From 191fa885d9e0a2dce92dd8727cddd18495e62409 Mon Sep 17 00:00:00 2001 From: Behdad Esfahbod Date: Tue, 28 Aug 2012 22:58:55 -0400 Subject: [PATCH] [GSUB] Merge Ligature and context input matching Looks better now... --- TODO | 4 -- src/hb-ot-layout-gsub-table.hh | 71 +++++++----------------------------- src/hb-ot-layout-gsubgpos-private.hh | 29 ++++++++++++++- 3 files changed, 40 insertions(+), 64 deletions(-) diff --git a/TODO b/TODO index 441eb4d..1ca2480 100644 --- a/TODO +++ b/TODO @@ -1,10 +1,6 @@ General fixes: ============= -- Ligature matching and match_input() of (Chain)Context should use the - same logic. Right now the Ligature logic is more involved. Possibly - merge, or duplicate. - - mask propagation? (when ligation, "or" the masks). - Warn at compile time (and runtime with HB_DEBUG?) if no Unicode / font diff --git a/src/hb-ot-layout-gsub-table.hh b/src/hb-ot-layout-gsub-table.hh index c85df46..ac3e8a9 100644 --- a/src/hb-ot-layout-gsub-table.hh +++ b/src/hb-ot-layout-gsub-table.hh @@ -491,66 +491,21 @@ struct Ligature unsigned int count = component.len; if (unlikely (count < 1)) return TRACE_RETURN (false); - hb_apply_context_t::mark_skipping_forward_iterator_t skippy_iter (c, c->buffer->idx, count - 1); - if (skippy_iter.has_no_chance ()) return TRACE_RETURN (false); - - /* - * This is perhaps the trickiest part of OpenType... Remarks: - * - * - If all components of the ligature were marks, we call this a mark ligature. - * - * - If there is no GDEF, and the ligature is NOT a mark ligature, we categorize - * it as a ligature glyph. - * - * - Ligatures cannot be formed across glyphs attached to different components - * of previous ligatures. Eg. the sequence is LAM,SHADDA,LAM,FATHA,HEH, and - * LAM,LAM,HEH form a ligature, leaving SHADDA,FATHA next to eachother. - * However, it would be wrong to ligate that SHADDA,FATHA sequence.o - * There is an exception to this: If a ligature tries ligating with marks that - * belong to it itself, go ahead, assuming that the font designer knows what - * they are doing (otherwise it can break Indic stuff when a matra wants to - * ligate with a conjunct...) - */ - - bool is_mark_ligature = !!(c->property & HB_OT_LAYOUT_GLYPH_CLASS_MARK); - - unsigned int total_component_count = 0; - total_component_count += get_lig_num_comps (c->buffer->cur()); - - unsigned int first_lig_id = get_lig_id (c->buffer->cur()); - unsigned int first_lig_comp = get_lig_comp (c->buffer->cur()); - - for (unsigned int i = 1; i < count; i++) - { - unsigned int property; - - if (!skippy_iter.next (&property)) return TRACE_RETURN (false); - - if (likely (c->buffer->info[skippy_iter.idx].codepoint != component[i])) return TRACE_RETURN (false); - - unsigned int this_lig_id = get_lig_id (c->buffer->info[skippy_iter.idx]); - unsigned int this_lig_comp = get_lig_comp (c->buffer->info[skippy_iter.idx]); - - if (first_lig_id && first_lig_comp) { - /* If first component was attached to a previous ligature component, - * all subsequent components should be attached to the same ligature - * component, otherwise we shouldn't ligate them. */ - if (first_lig_id != this_lig_id || first_lig_comp != this_lig_comp) - return TRACE_RETURN (false); - } else { - /* If first component was NOT attached to a previous ligature component, - * all subsequent components should also NOT be attached to any ligature - * component, unless they are attached to the first component itself! */ - if (this_lig_id && this_lig_comp && (this_lig_id != first_lig_id)) - return TRACE_RETURN (false); - } - - is_mark_ligature = is_mark_ligature && (property & HB_OT_LAYOUT_GLYPH_CLASS_MARK); - total_component_count += get_lig_num_comps (c->buffer->info[skippy_iter.idx]); - } + unsigned int end_offset; + bool is_mark_ligature; + unsigned int total_component_count; + + if (likely (!match_input (c, count, + &component[1], + match_glyph, + NULL, + &end_offset, + &is_mark_ligature, + &total_component_count))) + return TRACE_RETURN (false); /* Deal, we are forming the ligature. */ - c->buffer->merge_clusters (c->buffer->idx, skippy_iter.idx + 1); + c->buffer->merge_clusters (c->buffer->idx, c->buffer->idx + end_offset); /* * - If it *is* a mark ligature, we don't allocate a new ligature id, and leave diff --git a/src/hb-ot-layout-gsubgpos-private.hh b/src/hb-ot-layout-gsubgpos-private.hh index 5032d30..3d69736 100644 --- a/src/hb-ot-layout-gsubgpos-private.hh +++ b/src/hb-ot-layout-gsubgpos-private.hh @@ -421,13 +421,33 @@ static inline bool match_input (hb_apply_context_t *c, const USHORT input[], /* Array of input values--start with second glyph */ match_func_t match_func, const void *match_data, - unsigned int *end_offset = NULL) + unsigned int *end_offset = NULL, + bool *p_is_mark_ligature = NULL, + unsigned int *p_total_component_count = NULL) { hb_auto_trace_t trace (&c->debug_depth, "APPLY", NULL, HB_FUNC, "idx %d codepoint %u", c->buffer->idx, c->buffer->cur().codepoint); hb_apply_context_t::mark_skipping_forward_iterator_t skippy_iter (c, c->buffer->idx, count - 1); if (skippy_iter.has_no_chance ()) return TRACE_RETURN (false); + /* + * This is perhaps the trickiest part of OpenType... Remarks: + * + * - If all components of the ligature were marks, we call this a mark ligature. + * + * - If there is no GDEF, and the ligature is NOT a mark ligature, we categorize + * it as a ligature glyph. + * + * - Ligatures cannot be formed across glyphs attached to different components + * of previous ligatures. Eg. the sequence is LAM,SHADDA,LAM,FATHA,HEH, and + * LAM,LAM,HEH form a ligature, leaving SHADDA,FATHA next to eachother. + * However, it would be wrong to ligate that SHADDA,FATHA sequence.o + * There is an exception to this: If a ligature tries ligating with marks that + * belong to it itself, go ahead, assuming that the font designer knows what + * they are doing (otherwise it can break Indic stuff when a matra wants to + * ligate with a conjunct...) + */ + bool is_mark_ligature = !!(c->property & HB_OT_LAYOUT_GLYPH_CLASS_MARK); unsigned int total_component_count = 0; @@ -443,7 +463,6 @@ static inline bool match_input (hb_apply_context_t *c, if (!skippy_iter.next (&property)) return TRACE_RETURN (false); if (likely (!match_func (c->buffer->info[skippy_iter.idx].codepoint, input[i - 1], match_data))) return false; -// if (likely (c->buffer->info[skippy_iter.idx].codepoint != component[i])) return TRACE_RETURN (false); unsigned int this_lig_id = get_lig_id (c->buffer->info[skippy_iter.idx]); unsigned int this_lig_comp = get_lig_comp (c->buffer->info[skippy_iter.idx]); @@ -469,6 +488,12 @@ static inline bool match_input (hb_apply_context_t *c, if (end_offset) *end_offset = skippy_iter.idx - c->buffer->idx + 1; + if (p_is_mark_ligature) + *p_is_mark_ligature = is_mark_ligature; + + if (p_total_component_count) + *p_total_component_count = total_component_count; + return true; } -- 2.7.4