* Google Author(s): Behdad Esfahbod
*/
-#include "hb-ot-shape-complex-private.hh"
+#include "hb-private.hh"
+#include "hb-debug.hh"
+#include "hb-ot-shape-complex-arabic-private.hh"
#include "hb-ot-shape-private.hh"
-
/* buffer var allocations */
#define arabic_shaping_action() complex_var_u8_0() /* arabic shaping action */
+#define HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH HB_BUFFER_SCRATCH_FLAG_COMPLEX0
+
+/* See:
+ * https://github.com/harfbuzz/harfbuzz/commit/6e6f82b6f3dde0fc6c3c7d991d9ec6cfff57823d#commitcomment-14248516 */
+#define HB_ARABIC_GENERAL_CATEGORY_IS_WORD(gen_cat) \
+ (FLAG_UNSAFE (gen_cat) & \
+ (FLAG (HB_UNICODE_GENERAL_CATEGORY_UNASSIGNED) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_PRIVATE_USE) | \
+ /*FLAG (HB_UNICODE_GENERAL_CATEGORY_LOWERCASE_LETTER) |*/ \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_LETTER) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_LETTER) | \
+ /*FLAG (HB_UNICODE_GENERAL_CATEGORY_TITLECASE_LETTER) |*/ \
+ /*FLAG (HB_UNICODE_GENERAL_CATEGORY_UPPERCASE_LETTER) |*/ \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_SPACING_MARK) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_DECIMAL_NUMBER) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_LETTER_NUMBER) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_NUMBER) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_CURRENCY_SYMBOL) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_MODIFIER_SYMBOL) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_MATH_SYMBOL) | \
+ FLAG (HB_UNICODE_GENERAL_CATEGORY_OTHER_SYMBOL)))
+
+
+/*
+ * Joining types:
+ */
/*
* Bits used in the joining tables
*/
-enum {
+enum hb_arabic_joining_type_t {
JOINING_TYPE_U = 0,
- JOINING_TYPE_R = 1,
- JOINING_TYPE_D = 2,
+ JOINING_TYPE_L = 1,
+ JOINING_TYPE_R = 2,
+ JOINING_TYPE_D = 3,
JOINING_TYPE_C = JOINING_TYPE_D,
- JOINING_GROUP_ALAPH = 3,
- JOINING_GROUP_DALATH_RISH = 4,
- NUM_STATE_MACHINE_COLS = 5,
+ JOINING_GROUP_ALAPH = 4,
+ JOINING_GROUP_DALATH_RISH = 5,
+ NUM_STATE_MACHINE_COLS = 6,
- /* We deliberately don't have a JOINING_TYPE_L since that's unused in Unicode. */
-
- JOINING_TYPE_T = 6,
- JOINING_TYPE_X = 7 /* means: use general-category to choose between U or T. */
+ JOINING_TYPE_T = 7,
+ JOINING_TYPE_X = 8 /* means: use general-category to choose between U or T. */
};
-/*
- * Joining types:
- */
-
#include "hb-ot-shape-complex-arabic-table.hh"
static unsigned int get_joining_type (hb_codepoint_t u, hb_unicode_general_category_t gen_cat)
{
- if (likely (hb_in_range<hb_codepoint_t> (u, JOINING_TABLE_FIRST, JOINING_TABLE_LAST))) {
- unsigned int j_type = joining_table[u - JOINING_TABLE_FIRST];
- if (likely (j_type != JOINING_TYPE_X))
- return j_type;
- }
-
- /* Mongolian joining data is not in ArabicJoining.txt yet */
- if (unlikely (hb_in_range<hb_codepoint_t> (u, 0x1800, 0x18AF)))
- {
- /* All letters, SIBE SYLLABLE BOUNDARY MARKER, and NIRUGU are D */
- if (gen_cat == HB_UNICODE_GENERAL_CATEGORY_OTHER_LETTER || u == 0x1807 || u == 0x180A)
- return JOINING_TYPE_D;
- }
-
- if (unlikely (hb_in_range<hb_codepoint_t> (u, 0x200C, 0x200D))) {
- return u == 0x200C ? JOINING_TYPE_U : JOINING_TYPE_C;
- }
-
- return (FLAG(gen_cat) & (FLAG(HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) | FLAG(HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) | FLAG(HB_UNICODE_GENERAL_CATEGORY_FORMAT))) ?
