forked from mirrors/qmk_userspace
Improve consistency in UNICODEMAP code and docs, update docs/understanding_qmk (#4774)
* Remove unused UNICODE(n) macro, update docs * Add note about max length of unicode_map to docs * QK_UNICODE_MAP → QK_UNICODEMAP * Refactor process_unicode_map * process_unicode_map → process_unicodemap This is done for consistency: UNICODEMAP is the method (hence UNICODEMAP_ENABLE, process_unicodemap), whereas unicode_map is the mapping table itself. * Update references and ordering in docs/understanding_qmk * Add additional note to docs/understanding_qmk * &unicode_map[index] → unicode_map + index This avoids the issue of the compiler sometimes complaining about the array index being out of range * Update docs/getting_started_make_guide * Update method sections in docs/feature_unicode
This commit is contained in:
parent
6e984a8b5e
commit
cd9262d7b2
8 changed files with 88 additions and 93 deletions
|
@ -18,8 +18,7 @@
|
|||
#include "process_unicode_common.h"
|
||||
|
||||
__attribute__((weak))
|
||||
const uint32_t PROGMEM unicode_map[] = {
|
||||
};
|
||||
const uint32_t PROGMEM unicode_map[] = {};
|
||||
|
||||
void register_hex32(uint32_t hex) {
|
||||
bool onzerostart = true;
|
||||
|
@ -42,26 +41,26 @@ void register_hex32(uint32_t hex) {
|
|||
}
|
||||
|
||||
__attribute__((weak))
|
||||
void unicode_map_input_error() {}
|
||||
void unicodemap_input_error() {}
|
||||
|
||||
bool process_unicode_map(uint16_t keycode, keyrecord_t *record) {
|
||||
uint8_t input_mode = get_unicode_input_mode();
|
||||
if ((keycode & QK_UNICODE_MAP) == QK_UNICODE_MAP && record->event.pressed) {
|
||||
const uint32_t* map = unicode_map;
|
||||
uint16_t index = keycode - QK_UNICODE_MAP;
|
||||
uint32_t code = pgm_read_dword(&map[index]);
|
||||
if (code > 0xFFFF && code <= 0x10ffff && input_mode == UC_OSX) {
|
||||
bool process_unicodemap(uint16_t keycode, keyrecord_t *record) {
|
||||
if ((keycode & QK_UNICODEMAP) == QK_UNICODEMAP && record->event.pressed) {
|
||||
uint16_t index = keycode - QK_UNICODEMAP;
|
||||
uint32_t code = pgm_read_dword(unicode_map + index);
|
||||
uint8_t input_mode = get_unicode_input_mode();
|
||||
|
||||
if (code > 0xFFFF && code <= 0x10FFFF && input_mode == UC_OSX) {
|
||||
// Convert to UTF-16 surrogate pair
|
||||
code -= 0x10000;
|
||||
uint32_t lo = code & 0x3ff;
|
||||
uint32_t hi = (code & 0xffc00) >> 10;
|
||||
uint32_t lo = code & 0x3FF, hi = (code & 0xFFC00) >> 10;
|
||||
|
||||
unicode_input_start();
|
||||
register_hex32(hi + 0xd800);
|
||||
register_hex32(lo + 0xdc00);
|
||||
register_hex32(hi + 0xD800);
|
||||
register_hex32(lo + 0xDC00);
|
||||
unicode_input_finish();
|
||||
} else if ((code > 0x10ffff && input_mode == UC_OSX) || (code > 0xFFFFF && input_mode == UC_LNX)) {
|
||||
// when character is out of range supported by the OS
|
||||
unicode_map_input_error();
|
||||
} else if ((code > 0x10FFFF && input_mode == UC_OSX) || (code > 0xFFFFF && input_mode == UC_LNX)) {
|
||||
// Character is out of range supported by the OS
|
||||
unicodemap_input_error();
|
||||
} else {
|
||||
unicode_input_start();
|
||||
register_hex32(code);
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue