harfbuzz/src/hb-ot-shape-complex-hangul.cc

427 lines
12 KiB
C++
Raw Normal View History

/*
* Copyright © 2013 Google, Inc.
*
* This is part of HarfBuzz, a text shaping library.
*
* Permission is hereby granted, without written agreement and without
* license or royalty fees, to use, copy, modify, and distribute this
* software and its documentation for any purpose, provided that the
* above copyright notice and the following two paragraphs appear in
* all copies of this software.
*
* IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE TO ANY PARTY FOR
* DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES
* ARISING OUT OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN
* IF THE COPYRIGHT HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH
* DAMAGE.
*
* THE COPYRIGHT HOLDER SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
* BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
* FITNESS FOR A PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS
* ON AN "AS IS" BASIS, AND THE COPYRIGHT HOLDER HAS NO OBLIGATION TO
* PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS.
*
* Google Author(s): Behdad Esfahbod
*/
#include "hb-ot-shape-complex-private.hh"
/* Hangul shaper */
/* Same order as the feature array below */
enum {
NONE,
LJMO,
VJMO,
TJMO,
FIRST_HANGUL_FEATURE = LJMO,
HANGUL_FEATURE_COUNT = TJMO + 1
};
static const hb_tag_t hangul_features[HANGUL_FEATURE_COUNT] =
{
HB_TAG_NONE,
HB_TAG('l','j','m','o'),
HB_TAG('v','j','m','o'),
HB_TAG('t','j','m','o')
};
static void
collect_features_hangul (hb_ot_shape_planner_t *plan)
{
hb_ot_map_builder_t *map = &plan->map;
for (unsigned int i = FIRST_HANGUL_FEATURE; i < HANGUL_FEATURE_COUNT; i++)
map->add_feature (hangul_features[i], 1, F_NONE);
}
static void
override_features_hangul (hb_ot_shape_planner_t *plan)
{
/* Uniscribe does not apply 'calt' for Hangul, and certain fonts
* (Noto Sans CJK, Source Sans Han, etc) apply all of jamo lookups
* in calt, which is not desirable. */
plan->map.add_feature (HB_TAG('c','a','l','t'), 0, F_GLOBAL);
}
struct hangul_shape_plan_t
{
ASSERT_POD ();
hb_mask_t mask_array[HANGUL_FEATURE_COUNT];
};
static void *
data_create_hangul (const hb_ot_shape_plan_t *plan)
{
hangul_shape_plan_t *hangul_plan = (hangul_shape_plan_t *) calloc (1, sizeof (hangul_shape_plan_t));
if (unlikely (!hangul_plan))
return NULL;
for (unsigned int i = 0; i < HANGUL_FEATURE_COUNT; i++)
hangul_plan->mask_array[i] = plan->map.get_1_mask (hangul_features[i]);
return hangul_plan;
}
static void
data_destroy_hangul (void *data)
{
free (data);
}
/* Constants for algorithmic hangul syllable [de]composition. */
#define LBase 0x1100u
#define VBase 0x1161u
#define TBase 0x11A7u
#define LCount 19u
#define VCount 21u
#define TCount 28u
#define SBase 0xAC00u
#define NCount (VCount * TCount)
#define SCount (LCount * NCount)
#define isCombiningL(u) (hb_in_range ((u), LBase, LBase+LCount-1))
#define isCombiningV(u) (hb_in_range ((u), VBase, VBase+VCount-1))
#define isCombiningT(u) (hb_in_range ((u), TBase+1, TBase+TCount-1))
#define isCombinedS(u) (hb_in_range ((u), SBase, SBase+SCount-1))
#define isL(u) (hb_in_ranges ((u), 0x1100u, 0x115Fu, 0xA960u, 0xA97Cu))
#define isV(u) (hb_in_ranges ((u), 0x1160u, 0x11A7u, 0xD7B0u, 0xD7C6u))
#define isT(u) (hb_in_ranges ((u), 0x11A8u, 0x11FFu, 0xD7CBu, 0xD7FBu))
#define isHangulTone(u) (hb_in_range ((u), 0x302Eu, 0x302Fu))
/* buffer var allocations */
#define hangul_shaping_feature() complex_var_u8_0() /* hangul jamo shaping feature */
static bool
is_zero_width_char (hb_font_t *font,
hb_codepoint_t unicode)
{
hb_codepoint_t glyph;
return hb_font_get_glyph (font, unicode, 0, &glyph) && hb_font_get_glyph_h_advance (font, glyph) == 0;
}
static void
preprocess_text_hangul (const hb_ot_shape_plan_t *plan,
hb_buffer_t *buffer,
hb_font_t *font)
{
HB_BUFFER_ALLOCATE_VAR (buffer, hangul_shaping_feature);
/* Hangul syllables come in two shapes: LV, and LVT. Of those:
*
* - LV can be precomposed, or decomposed. Lets call those
* <LV> and <L,V>,
* - LVT can be fully precomposed, partically precomposed, or
* fully decomposed. Ie. <LVT>, <LV,T>, or <L,V,T>.