- JOINING_TYPE_T : JOINING_TYPE_U;
+ unsigned int j_type = joining_type(u);
+ if (likely (j_type != JOINING_TYPE_X))
+ return j_type;
+
+ return (FLAG_UNSAFE(gen_cat) &
+ (FLAG(HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK) |
+ FLAG(HB_UNICODE_GENERAL_CATEGORY_ENCLOSING_MARK) |
+ FLAG(HB_UNICODE_GENERAL_CATEGORY_FORMAT))
+ ) ? JOINING_TYPE_T : JOINING_TYPE_U;
}
-static hb_codepoint_t get_arabic_shape (hb_codepoint_t u, unsigned int shape)
-{
- if (likely (hb_in_range<hb_codepoint_t> (u, SHAPING_TABLE_FIRST, SHAPING_TABLE_LAST)) && shape < 4)
- return shaping_table[u - SHAPING_TABLE_FIRST][shape];
- return u;
-}
-
-static uint16_t get_ligature (hb_codepoint_t first, hb_codepoint_t second)
-{
- if (unlikely (!second)) return 0;
- for (unsigned i = 0; i < ARRAY_LENGTH (ligature_table); i++)
- if (ligature_table[i].first == first)
- for (unsigned j = 0; j < ARRAY_LENGTH (ligature_table[i].ligatures); j++)
- if (ligature_table[i].ligatures[j].second == second)
- return ligature_table[i].ligatures[j].ligature;
- return 0;
-}
+#define FEATURE_IS_SYRIAC(tag) hb_in_range<unsigned char> ((unsigned char) (tag), '2', '3')
static const hb_tag_t arabic_features[] =
{
- HB_TAG('i','n','i','t'),
- HB_TAG('m','e','d','i'),
- HB_TAG('f','i','n','a'),
HB_TAG('i','s','o','l'),
- /* Syriac */
- HB_TAG('m','e','d','2'),
+ HB_TAG('f','i','n','a'),
HB_TAG('f','i','n','2'),
HB_TAG('f','i','n','3'),
+ HB_TAG('m','e','d','i'),
+ HB_TAG('m','e','d','2'),
+ HB_TAG('i','n','i','t'),
HB_TAG_NONE
};
/* Same order as the feature array */
-enum {
- INIT,
- MEDI,
- FINA,
+enum arabic_action_t {
ISOL,
-
- /* Syriac */
- MED2,
+ FINA,
FIN2,
FIN3,
+ MEDI,
+ MED2,
+ INIT,
NONE,
- ARABIC_NUM_FEATURES = NONE
+ ARABIC_NUM_FEATURES = NONE,
+
+ /* We abuse the same byte for other things... */
+ STCH_FIXED,
+ STCH_REPEATING,
};
static const struct arabic_state_table_entry {
uint16_t next_state;
} arabic_state_table[][NUM_STATE_MACHINE_COLS] =
{
- /* jt_U, jt_R, jt_D, jg_ALAPH, jg_DALATH_RISH */
+ /* jt_U, jt_L, jt_R, jt_D, jg_ALAPH, jg_DALATH_RISH */
/* State 0: prev was U, not willing to join. */
- { {NONE,NONE,0}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,6}, },
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,6}, },
/* State 1: prev was R or ISOL/ALAPH, not willing to join. */
- { {NONE,NONE,0}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN2,5}, {NONE,ISOL,6}, },
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN2,5}, {NONE,ISOL,6}, },
- /* State 2: prev was D/ISOL, willing to join. */
- { {NONE,NONE,0}, {INIT,FINA,1}, {INIT,FINA,3}, {INIT,FINA,4}, {INIT,FINA,6}, },
+ /* State 2: prev was D/L in ISOL form, willing to join. */
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {INIT,FINA,1}, {INIT,FINA,3}, {INIT,FINA,4}, {INIT,FINA,6}, },
- /* State 3: prev was D/FINA, willing to join. */
- { {NONE,NONE,0}, {MEDI,FINA,1}, {MEDI,FINA,3}, {MEDI,FINA,4}, {MEDI,FINA,6}, },
+ /* State 3: prev was D in FINA form, willing to join. */
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {MEDI,FINA,1}, {MEDI,FINA,3}, {MEDI,FINA,4}, {MEDI,FINA,6}, },
/* State 4: prev was FINA ALAPH, not willing to join. */
- { {NONE,NONE,0}, {MED2,ISOL,1}, {MED2,ISOL,2}, {MED2,FIN2,5}, {MED2,ISOL,6}, },
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {MED2,ISOL,1}, {MED2,ISOL,2}, {MED2,FIN2,5}, {MED2,ISOL,6}, },
/* State 5: prev was FIN2/FIN3 ALAPH, not willing to join. */
- { {NONE,NONE,0}, {ISOL,ISOL,1}, {ISOL,ISOL,2}, {ISOL,FIN2,5}, {ISOL,ISOL,6}, },
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {ISOL,ISOL,1}, {ISOL,ISOL,2}, {ISOL,FIN2,5}, {ISOL,ISOL,6}, },
/* State 6: prev was DALATH/RISH, not willing to join. */
