Space fallback in vertical text: set the y_advance?

[emuller-amazon]

In _hb_ot_shape_fallback_spaces, it seems that it is always the x_advance that is set. I believe the y_advance should be set for vertical text.

[behdad]

Any suggestions on what advance width various spaces should get in vertical?

[emuller-amazon]

Of the characters currently handled, U+3000 IDEOGRAPHIC SPACE has a default vertical orientation U, all the others have R.

For U+3000, I have a strong preference for 1em. That's better than the default value in case vmtx is missing (ascent - descent), imho: the most likely scenario is CJK text (since U+3000 is present in the text), set in a non CJK font (since U+3000 is not cmap'ed by the font), and the other (CJK) characters will fallback to a CJK font, with a typical 1em vertical advance.

All the other characters will probably be very rare in actual U orientation. One could argue for the same value as in the case of a missing vmtx table (ascent - descent), or one could argue for the same amount as in horizontal. I suppose that the most likely scenario for those characters is non-CJK text set vertically, and it's useful to have easy access to multiple widths (just like in horizontal), so I would suggest the second option.

[jfkthame]

In general, I'd agree with this. One exception, perhaps, would be U+2007 FIGURE SPACE, which arguably should have the same advance as numerals, so in U orientation that would default to ascent-descent if vmtx is absent. (I expect such occurrences to be so rare that I don't care strongly about it, though.)

[behdad]

FWIW, I think I switched default advance in absence of vmtx to 1em instead of ascent+descent.