*
* The composition / decomposition is mechanical. However, not
* all <L,V> sequences compose, and not all <LV,T> sequences
* compose.
*
* Here are the specifics:
*
* - <L>: U+1100..115F, U+A960..A97F
* - <V>: U+1160..11A7, U+D7B0..D7C7
* - <T>: U+11A8..11FF, U+D7CB..D7FB
*
* - Only the <L,V> sequences for the 11xx ranges combine.
* - Only <LV,T> sequences for T in U+11A8..11C3 combine.
*
* Here is what we want to accomplish in this shaper:
*
* - If the whole syllable can be precomposed, do that,
* - Otherwise, fully decompose and apply ljmo/vjmo/tjmo features.
* - If a valid syllable is followed by a Hangul tone mark, reorder the tone
* mark to precede the whole syllable - unless it is a zero-width glyph, in
* which case we leave it untouched, assuming it's designed to overstrike.
*
* That is, of the different possible syllables:
*
* <L>
* <L,V>
* <L,V,T>
* <LV>
* <LVT>
* <LV, T>
*
* - <L> needs no work.
*
* - <LV> and <LVT> can stay the way they are if the font supports them, otherwise we
* should fully decompose them if font supports.
*
* - <L,V> and <L,V,T> we should compose if the whole thing can be composed.
*
* - <LV,T> we should compose if the whole thing can be composed, otherwise we should
* decompose.
*/
buffer->clear_output ();
unsigned int start = 0, end = 0; /* Extent of most recently seen syllable;
* valid only if start < end
*/
unsigned int count = buffer->len;
for (buffer->idx = 0; buffer->idx < count;)
{
hb_codepoint_t u = buffer->cur().codepoint;
if (isHangulTone (u))
{
/*
* We could cache the width of the tone marks and the existence of dotted-circle,
* but the use of the Hangul tone mark characters seems to be rare enough that
* I didn't bother for now.
*/
if (start < end && end == buffer->out_len)
{
/* Tone mark follows a valid syllable; move it in front, unless it's zero width. */
buffer->next_glyph ();
if (!is_zero_width_char (font, u))
{
hb_glyph_info_t *info = buffer->out_info;
hb_glyph_info_t tone = info[end];
memmove (&info[start + 1], &info[start], (end - start) * sizeof (hb_glyph_info_t));
info[start] = tone;
}
/* Merge clusters across the (possibly reordered) syllable+tone.
* We want to merge even in the zero-width tone mark case here,
* so that clustering behavior isn't dependent on how the tone mark
* is handled by the font.
*/
buffer->merge_out_clusters (start, end + 1);
}
else
{
/* No valid syllable as base for tone mark; try to insert dotted circle. */
if (font->has_glyph (0x25CCu))
{
hb_codepoint_t chars[2];
if (!is_zero_width_char (font, u)) {
chars[0] = u;
chars[1] = 0x25CCu;
} else {
chars[0] = 0x25CCu;
chars[1] = u;
}
buffer->replace_glyphs (1, 2, chars);
}
else
{
/* No dotted circle available in the font; just leave tone mark untouched. */
buffer->next_glyph ();
}
}
start = end = buffer->out_len;
continue;
}
start = buffer->out_len; /* Remember current position as a potential syllable start;
* will only be used if we set end to a later position.
*/
if (isL (u) && buffer->idx + 1 < count)
{
hb_codepoint_t l = u;
hb_codepoint_t v = buffer->cur(+1).codepoint;
if (isV (v))
{
/* Have <L,V> or <L,V,T>. */
hb_codepoint_t t = 0;
unsigned int tindex = 0;
if (buffer->idx + 2 < count)
{
t = buffer->cur(+2).codepoint;
if (isT (t))
tindex = t - TBase; /* Only used if isCombiningT (t); otherwise invalid. */
else
t = 0; /* The next character was not a trailing jamo. */
}
/* We've got a syllable <L,V,T?>; see if it can potentially be composed. */
if (isCombiningL (l) && isCombiningV (v) && (t == 0 || isCombiningT (t)))
{
/* Try to compose; if this succeeds, end is set to start+1. */
hb_codepoint_t s = SBase + (l - LBase) * NCount + (v - VBase) * TCount + tindex;
2014-01-02 07:30:45 +01:00
if (font->has_glyph (s))
{
buffer->replace_glyphs (t ? 3 : 2, 1, &s);
if (unlikely (buffer->in_error))
return;
end = start + 1;
continue;
}
}
/* We didn't compose, either because it's an Old Hangul syllable without a
* precomposed character in Unicode, or because the font didn't support the
* necessary precomposed glyph.