- { {NONE,NONE,0}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN3,5}, {NONE,ISOL,6}, }
+ { {NONE,NONE,0}, {NONE,ISOL,2}, {NONE,ISOL,1}, {NONE,ISOL,2}, {NONE,FIN3,5}, {NONE,ISOL,6}, }
};
+static void
+nuke_joiners (const hb_ot_shape_plan_t *plan,
+ hb_font_t *font,
+ hb_buffer_t *buffer);
+
+static void
+arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
+ hb_font_t *font,
+ hb_buffer_t *buffer);
+
+static void
+record_stch (const hb_ot_shape_plan_t *plan,
+ hb_font_t *font,
+ hb_buffer_t *buffer);
static void
collect_features_arabic (hb_ot_shape_planner_t *plan)
{
hb_ot_map_builder_t *map = &plan->map;
- /* For Language forms (in ArabicOT speak), we do the iso/fina/medi/init together,
- * then rlig and calt each in their own stage. This makes IranNastaliq's ALLAH
- * ligature work correctly. It's unfortunate though...
+ /* We apply features according to the Arabic spec, with pauses
+ * in between most.
*
- * This also makes Arial Bold in Windows7 work. See:
+ * The pause between init/medi/... and rlig is required. See eg:
* https://bugzilla.mozilla.org/show_bug.cgi?id=644184
*
- * TODO: Add test cases for these two.
+ * The pauses between init/medi/... themselves are not necessarily
+ * needed as only one of those features is applied to any character.
+ * The only difference it makes is when fonts have contextual
+ * substitutions. We now follow the order of the spec, which makes
+ * for better experience if that's what Uniscribe is doing.
+ *
+ * At least for Arabic, looks like Uniscribe has a pause between
+ * rlig and calt. Otherwise the IranNastaliq's ALLAH ligature won't
+ * work. However, testing shows that rlig and calt are applied
+ * together for Mongolian in Uniscribe. As such, we only add a
+ * pause for Arabic, not other scripts.
+ *
+ * A pause after calt is required to make KFGQPC Uthmanic Script HAFS
+ * work correctly. See https://github.com/harfbuzz/harfbuzz/issues/505
*/
- map->add_bool_feature (HB_TAG('c','c','m','p'));
- map->add_bool_feature (HB_TAG('l','o','c','l'));
+ map->add_gsub_pause (nuke_joiners);
- map->add_gsub_pause (NULL);
+ map->add_global_bool_feature (HB_TAG('s','t','c','h'));
+ map->add_gsub_pause (record_stch);
- for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++)
- map->add_bool_feature (arabic_features[i], false);
+ map->add_global_bool_feature (HB_TAG('c','c','m','p'));
+ map->add_global_bool_feature (HB_TAG('l','o','c','l'));
- map->add_gsub_pause (NULL);
+ map->add_gsub_pause (nullptr);
- map->add_bool_feature (HB_TAG('r','l','i','g'));
- map->add_gsub_pause (NULL);
-
- map->add_bool_feature (HB_TAG('c','a','l','t'));
- map->add_gsub_pause (NULL);
+ for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++)
+ {
+ bool has_fallback = plan->props.script == HB_SCRIPT_ARABIC && !FEATURE_IS_SYRIAC (arabic_features[i]);
+ map->add_feature (arabic_features[i], 1, has_fallback ? F_HAS_FALLBACK : F_NONE);
+ map->add_gsub_pause (nullptr);
+ }
- /* ArabicOT spec enables 'cswh' for Arabic where as for basic shaper it's disabled by default. */
- map->add_bool_feature (HB_TAG('c','s','w','h'));
+ map->add_feature (HB_TAG('r','l','i','g'), 1, F_GLOBAL|F_HAS_FALLBACK);
+ if (plan->props.script == HB_SCRIPT_ARABIC)
+ map->add_gsub_pause (arabic_fallback_shape);
+
+ /* No pause after rclt. See 98460779bae19e4d64d29461ff154b3527bf8420. */
+ map->add_global_bool_feature (HB_TAG('r','c','l','t'));
+ map->add_global_bool_feature (HB_TAG('c','a','l','t'));
+ map->add_gsub_pause (nullptr);
+
+ /* The spec includes 'cswh'. Earlier versions of Windows
+ * used to enable this by default, but testing suggests
+ * that Windows 8 and later do not enable it by default,
+ * and spec now says 'Off by default'.
