Harfbuzz-ng中的语言shaper
在harfbuzz-ng中，实际上存在两种层面的shaper，一种是比较宏观的字库shaper，这类shaper一般都针对于特定的字库文件类型来写，比如graphite2 shaper，ot shaper等。还有一种就是语言的shaper，这种shaper是harfbuzz-ng所实现的ot shaper的一个子部分，这种shaper一般都针对于特定的语言或script来写，比如印度语系的shaper，泰语/老挝语的shaper，阿拉伯语系的shaper等。harfbuzz-ng如何选择一个shaper一文中有对字体shaper的选择做过一个简单的说明。那harfbuzz-ng又是在何处来决定要采用哪一个语言shaper的呢？作出决定的依据是什么？语言shaper的结构中又有些什么内容呢？接下来，我们就来尝试解答这些问题。
在harfbuzz-ng如何选择一个shaper一文中我们有提过，harfbuzz-ng在创建shape_plan时，会通过一个函数创建一个shape_plan data，对于ot shaper而言，这个函数就是_hb_ot_shaper_shape_plan_data_create()，实际上，选择语言shaper的动作也正是在这个部分完成的。我们来看这个函数的定义：
hb_ot_shaper_shape_plan_data_t *
_hb_ot_shaper_shape_plan_data_create (hb_shape_plan_t *shape_plan,
const hb_feature_t *user_features,
unsigned int num_user_features)
{
hb_ot_shape_plan_t *plan = (hb_ot_shape_plan_t *) calloc (1, sizeof (hb_ot_shape_plan_t));
if (unlikely (!plan))
return NULL;
hb_ot_shape_planner_t planner (shape_plan);
planner.shaper = hb_ot_shape_complex_categorize (&planner);
hb_ot_shape_collect_features (&planner, &shape_plan->props, user_features, num_user_features);
planner.compile (*plan);
if (plan->shaper->data_create) {
plan->data = plan->shaper->data_create (plan);
if (unlikely (!plan->data))
return NULL;
}
return plan;
}
是在这个函数中，通过调用
hb_ot_shape_complex_categorize()函数来选定一个语言shaper的，该函数的定义如下：
static inline const hb_ot_complex_shaper_t *
hb_ot_shape_complex_categorize (const hb_ot_shape_planner_t *planner)
{
switch ((hb_tag_t) planner->props.script)
{
default:
return &_hb_ot_complex_shaper_default;
/* Unicode-1.1 additions */
case HB_SCRIPT_ARABIC:
case HB_SCRIPT_MONGOLIAN:
case HB_SCRIPT_SYRIAC:
/* Unicode-5.0 additions */
case HB_SCRIPT_NKO:
case HB_SCRIPT_PHAGS_PA:
/* Unicode-6.0 additions */
case HB_SCRIPT_MANDAIC:
/* For Arabic script, use the Arabic shaper even if no OT script tag was found.
* This is because we do fallback shaping for Arabic script (and not others). */
if (planner->map.chosen_script[0] != HB_OT_TAG_DEFAULT_SCRIPT ||
planner->props.script == HB_SCRIPT_ARABIC)
return &_hb_ot_complex_shaper_arabic;
else
return &_hb_ot_complex_shaper_default;
/* Unicode-1.1 additions */
case HB_SCRIPT_THAI:
case HB_SCRIPT_LAO:
return &_hb_ot_complex_shaper_thai;
/* ^--- Add new shapers here */
#if 0
/* Note:
*
* These disabled scripts are listed in ucd/IndicSyllabicCategory.txt, but according
* to Martin Hosken and Jonathan Kew do not require complex shaping.
*
* TODO We should automate figuring out which scripts do not need complex shaping
*
* TODO We currently keep data for these scripts in our indic table. Need to fix the
* generator to not do that.
*/
/* Simple? */
/* Unicode-3.2 additions */
case HB_SCRIPT_BUHID:
case HB_SCRIPT_HANUNOO:
/* Unicode-5.1 additions */
case HB_SCRIPT_SAURASHTRA:
/* Unicode-6.0 additions */
case HB_SCRIPT_BATAK:
case HB_SCRIPT_BRAHMI:
/* Simple */
/* Unicode-1.1 additions */
/* These have their own shaper now. */
case HB_SCRIPT_LAO:
case HB_SCRIPT_THAI:
/* Unicode-2.0 additions */
case HB_SCRIPT_TIBETAN:
/* Unicode-3.2 additions */
case HB_SCRIPT_TAGALOG:
case HB_SCRIPT_TAGBANWA:
/* Unicode-4.0 additions */
case HB_SCRIPT_LIMBU:
case HB_SCRIPT_TAI_LE:
/* Unicode-4.1 additions */
case HB_SCRIPT_KHAROSHTHI:
case HB_SCRIPT_SYLOTI_NAGRI:
/* Unicode-5.1 additions */
case HB_SCRIPT_KAYAH_LI:
/* Unicode-5.2 additions */
case HB_SCRIPT_TAI_VIET:
#endif
/* Unicode-1.1 additions */
case HB_SCRIPT_BENGALI:
case HB_SCRIPT_DEVANAGARI:
case HB_SCRIPT_GUJARATI:
case HB_SCRIPT_GURMUKHI:
case HB_SCRIPT_KANNADA:
case HB_SCRIPT_MALAYALAM:
case HB_SCRIPT_ORIYA:
case HB_SCRIPT_TAMIL:
case HB_SCRIPT_TELUGU:
/* Unicode-3.0 additions */
case HB_SCRIPT_SINHALA:
/* Unicode-4.1 additions */
case HB_SCRIPT_BUGINESE:
case HB_SCRIPT_NEW_TAI_LUE:
/* Unicode-5.0 additions */
case HB_SCRIPT_BALINESE:
/* Unicode-5.1 additions */
case HB_SCRIPT_CHAM:
case HB_SCRIPT_LEPCHA:
case HB_SCRIPT_REJANG:
case HB_SCRIPT_SUNDANESE:
/* Unicode-5.2 additions */
case HB_SCRIPT_JAVANESE:
case HB_SCRIPT_KAITHI:
case HB_SCRIPT_MEETEI_MAYEK:
case HB_SCRIPT_TAI_THAM:
/* Unicode-6.1 additions */
case HB_SCRIPT_CHAKMA:
case HB_SCRIPT_SHARADA:
case HB_SCRIPT_TAKRI:
/* Only use Indic shaper if the font has Indic tables. */
if (planner->map.found_script[0])
return &_hb_ot_complex_shaper_indic;
else
return &_hb_ot_complex_shaper_default;
case HB_SCRIPT_KHMER:
/* A number of Khmer fonts in the wild don't have a 'pref' feature,
* and as such won't shape properly via the Indic shaper;
* however, they typically have 'liga' / 'clig' features that implement
* the necessary "reordering" by means of ligature substitutions.