* Set jamo features on the individual glyphs, and advance past them.
*/
buffer->cur().hangul_shaping_feature() = LJMO;
buffer->next_glyph ();
buffer->cur().hangul_shaping_feature() = VJMO;
buffer->next_glyph ();
if (t)
{
buffer->cur().hangul_shaping_feature() = TJMO;
buffer->next_glyph ();
end = start + 3;
}
else
end = start + 2;
buffer->merge_out_clusters (start, end);
continue;
}
}
else if (isCombinedS (u))
{
/* Have <LV>, <LVT>, or <LV,T> */
hb_codepoint_t s = u;
2014-01-02 07:30:45 +01:00
bool has_glyph = font->has_glyph (s);
unsigned int lindex = (s - SBase) / NCount;
unsigned int nindex = (s - SBase) % NCount;
unsigned int vindex = nindex / TCount;
unsigned int tindex = nindex % TCount;
if (!tindex &&
buffer->idx + 1 < count &&
isCombiningT (buffer->cur(+1).codepoint))
{
/* <LV,T>, try to combine. */
unsigned int new_tindex = buffer->cur(+1).codepoint - TBase;
hb_codepoint_t new_s = s + new_tindex;
if (font->has_glyph (new_s))
{
buffer->replace_glyphs (2, 1, &new_s);
if (unlikely (buffer->in_error))
return;
end = start + 1;
continue;
}
}
/* Otherwise, decompose if font doesn't support <LV> or <LVT>,
* or if having non-combining <LV,T>. Note that we already handled
* combining <LV,T> above. */
if (!has_glyph ||
(!tindex &&
buffer->idx + 1 < count &&
isT (buffer->cur(+1).codepoint)))
{
hb_codepoint_t decomposed[3] = {LBase + lindex,
VBase + vindex,
TBase + tindex};
if (font->has_glyph (decomposed[0]) &&
2014-01-02 07:30:45 +01:00
font->has_glyph (decomposed[1]) &&
(!tindex || font->has_glyph (decomposed[2])))
{
unsigned int s_len = tindex ? 3 : 2;
buffer->replace_glyphs (1, s_len, decomposed);
if (unlikely (buffer->in_error))
return;
/* We decomposed S: apply jamo features to the individual glyphs
* that are now in buffer->out_info.
*/
hb_glyph_info_t *info = buffer->out_info;
/* If we decomposed an LV because of a non-combining T following,
* we want to include this T in the syllable.
*/
if (has_glyph && !tindex)
{
buffer->next_glyph ();
s_len++;
}
end = start + s_len;
unsigned int i = start;
info[i++].hangul_shaping_feature() = LJMO;
info[i++].hangul_shaping_feature() = VJMO;
if (i < end)
info[i++].hangul_shaping_feature() = TJMO;
buffer->merge_out_clusters (start, end);
continue;
}
}
if (has_glyph)
{
/* We didn't decompose the S, so just advance past it. */
end = start + 1;
buffer->next_glyph ();
continue;
}
}
/* Didn't find a recognizable syllable, so we leave end <= start;
* this will prevent tone-mark reordering happening.
*/
buffer->next_glyph ();
}
buffer->swap_buffers ();
}
static void
setup_masks_hangul (const hb_ot_shape_plan_t *plan,
hb_buffer_t *buffer,
hb_font_t *font HB_UNUSED)
{
const hangul_shape_plan_t *hangul_plan = (const hangul_shape_plan_t *) plan->data;
if (likely (hangul_plan))
{
unsigned int count = buffer->len;
hb_glyph_info_t *info = buffer->info;
for (unsigned int i = 0; i < count; i++, info++)
info->mask |= hangul_plan->mask_array[info->hangul_shaping_feature()];
}
HB_BUFFER_DEALLOCATE_VAR (buffer, hangul_shaping_feature);
}
const hb_ot_complex_shaper_t _hb_ot_complex_shaper_hangul =
{
"hangul",
collect_features_hangul,
override_features_hangul,
data_create_hangul, /* data_create */
data_destroy_hangul, /* data_destroy */
preprocess_text_hangul,
HB_OT_SHAPE_NORMALIZATION_MODE_NONE,
NULL, /* decompose */
NULL, /* compose */
setup_masks_hangul, /* setup_masks */
HB_OT_SHAPE_ZERO_WIDTH_MARKS_NONE,
false, /* fallback_position */
};