+ * We disabled this in ae23c24c32.
+ * Note that IranNastaliq uses this feature extensively
+ * to fixup broken glyph sequences. Oh well...
+ * Test case: U+0643,U+0640,U+0631. */
+ //map->add_global_bool_feature (HB_TAG('c','s','w','h'));
+ map->add_global_bool_feature (HB_TAG('m','s','e','t'));
}
+#include "hb-ot-shape-complex-arabic-fallback.hh"
+
struct arabic_shape_plan_t
{
ASSERT_POD ();
- bool do_fallback;
/* The "+ 1" in the next array is to accommodate for the "NONE" command,
* which is not an OpenType feature, but this simplifies the code by not
* having to do a "if (... < NONE) ..." and just rely on the fact that
* mask_array[NONE] == 0. */
hb_mask_t mask_array[ARABIC_NUM_FEATURES + 1];
+
+ arabic_fallback_plan_t *fallback_plan;
+
+ unsigned int do_fallback : 1;
+ unsigned int has_stch : 1;
};
-static void *
+void *
data_create_arabic (const hb_ot_shape_plan_t *plan)
{
arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) calloc (1, sizeof (arabic_shape_plan_t));
if (unlikely (!arabic_plan))
- return NULL;
+ return nullptr;
- hb_mask_t total_masks = 0;
+ arabic_plan->do_fallback = plan->props.script == HB_SCRIPT_ARABIC;
+ arabic_plan->has_stch = !!plan->map.get_1_mask (HB_TAG ('s','t','c','h'));
for (unsigned int i = 0; i < ARABIC_NUM_FEATURES; i++) {
arabic_plan->mask_array[i] = plan->map.get_1_mask (arabic_features[i]);
- total_masks |= arabic_plan->mask_array[i];
+ arabic_plan->do_fallback = arabic_plan->do_fallback &&
+ (FEATURE_IS_SYRIAC (arabic_features[i]) ||
+ plan->map.needs_fallback (arabic_features[i]));
}
- /* Pitfalls:
- * - This path fires if user force-set init/medi/fina/isol off,
- * - If font does not declare script 'arab', well, what to do?
- * Most probably it's safe to assume that init/medi/fina/isol
- * still mean Arabic shaping, although they do not have to.
- */
- arabic_plan->do_fallback = 0 == total_masks;
-
return arabic_plan;
}
-static void
+void
data_destroy_arabic (void *data)
{
+ arabic_shape_plan_t *arabic_plan = (arabic_shape_plan_t *) data;
+
+ arabic_fallback_plan_destroy (arabic_plan->fallback_plan);
+
free (data);
}
static void
-arabic_fallback_shape (hb_font_t *font, hb_buffer_t *buffer)
+arabic_joining (hb_buffer_t *buffer)
{
- /* Only Arabic has presentation forms encoded in Unicode. */
- if (buffer->props.script != HB_SCRIPT_ARABIC)
- return;
-
unsigned int count = buffer->len;
- hb_codepoint_t glyph;
-
- /* Shape to presentation forms */
- for (unsigned int i = 0; i < count; i++) {
- hb_codepoint_t u = buffer->info[i].codepoint;
- hb_codepoint_t shaped = get_arabic_shape (u, buffer->info[i].arabic_shaping_action());
- if (shaped != u && font->get_glyph (shaped, 0, &glyph))
- buffer->info[i].codepoint = shaped;
+ hb_glyph_info_t *info = buffer->info;
+ unsigned int prev = (unsigned int) -1, state = 0;
+
+ /* Check pre-context */
+ for (unsigned int i = 0; i < buffer->context_len[0]; i++)
+ {
+ unsigned int this_type = get_joining_type (buffer->context[0][i], buffer->unicode->general_category (buffer->context[0][i]));
+
+ if (unlikely (this_type == JOINING_TYPE_T))
+ continue;
+
+ const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
+ state = entry->next_state;
+ break;
}
- /* Mandatory ligatures */
- buffer->clear_output ();
- for (buffer->idx = 0; buffer->idx + 1 < count;) {
- hb_codepoint_t ligature = get_ligature (buffer->cur().codepoint,
- buffer->cur(+1).codepoint);
- if (likely (!ligature) || !(font->get_glyph (ligature, 0, &glyph))) {
- buffer->next_glyph ();
+ for (unsigned int i = 0; i < count; i++)
+ {
+ unsigned int this_type = get_joining_type (info[i].codepoint, _hb_glyph_info_get_general_category (&info[i]));
+
+ if (unlikely (this_type == JOINING_TYPE_T)) {
+ info[i].arabic_shaping_action() = NONE;
continue;
}
- buffer->replace_glyphs (2, 1, &ligature);
+ const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
- /* Technically speaking we can skip marks and stuff, like the GSUB path does.