* So we send such pref-less fonts through the generic shaper instead. */
if (planner->map.found_script[0] &&
hb_ot_layout_language_find_feature (planner->face, HB_OT_TAG_GSUB,
planner->map.script_index[0],
planner->map.language_index[0],
HB_TAG ('p','r','e','f'),
NULL))
return &_hb_ot_complex_shaper_indic;
else
return &_hb_ot_complex_shaper_default;
case HB_SCRIPT_MYANMAR:
/* For Myanmar, we only want to use the Indic shaper if the "new" script
* tag is found. For "old" script tag we want to use the default shaper. */
if (planner->map.chosen_script[0] == HB_TAG ('m','y','m','2'))
return &_hb_ot_complex_shaper_indic;
else
return &_hb_ot_complex_shaper_default;
}
}
可以看到，这个函数选择语言shaper的依据就只有一个，那就是输入字串的script，这个函数用一个switch-case结构来选择一个语言shaper。在harfbuzz-0.9.12中，主要有如下的几种语言shaper：
- _hb_ot_complex_shaper_default
- _hb_ot_complex_shaper_arabic
- _hb_ot_complex_shaper_thai
- _hb_ot_complex_shaper_indic
在最新版本的harfbuzz中，有又添加一些新的语言shaper。
那语言shaper又是一个什么样的东西呢？我们来看hb_ot_complex_shaper_t结构的定义：
struct hb_ot_complex_shaper_t
{
char name[8];
/* collect_features()
* Called during shape_plan().
* Shapers should use plan->map to add their features and callbacks.
* May be NULL.
*/
void (*collect_features) (hb_ot_shape_planner_t *plan);
/* override_features()
* Called during shape_plan().
* Shapers should use plan->map to override features and add callbacks after
* common features are added.
* May be NULL.
*/
void (*override_features) (hb_ot_shape_planner_t *plan);
/* data_create()
* Called at the end of shape_plan().
* Whatever shapers return will be accessible through plan->data later.
* If NULL is returned, means a plan failure.
*/
void *(*data_create) (const hb_ot_shape_plan_t *plan);
/* data_destroy()
* Called when the shape_plan is being destroyed.
* plan->data is passed here for destruction.
* If NULL is returned, means a plan failure.
* May be NULL.
*/
void (*data_destroy) (void *data);
/* preprocess_text()
* Called during shape().
* Shapers can use to modify text before shaping starts.
* May be NULL.
*/
void (*preprocess_text) (const hb_ot_shape_plan_t *plan,
hb_buffer_t *buffer,
hb_font_t *font);
/* normalization_preference()
* Called during shape().
* May be NULL.
*/
hb_ot_shape_normalization_mode_t
(*normalization_preference) (const hb_segment_properties_t *props);
/* decompose()
* Called during shape()'s normalization.
* May be NULL.
*/
bool (*decompose) (const hb_ot_shape_normalize_context_t *c,
hb_codepoint_t ab,
hb_codepoint_t *a,
hb_codepoint_t *b);
/* compose()
* Called during shape()'s normalization.
* May be NULL.
*/
bool (*compose) (const hb_ot_shape_normalize_context_t *c,
hb_codepoint_t a,
hb_codepoint_t b,
hb_codepoint_t *ab);
/* setup_masks()
* Called during shape().
* Shapers should use map to get feature masks and set on buffer.
* Shapers may NOT modify characters.
* May be NULL.
*/
void (*setup_masks) (const hb_ot_shape_plan_t *plan,
hb_buffer_t *buffer,
hb_font_t *font);
bool zero_width_attached_marks;
bool fallback_position;
};
可以看到，这个结构里面主要就是提供了一些callback，以方便shape等过程在需要的时候来调用。这个结构的注释还是提供了比较多的信息，各个callback的作用都有说明，此处不再罗嗦。
Done.