compose & decompose from indic shaper need port to other shapers?

[behdad]

[Motivated by https://github.com//issues/381]

Back when we were using indic shaper for many more scripts, we added a decompose() function that decomposes some characters into two where they don't have a Unicode-sanctioned decomposition. Moreover, we have a compose() function that avoids recomposing matras; otherwise the split matras will be recomposed again and shape incorrectly.

However, when we added more shapers, we didn't add compose / decompose to them. In particular, the following characters are decomposed in indic shaper where these days they'll never reach that shaper:

    /* Tibetan */ 
    case 0x0F77u  : *a = 0x0FB2u; *b= 0x0F81u; return true; 
    case 0x0F79u  : *a = 0x0FB3u; *b= 0x0F81u; return true; 

    /* Limbu */ 
    case 0x1925u  : *a = 0x1920u; *b= 0x1923u; return true; 
    case 0x1926u  : *a = 0x1920u; *b= 0x1924u; return true; 
 
    /* Balinese */ 
    case 0x1B3Cu  : *a = 0x1B42u; *b= 0x1B3Cu; return true; 
 
    /* Chakma */ 
    case 0x1112Eu  : *a = 0x11127u; *b= 0x11131u; return true; 
    case 0x1112Fu  : *a = 0x11127u; *b= 0x11132u; return true; 

• Myanmar shaper: Myanmar doesn't seem to have split vowels. So maybe this one is fine.

• Tibetan: Do we need the decomposition? We should test with Windows. MS sends Tibetan through USE now. Maybe we should consider that.

• USE: The spec clearly says the font designer is responsive for decomposition if Unicode doesn't decompose. However, we should test what Windows does and adjust.

cc @jfkthame

[behdad]

/* Chakma */ case 0x1112Eu : *a = 0x11127u; *b= 0x11131u; return true; case 0x1112Fu : *a = 0x11127u; *b= 0x11132u; return true;

These seem to be in Unicode already (in the reverse order, but that shouldn't matter; right?). So, no idea why we put them here to begin with.