- * But who cares, we're in fallback! */
+ if (entry->prev_action != NONE && prev != (unsigned int) -1)
+ {
+ info[prev].arabic_shaping_action() = entry->prev_action;
+ buffer->unsafe_to_break (prev, i + 1);
+ }
+
+ info[i].arabic_shaping_action() = entry->curr_action;
+
+ prev = i;
+ state = entry->next_state;
+ }
+
+ for (unsigned int i = 0; i < buffer->context_len[1]; i++)
+ {
+ unsigned int this_type = get_joining_type (buffer->context[1][i], buffer->unicode->general_category (buffer->context[1][i]));
+
+ if (unlikely (this_type == JOINING_TYPE_T))
+ continue;
+
+ const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
+ if (entry->prev_action != NONE && prev != (unsigned int) -1)
+ info[prev].arabic_shaping_action() = entry->prev_action;
+ break;
}
- for (; buffer->idx < count;)
- buffer->next_glyph ();
- buffer->swap_buffers ();
}
static void
-arabic_joining (hb_buffer_t *buffer)
+mongolian_variation_selectors (hb_buffer_t *buffer)
{
+ /* Copy arabic_shaping_action() from base to Mongolian variation selectors. */
unsigned int count = buffer->len;
- unsigned int prev = 0, state = 0;
+ hb_glyph_info_t *info = buffer->info;
+ for (unsigned int i = 1; i < count; i++)
+ if (unlikely (hb_in_range<hb_codepoint_t> (info[i].codepoint, 0x180Bu, 0x180Du)))
+ info[i].arabic_shaping_action() = info[i - 1].arabic_shaping_action();
+}
+void
+setup_masks_arabic_plan (const arabic_shape_plan_t *arabic_plan,
+ hb_buffer_t *buffer,
+ hb_script_t script)
+{
HB_BUFFER_ALLOCATE_VAR (buffer, arabic_shaping_action);
+ arabic_joining (buffer);
+ if (script == HB_SCRIPT_MONGOLIAN)
+ mongolian_variation_selectors (buffer);
+
+ unsigned int count = buffer->len;
+ hb_glyph_info_t *info = buffer->info;
for (unsigned int i = 0; i < count; i++)
- {
- unsigned int this_type = get_joining_type (buffer->info[i].codepoint, _hb_glyph_info_get_general_category (&buffer->info[i]));
+ info[i].mask |= arabic_plan->mask_array[info[i].arabic_shaping_action()];
+}
- if (unlikely (this_type == JOINING_TYPE_T)) {
- buffer->info[i].arabic_shaping_action() = NONE;
- continue;
- }
+static void
+setup_masks_arabic (const hb_ot_shape_plan_t *plan,
+ hb_buffer_t *buffer,
+ hb_font_t *font HB_UNUSED)
+{
+ const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
+ setup_masks_arabic_plan (arabic_plan, buffer, plan->props.script);
+}
- const arabic_state_table_entry *entry = &arabic_state_table[state][this_type];
- if (entry->prev_action != NONE)
- buffer->info[prev].arabic_shaping_action() = entry->prev_action;
+static void
+nuke_joiners (const hb_ot_shape_plan_t *plan HB_UNUSED,
+ hb_font_t *font HB_UNUSED,
+ hb_buffer_t *buffer)
+{
+ unsigned int count = buffer->len;
+ hb_glyph_info_t *info = buffer->info;
+ for (unsigned int i = 0; i < count; i++)
+ if (_hb_glyph_info_is_zwj (&info[i]))
+ _hb_glyph_info_flip_joiners (&info[i]);
+}
- buffer->info[i].arabic_shaping_action() = entry->curr_action;
+static void
+arabic_fallback_shape (const hb_ot_shape_plan_t *plan,
+ hb_font_t *font,
+ hb_buffer_t *buffer)
+{
+ const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
- prev = i;
- state = entry->next_state;
+ if (!arabic_plan->do_fallback)
+ return;
+
+retry:
+ arabic_fallback_plan_t *fallback_plan = (arabic_fallback_plan_t *) hb_atomic_ptr_get (&arabic_plan->fallback_plan);
+ if (unlikely (!fallback_plan))
+ {
+ /* This sucks. We need a font to build the fallback plan... */
+ fallback_plan = arabic_fallback_plan_create (plan, font);
+ if (unlikely (!hb_atomic_ptr_cmpexch (&(const_cast<arabic_shape_plan_t *> (arabic_plan))->fallback_plan, nullptr, fallback_plan))) {
+ arabic_fallback_plan_destroy (fallback_plan);
+ goto retry;
+ }
}
- HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
+ arabic_fallback_plan_shape (fallback_plan, font, buffer);
}
+/*
+ * Stretch feature: "stch".
