Whitespace and USE dotted circles

[simoncozens]

6e059a4 introduced a surprising change. Before that, a mark following a space glyph got a dotted circle:

$ ./util/hb-shape NotoSansLimbu-Regular.ttf -u 1921,20,1921
[uni25CC=0+721|uni1921=0@-86,0+0|space=1+260|uni25CC=1+721|uni1921=1@-86,0+0]

Afterwards, no added dotted circle:

$ ./util/hb-shape NotoSansLimbu-Regular.ttf -u 1921,20,1921
[uni25CC=0+721|uni1921=0@-86,0+0|space=1+260|uni1921=1+0]

Is this by design? If seems strange to me that a space should constitute a cluster head.

[behdad]

Is this by design?

I don't think so.

[behdad]

Oops. That's because:

export O↦       = 0; # OTHER 

and:

symbol_cluster = (O | GB) tail?; 

@dscorbett I think the O there is wrong. WDYT?

[behdad]

The spec says instead of O there should be there:

UGC = So except U+25CC;

UGC = Sc

[dscorbett]

Using O there does indeed go against the spec. See #3249 for why I disagree with the spec.

Unicode recommends U+00A0 as the base for isolated combining marks. It used to recommend U+0020 but explicitly no longer does. It doesn’t say that U+0020 is invalid, just that there might be “potential conflicts” and “complications”.

One could make the case that HarfBuzz should interpret this discouragement as a prohibition, thus restoring the dotted circle. The simplest implementation is to override U+0020 to the class WJ. For other characters, though, I think dotted circles should only be added to text that is actually invalid, not merely strange.

[behdad]

I'm fine separating U+0020 only.

Should we do the same in other syllabic shapers as well?

[dscorbett]

It’s best to keep them consistent.