[HarfBuzz] HarfBuzz 1.0 API; the message you were hoping would never come

[Jonathan Kew]

On 3/1/14 11:57, Khaled Hosny wrote:
> On Fri, Jan 03, 2014 at 10:07:50AM +0000, Jonathan Kew wrote:
>> On 3/1/14 06:12, James Clark wrote:
>>> On Mon, Dec 30, 2013 at 7:26 PM, Jonathan Kew <jfkthame at googlemail.com
>>> <mailto:jfkthame at googlemail.com>> wrote:
>>>
>>>
>>>     There's overlap here with the process of font-matching (choosing the
>>>     font(s) to be used for a given text sequence), which is clearly out
>>>     of scope for harfbuzz. If a given Unicode character is not supported
>>>     (exactly, or via a *canonical* [de]composition) by a given font,
>>>     there are several possible outcomes: just render the font's .notdef
>>>     glyph; render some synthetic representation of the codepoint
>>>     (hexbox); render a compatibility-equivalent character/sequence, if
>>>     such exists; choose a different font.
>>>
>>>
>>> I agree with this model.  There's a general problem of what you might
>>> call "fallback": what to do if the requested font does not have a glyph
>>> for the requested character.
>>>
>>> This makes me realize that I don't understand the big picture of how
>>> this fallback process interacts with harfbuzz. In order to do fallback,
>>> you need to do character to glyph mapping.
>>
>> Not necessarily. You need to know the character repertoire supported by the
>> font, but you may not need to actually map to glyphs. In Firefox, for
>> instance, font fallback is done based on a per-font *bit* map of supported
>> Unicode codepoints. So at the font fallback stage, we know whether the
>> character is present, but do not map it to a glyph.
>
> One issue with that approach is you lose HarfBuzz’s ability to do
> Unicode composition/decomposition for unsupported characters, which
> should generally produce better resulting than font fallback. I wounder
> HarfBuzz can provide an API to help clients decide whether HarfBuzz can
> shape a given character with a given font or not? Or would that be too
> slow for practical use?

This is an interesting point, and one that we're still thinking about. 
Behdad has talked at some point about possibly having such an API, but 
we've never finalized anything.

> In LibreOffice we just pass the characters to HarfBuzz, and if returns a
> 0 glyph id for any character, font fallback is used for that one (some
> times with some of the surrounding characters to make sure clusters are
> rendered with the same font).

Yes, that's another possibility, though it wouldn't fit very readily 
into the current Gecko text architecture. It may well be a more natural 
approach for some other clients, though.

JK