+ * See example here:
+ * https://docs.microsoft.com/en-us/typography/script-development/syriac
+ * We implement this in a generic way, such that the Arabic subtending
+ * marks can use it as well.
+ */
+
static void
-preprocess_text_arabic (const hb_ot_shape_plan_t *plan,
- hb_buffer_t *buffer,
- hb_font_t *font)
+record_stch (const hb_ot_shape_plan_t *plan,
+ hb_font_t *font,
+ hb_buffer_t *buffer)
{
const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
+ if (!arabic_plan->has_stch)
+ return;
- if (unlikely (arabic_plan->do_fallback))
+ /* 'stch' feature was just applied. Look for anything that multiplied,
+ * and record it for stch treatment later. Note that rtlm, frac, etc
+ * are applied before stch, but we assume that they didn't result in
+ * anything multiplying into 5 pieces, so it's safe-ish... */
+
+ unsigned int count = buffer->len;
+ hb_glyph_info_t *info = buffer->info;
+ for (unsigned int i = 0; i < count; i++)
+ if (unlikely (_hb_glyph_info_multiplied (&info[i])))
+ {
+ unsigned int comp = _hb_glyph_info_get_lig_comp (&info[i]);
+ info[i].arabic_shaping_action() = comp % 2 ? STCH_REPEATING : STCH_FIXED;
+ buffer->scratch_flags |= HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH;
+ }
+}
+
+static void
+apply_stch (const hb_ot_shape_plan_t *plan,
+ hb_buffer_t *buffer,
+ hb_font_t *font)
+{
+ if (likely (!(buffer->scratch_flags & HB_BUFFER_SCRATCH_FLAG_ARABIC_HAS_STCH)))
+ return;
+
+ /* The Arabic shaper currently always processes in RTL mode, so we should
+ * stretch / position the stretched pieces to the left / preceding glyphs. */
+
+ /* We do a two pass implementation:
+ * First pass calculates the exact number of extra glyphs we need,
+ * We then enlarge buffer to have that much room,
+ * Second pass applies the stretch, copying things to the end of buffer.
+ */
+
+ int sign = font->x_scale < 0 ? -1 : +1;
+ unsigned int extra_glyphs_needed = 0; // Set during MEASURE, used during CUT
+ typedef enum { MEASURE, CUT } step_t;
+
+ for (step_t step = MEASURE; step <= CUT; step = (step_t) (step + 1))
{
- arabic_joining (buffer);
- arabic_fallback_shape (font, buffer);
+ unsigned int count = buffer->len;
+ hb_glyph_info_t *info = buffer->info;
+ hb_glyph_position_t *pos = buffer->pos;
+ unsigned int new_len = count + extra_glyphs_needed; // write head during CUT
+ unsigned int j = new_len;
+ for (unsigned int i = count; i; i--)
+ {
+ if (!hb_in_range<unsigned> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
+ {
+ if (step == CUT)
+ {
+ --j;
+ info[j] = info[i - 1];
+ pos[j] = pos[i - 1];
+ }
+ continue;
+ }
+
+ /* Yay, justification! */
+
+ hb_position_t w_total = 0; // Total to be filled
+ hb_position_t w_fixed = 0; // Sum of fixed tiles
+ hb_position_t w_repeating = 0; // Sum of repeating tiles
+ int n_fixed = 0;
+ int n_repeating = 0;
+
+ unsigned int end = i;
+ while (i &&
+ hb_in_range<unsigned> (info[i - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING))
+ {
+ i--;
+ hb_position_t width = font->get_glyph_h_advance (info[i].codepoint);
+ if (info[i].arabic_shaping_action() == STCH_FIXED)
+ {
+ w_fixed += width;
+ n_fixed++;
+ }
+ else
+ {
+ w_repeating += width;
+ n_repeating++;
+ }
+ }
+ unsigned int start = i;
+ unsigned int context = i;
+ while (context &&
+ !hb_in_range<unsigned> (info[context - 1].arabic_shaping_action(), STCH_FIXED, STCH_REPEATING) &&
+ (_hb_glyph_info_is_default_ignorable (&info[context - 1]) ||
+ HB_ARABIC_GENERAL_CATEGORY_IS_WORD (_hb_glyph_info_get_general_category (&info[context - 1]))))
+ {
+ context--;
+ w_total += pos[context].x_advance;
+ }
+ i++; // Don't touch i again.
