On 09/21/11 12:32, Khaled Hosny wrote: > On Tue, Sep 20, 2011 at 04:46:40PM -0400, Behdad Esfahbod wrote: >> On 09/20/11 14:32, Khaled Hosny wrote: >>> I just noticed that feature ranges assume byte not character count, e.g. >>> to apply a feature for first 10 Arabic characters I've to pass >>> [:20] not [:10], I can live with that but it wasn't obvious at first. >> >> Right. They are indices into the UTF-8 text. I can provide a mode to count >> characters. I agree that it would be easier that way. Maybe it should be >> the >> default? > > That would make more sense, I can't even think if use case where counting > by bytes would be preferred over characters.
Well, for machine-generated offsets, byte indices are easier. > I ended up using a csv file so I can specify options to pass to hb-shape > (I want to test optional and script/language sensitive features etc.) > and a python script to parse the file, run hb-shape and compare the > results (if anyone interested, check [1]) Great. I'm designing something similar today. behdad _______________________________________________ HarfBuzz mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/harfbuzz
