John Cowan noted: > > So formal canonical decompositions are almost entirely > > confined to separable, accent-like diacritics (acute, > > grave, diaeresis, and so on). The only significant exceptions are > > the cedilla and ogonek, which attach smoothly to letter > > bottoms without otherwise distorting them, and which > > often have graphic alternates that are, indeed, separated > > diacritics (comma-like and reverse-comma-like forms). > > And the Vietnamese horn.
Yep, and the Vienamese horn. That one was on the hairy cusp of the decision, and in my opinion was probably decided wrongly -- the horn is not really a productive diacritic, and is more akin to the Cyrillic descender in terms of base letterform distortion. But that is water under the bridge at this point. Doug Ewell asked: > but I still don't see why stroke overlays are lumped in > with that group. They don't distort the base form any more than > cedillas and ogoneks do -- and isn't this a glyph issue anyway? Well, sure, it *is* a glyph issue. But, again, you have to draw the line somewhere. Where, in the following continuum, would you draw the line: z (007A), z-acute (017A), z-bar (01B6), z-curl (0291), ezh (0292), ezh-curl (0293), ezh-tail (01BA), ... One of the reasons why the "framers" decided, for Latin, to draw the line before the bars is that unlike the free-floating accents, which can be dynamically placed reasonably well by referring to the overall bounding box and "ink" of the base character, the bars are rather more sensitive in placement to the actual shape of the glyph they have to cross. In some cases, this leads to structurally distinct solutions, as in the glyph families for b-bar and d-bar (cross the ascender? cross the bowl?), and so on. It was just far enough over into the *glyph design in a font* side of the issue to make decomposition uncomfortable, whereas the placement of floating accents is just far enough over into the *generative placement* side of the issue to make sense. But it was clearly a judgement call, and as Doug noted, the decision was made long ago, and second-guessing it now won't make any difference to the actual decompositions in implementations. --Ken