+
+ DEBUG_MSG (ARABIC, nullptr, "%s stretch at (%d,%d,%d)",
+ step == MEASURE ? "measuring" : "cutting", context, start, end);
+ DEBUG_MSG (ARABIC, nullptr, "rest of word: count=%d width %d", start - context, w_total);
+ DEBUG_MSG (ARABIC, nullptr, "fixed tiles: count=%d width=%d", n_fixed, w_fixed);
+ DEBUG_MSG (ARABIC, nullptr, "repeating tiles: count=%d width=%d", n_repeating, w_repeating);
+
+ /* Number of additional times to repeat each repeating tile. */
+ int n_copies = 0;
+
+ hb_position_t w_remaining = w_total - w_fixed;
+ if (sign * w_remaining > sign * w_repeating && sign * w_repeating > 0)
+ n_copies = (sign * w_remaining) / (sign * w_repeating) - 1;
+
+ /* See if we can improve the fit by adding an extra repeat and squeezing them together a bit. */
+ hb_position_t extra_repeat_overlap = 0;
+ hb_position_t shortfall = sign * w_remaining - sign * w_repeating * (n_copies + 1);
+ if (shortfall > 0 && n_repeating > 0)
+ {
+ ++n_copies;
+ hb_position_t excess = (n_copies + 1) * sign * w_repeating - sign * w_remaining;
+ if (excess > 0)
+ extra_repeat_overlap = excess / (n_copies * n_repeating);
+ }
+
+ if (step == MEASURE)
+ {
+ extra_glyphs_needed += n_copies * n_repeating;
+ DEBUG_MSG (ARABIC, nullptr, "will add extra %d copies of repeating tiles", n_copies);
+ }
+ else
+ {
+ buffer->unsafe_to_break (context, end);
+ hb_position_t x_offset = 0;
+ for (unsigned int k = end; k > start; k--)
+ {
+ hb_position_t width = font->get_glyph_h_advance (info[k - 1].codepoint);
+
+ unsigned int repeat = 1;
+ if (info[k - 1].arabic_shaping_action() == STCH_REPEATING)
+ repeat += n_copies;
+
+ DEBUG_MSG (ARABIC, nullptr, "appending %d copies of glyph %d; j=%d",
+ repeat, info[k - 1].codepoint, j);
+ for (unsigned int n = 0; n < repeat; n++)
+ {
+ x_offset -= width;
+ if (n > 0)
+ x_offset += extra_repeat_overlap;
+ pos[k - 1].x_offset = x_offset;
+ /* Append copy. */
+ --j;
+ info[j] = info[k - 1];
+ pos[j] = pos[k - 1];
+ }
+ }
+ }
+ }
+
+ if (step == MEASURE)
+ {
+ if (unlikely (!buffer->ensure (count + extra_glyphs_needed)))
+ break;
+ }
+ else
+ {
+ assert (j == 0);
+ buffer->len = new_len;
+ }
}
}
+
static void
-setup_masks_arabic (const hb_ot_shape_plan_t *plan,
- hb_buffer_t *buffer,
- hb_font_t *font)
+postprocess_glyphs_arabic (const hb_ot_shape_plan_t *plan,
+ hb_buffer_t *buffer,
+ hb_font_t *font)
{
- const arabic_shape_plan_t *arabic_plan = (const arabic_shape_plan_t *) plan->data;
+ apply_stch (plan, buffer, font);
+
+ HB_BUFFER_DEALLOCATE_VAR (buffer, arabic_shaping_action);
+}
- if (likely (!arabic_plan->do_fallback))
+/* https://unicode.org/reports/tr53/tr53-1.pdf */
+
+static hb_codepoint_t
+modifier_combining_marks[] =
+{
+ 0x0654u, /* ARABIC HAMZA ABOVE */
+ 0x0655u, /* ARABIC HAMZA BELOW */
+ 0x0658u, /* ARABIC MARK NOON GHUNNA */
+ 0x06DCu, /* ARABIC SMALL HIGH SEEN */
+ 0x06E3u, /* ARABIC SMALL LOW SEEN */
+ 0x06E7u, /* ARABIC SMALL HIGH YEH */
+ 0x06E8u, /* ARABIC SMALL HIGH NOON */
+ 0x08F3u, /* ARABIC SMALL HIGH WAW */
+};
+
+static inline bool
+info_is_mcm (const hb_glyph_info_t &info)
+{
+ hb_codepoint_t u = info.codepoint;
+ for (unsigned int i = 0; i < ARRAY_LENGTH (modifier_combining_marks); i++)
