklmnop > result1.png

gives:

And

hb-view.exe "Jellee Roman Bold.ttf" ckckck > result2.png

Notice the wrong advance of the c and k letters.

Otherwise, if this is an intended change, then it is pretty unexpected.

[ebraminio]

1a5ef84#diff-05711f830ee489547bf61904b5733b13R463 I guess

[ebraminio]

Fixes if I revert the commit

[ebraminio]

A format2 subtable, states itself as AAT kern (version 1) but expects OT one (version 0), interesting

[ebraminio]

Hmm, actually it got broken when format 2 support is added apparently, broken font I guess, we should see how CoreText and Uniscribe deals with it I guess however.

[behdad]

Our Format2 impl in kern table is untested. Thanks. I'll debug.

[ebraminio]

OK with CoreText

[behdad]

Can you paste the CoreText output with a few sequences? Those in the report, and maybe: AV VA Te eT Tr rT Td TA LV Vd pV Vq bV

[ebraminio]

$ util/hb-shape "Jellee Roman Bold.ttf" "abcdefghijklmnop ckckck AV VA Te eT Tr rT Td TA LV Vd pV Vq bV" --shaper coretext        
[a=0+626|b=1+672|c=2+549|d=3+672|e=4+620|f=5+393|g=6+672|h=7+665|i=8+316|j=9+316|k=10+590|l=11+316|m=12+1020|n=13+666|o=14+643|p=15+671|space=16+250|c=17+579|k=18+591|c=19+579|k=20+591|c=21+579|k=22+591|space=23+250|A=24+660|V=25+742|space=26+250|V=27+662|A=28+740|space=29+250|T=30+522|e=31+620|space=32+250|e=33+620|T=34+652|space=35+250|T=36+652|r=37+453|space=38+250|r=39+452|T=40+652|space=41+250|T=42+522|d=43+672|space=44+250|T=45+552|A=46+740|space=47+250|L=48+629|V=49+742|space=50+250|V=51+683|d=52+672|space=53+250|p=54+601|V=55+742|space=56+250|V=57+683|q=58+672|space=59+250|b=60+601|V=61+742]

[khaledhosny]

FontTools does not seem to be able to dump such kern subtable format, however FontForge can and the classes seem to be OK and I don’t seen any obvious bad values.

[behdad]

@ebraminio can I get those on separate lines please? Very hard to use as is. Thanks.

[behdad]

Oh, OFL. Great. We should add it to test suite. And yes, PITA that fonttools doesn't implement this format yet. Volunteers?

[ebraminio]

abcdefghijklmnop
ckckck
AV
VA
Te
eT
Tr
rT
Td
TA
LV
Vd
pV
Vq
bV

util/hb-shape --text-file a.txt Jellee\ Roman\ Bold.ttf --shaper coretext

[a=0+626|b=1+672|c=2+549|d=3+672|e=4+620|f=5+393|g=6+672|h=7+665|i=8+316|j=9+316|k=10+590|l=11+316|m=12+1020|n=13+666|o=14+643|p=15+671]
[c=0+579|k=1+591|c=2+578|k=3+590|c=4+579|k=5+590]
[A=0+660|V=1+743]
[V=0+663|A=1+740]
[T=0+521|e=1+621]
[e=0+621|T=1+652]
[T=0+652|r=1+452]
[r=0+452|T=1+652]
[T=0+521|d=1+672]
[T=0+551|A=1+741]
[L=0+629|V=1+742]
[V=0+682|d=1+671]
[p=0+602|V=1+743]
[V=0+682|q=1+671]
[b=0+602|V=1+743]

[behdad]

Any chance you can also provide me with the ftxdumperfuser of the kern table?