All NotoSerifCJK latin glyphs have unsafe to break flag

[khaledhosny]

Not sure if this is expected or not, but it seems all glyphs in the buffer except the first glyph and the glyph after space has unsafe to break flag set:

$ hb-shape NotoSerifCJK-Regular.ttc "abc abc" --show-flags --no-positions
[gid66=0|gid67=1#1|gid68=2#1|gid1=3#1|gid66=4|gid67=5#1|gid68=6#1]

Not sure how to debug this.

[behdad]

Try --trace?

[khaledhosny]

$ hb-shape NotoSerifCJK-Regular.ttc "abc" --trace --show-flags
trace: start lookup 2	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 2	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 3	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 3	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 4	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 4	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 5	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 5	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 30	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 30	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 32	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 32	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 44	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 44	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 3	buffer: [gid66=0+552|gid67=1+627|gid68=2+529]
trace: end lookup 3	buffer: [gid66=0+552|gid67=1+627#1|gid68=2+529#1]
[gid66=0+552|gid67=1+627#1|gid68=2+529#1]

This seems to be a lookup used in kern feature, and indeed disabling the kerning drops the flag:

$ hb-shape NotoSerifCJK-Regular.ttc "abc" --trace --show-flags --features="-kern"
trace: start lookup 2	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 2	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 3	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 3	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 4	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 4	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 5	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 5	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 30	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 30	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 32	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 32	buffer: [gid66=0|gid67=1|gid68=2]
trace: start lookup 44	buffer: [gid66=0|gid67=1|gid68=2]
trace: end lookup 44	buffer: [gid66=0|gid67=1|gid68=2]
[gid66=0+552|gid67=1+627|gid68=2+529]

A bit of looking around, this is coming from PairPosFormat2::apply(), so I guess it is intentional. I wasn’t expecting GPOS lookups to set the flag, and it didn’t look like there is any glyph substitution involved.

Kerning shouldn’t be an issue for what I’m trying to do (checking how much I need to reshape after hyphenation) because I can reuse the old kerning value, but I guess I have to live with it.

[behdad]

Is it actually kerning? We can make the code not flag if kern value was zero. We should.

[behdad]

It's probably a class kerning table that doesn't actually kern ideographs next to each other.

[jfkthame]

Kerning shouldn’t be an issue for what I’m trying to do (checking how much I need to reshape after hyphenation) because I can reuse the old kerning value, but I guess I have to live with it.

In general, I'd have thought you do need to reshape after hyphenation where kerning was present; if a glyph pair such as "AV" was kerned, but then you hyphenate so that one line ends "A-" and the next begins "V", the adjusted advances of the "A" and "V" glyphs resulting from the kerning will no longer be appropriate to use.

(Of course, if the kern value was zero we needn't set the flag.)

[khaledhosny]

Right, there is no difference in the positioning with or without kerning in the example above.

[khaledhosny]

Kerning shouldn’t be an issue for what I’m trying to do (checking how much I need to reshape after hyphenation) because I can reuse the old kerning value, but I guess I have to live with it.

In general, I'd have thought you do need to reshape after hyphenation where kerning was present; if a glyph pair such as "AV" was kerned, but then you hyphenate so that one line ends "A-" and the next begins "V", the adjusted advances of the "A" and "V" glyphs resulting from the kerning will no longer be appropriate to use.

(Of course, if the kern value was zero we needn't set the flag.)

This is for TeX’s discretionary nodes, what I do is to put the kern (kerning is done by a separate node that adds +ve or -ve space, not by changing the glyph advance) in the the “replace” list of the node, and the hyphen in the “pre” list, so if line breaking happens the kerning goes away, if not it gets kept.

[behdad]

Fixed in my cross-kern branch. Will reach master later today.