+ if (u == modifier_combining_marks[i])
+ return true;
+ return false;
+}
+
+static void
+reorder_marks_arabic (const hb_ot_shape_plan_t *plan,
+ hb_buffer_t *buffer,
+ unsigned int start,
+ unsigned int end)
+{
+ hb_glyph_info_t *info = buffer->info;
+
+ DEBUG_MSG (ARABIC, buffer, "Reordering marks from %d to %d", start, end);
+
+ unsigned int i = start;
+ for (unsigned int cc = 220; cc <= 230; cc += 10)
{
- arabic_joining (buffer);
- unsigned int count = buffer->len;
- for (unsigned int i = 0; i < count; i++)
- buffer->info[i].mask |= arabic_plan->mask_array[buffer->info[i].arabic_shaping_action()];
+ DEBUG_MSG (ARABIC, buffer, "Looking for %d's starting at %d", cc, i);
+ while (i < end && info_cc(info[i]) < cc)
+ i++;
+ DEBUG_MSG (ARABIC, buffer, "Looking for %d's stopped at %d", cc, i);
+
+ if (i == end)
+ break;
+
+ if (info_cc(info[i]) > cc)
+ continue;
+
+ unsigned int j = i;
+ while (j < end && info_cc(info[j]) == cc && info_is_mcm (info[j]))
+ j++;
+
+ if (i == j)
+ continue;
+
+ DEBUG_MSG (ARABIC, buffer, "Found %d's from %d to %d", cc, i, j);
+
+ /* Shift it! */
+ DEBUG_MSG (ARABIC, buffer, "Shifting %d's: %d %d", cc, i, j);
+ hb_glyph_info_t temp[HB_OT_SHAPE_COMPLEX_MAX_COMBINING_MARKS];
+ assert (j - i <= ARRAY_LENGTH (temp));
+ buffer->merge_clusters (start, j);
+ memmove (temp, &info[i], (j - i) * sizeof (hb_glyph_info_t));
+ memmove (&info[start + j - i], &info[start], (i - start) * sizeof (hb_glyph_info_t));
+ memmove (&info[start], temp, (j - i) * sizeof (hb_glyph_info_t));
+
+ /* Renumber CC such that the reordered sequence is still sorted.
+ * 22 and 26 are chosen because they are smaller than all Arabic categories,
+ * and are folded back to 220/230 respectively during fallback mark positioning.
+ *
+ * We do this because the CGJ-handling logic in the normalizer relies on
+ * mark sequences having an increasing order even after this reordering.
+ * https://github.com/harfbuzz/harfbuzz/issues/554
+ * This, however, does break some obscure sequences, where the normalizer
+ * might compose a sequence that it should not. For example, in the seequence
+ * ALEF, HAMZAH, MADDAH, we should NOT try to compose ALEF+MADDAH, but with this
+ * renumbering, we will.
+ */
+ unsigned int new_start = start + j - i;
+ unsigned int new_cc = cc == 220 ? HB_MODIFIED_COMBINING_CLASS_CCC22 : HB_MODIFIED_COMBINING_CLASS_CCC26;
+ while (start < new_start)
+ {
+ _hb_glyph_info_set_modified_combining_class (&info[start], new_cc);
+ start++;
+ }
+
+ i = j;
}
}
const hb_ot_complex_shaper_t _hb_ot_complex_shaper_arabic =
{
- "arabic",
collect_features_arabic,
- NULL, /* override_features */
+ nullptr, /* override_features */
data_create_arabic,
data_destroy_arabic,
- preprocess_text_arabic,
- NULL, /* normalization_preference */
+ nullptr, /* preprocess_text */
+ postprocess_glyphs_arabic,
+ HB_OT_SHAPE_NORMALIZATION_MODE_DEFAULT,
+ nullptr, /* decompose */
+ nullptr, /* compose */
setup_masks_arabic,
- true, /* zero_width_attached_marks */
+ nullptr, /* disable_otl */
+ reorder_marks_arabic,
+ HB_OT_SHAPE_ZERO_WIDTH_MARKS_BY_GDEF_LATE,
+ true, /* fallback_position */
};