summaryrefslogtreecommitdiff
path: root/buffer.c
diff options
context:
space:
mode:
Diffstat (limited to 'buffer.c')
-rw-r--r--buffer.c48
1 files changed, 23 insertions, 25 deletions
diff --git a/buffer.c b/buffer.c
index 03d9b9e..41ad74c 100644
--- a/buffer.c
+++ b/buffer.c
@@ -133,7 +133,7 @@ char32_t buffer_char_at_pos(TextBuffer *buffer, BufferPos p) {
// invalid (col too large)
return 0;
} else {
- return U'\n';
+ return '\n';
}
}
@@ -200,7 +200,7 @@ size_t buffer_get_text_at_pos(TextBuffer *buffer, BufferPos pos, char32_t *text,
if (p) {
memcpy(p, line->str + index, chars_from_this_line * sizeof *p);
p += chars_from_this_line;
- *p++ = U'\n';
+ *p++ = '\n';
}
chars_left -= chars_from_this_line+1;
}
@@ -355,7 +355,7 @@ static void buffer_edit_print(BufferEdit *edit) {
printf(" (" U32_FMT " chars): ", edit->prev_len);
for (size_t i = 0; i < edit->prev_len; ++i) {
char32_t c = edit->prev_text[i];
- if (c == U'\n')
+ if (c == '\n')
printf("\\n");
else
printf("%lc", (wint_t)c);
@@ -547,22 +547,19 @@ Status buffer_load_file(TextBuffer *buffer, char const *filename) {
size_t bytes_read = fread(file_contents, 1, file_size, fp);
if (bytes_read == file_size) {
char32_t c = 0;
- mbstate_t mbstate = {0};
for (u8 *p = file_contents, *end = p + file_size; p != end; ) {
if (*p == '\r' && p != end-1 && p[1] == '\n') {
// CRLF line endings
p += 2;
- c = U'\n';
+ c = '\n';
} else {
- size_t n = mbrtoc32(&c, (char *)p, (size_t)(end - p), &mbstate);
+ size_t n = unicode_utf8_to_utf32(&c, (char *)p, (size_t)(end - p));
if (n == 0) {
// null character
c = 0;
++p;
- } else if (n == (size_t)(-3)) {
- // no bytes consumed, but a character was produced
- } else if (n == (size_t)(-2) || n == (size_t)(-1)) {
- // incomplete character at end of file or invalid UTF-8 respectively; fail
+ } else if (n == (size_t)(-1)) {
+ // invalid UTF-8
success = false;
buffer_seterr(buffer, "Invalid UTF-8 (position: %td).", p - file_contents);
break;
@@ -570,7 +567,7 @@ Status buffer_load_file(TextBuffer *buffer, char const *filename) {
p += n;
}
}
- if (c == U'\n') {
+ if (c == '\n') {
if (buffer_lines_set_min_capacity(buffer, &lines, &lines_capacity, nlines + 1))
++nlines;
} else {
@@ -634,11 +631,12 @@ bool buffer_save(TextBuffer *buffer) {
if (out) {
bool success = true;
for (Line *line = buffer->lines, *end = line + buffer->nlines; line != end; ++line) {
- mbstate_t state = {0};
for (char32_t *p = line->str, *p_end = p + line->len; p != p_end; ++p) {
- char utf8[MB_LEN_MAX] = {0};
- size_t bytes = c32rtomb(utf8, *p, &state);
- fwrite(utf8, 1, bytes, out);
+ char utf8[4] = {0};
+ size_t bytes = unicode_utf32_to_utf8(utf8, *p);
+ if (bytes != (size_t)-1) {
+ fwrite(utf8, 1, bytes, out);
+ }
}
if (line != end-1) {
@@ -694,7 +692,7 @@ static u32 buffer_index_to_column(TextBuffer *buffer, u32 line, u32 index) {
uint tab_width = buffer_settings(buffer)->tab_width;
for (u32 i = 0; i < index; ++i) {
switch (str[i]) {
- case U'\t': {
+ case '\t': {
do
++col;
while (col % tab_width);
@@ -718,7 +716,7 @@ static u32 buffer_column_to_index(TextBuffer *buffer, u32 line, u32 column) {
uint tab_width = buffer_settings(buffer)->tab_width;
for (u32 i = 0; i < len; ++i) {
switch (str[i]) {
- case U'\t': {
+ case '\t': {
do {
if (col == column)
return i;
@@ -1059,7 +1057,7 @@ i64 buffer_cursor_move_down(TextBuffer *buffer, i64 by) {
// Is this character a "word" character?
// This determines how buffer_pos_move_words (i.e. ctrl+left/right) works
static bool is_word(char32_t c) {
- return c > WCHAR_MAX || c == U'_' || iswalnum((wint_t)c);
+ return c > WCHAR_MAX || c == '_' || iswalnum((wint_t)c);
}
static bool is_space(char32_t c) {
@@ -1229,7 +1227,7 @@ BufferPos buffer_insert_text_at_pos(TextBuffer *buffer, BufferPos pos, String32
if (buffer->is_line_buffer) {
// remove all the newlines from str.
- str32_remove_all_instances_of_char(&str, U'\n');
+ str32_remove_all_instances_of_char(&str, '\n');
}
if (str.len == 0) {
@@ -1260,7 +1258,7 @@ BufferPos buffer_insert_text_at_pos(TextBuffer *buffer, BufferPos pos, String32
// `text` could consist of multiple lines, e.g. U"line 1\nline 2",
// so we need to go through them one by one
- u32 n_added_lines = (u32)str32_count_char(str, U'\n');
+ u32 n_added_lines = (u32)str32_count_char(str, '\n');
if (n_added_lines) {
if (buffer_insert_lines(buffer, line_idx + 1, n_added_lines)) {
line = &buffer->lines[line_idx]; // fix pointer
@@ -1279,7 +1277,7 @@ BufferPos buffer_insert_text_at_pos(TextBuffer *buffer, BufferPos pos, String32
while (str.len) {
- u32 text_line_len = (u32)str32chr(str, U'\n');
+ u32 text_line_len = (u32)str32chr(str, '\n');
u32 old_len = line->len;
u32 new_len = old_len + text_line_len;
if (new_len > old_len) { // handles both overflow and empty text lines
@@ -1947,12 +1945,12 @@ void buffer_render(TextBuffer *buffer, float x1, float y1, float x2, float y2) {
char32_t c = *p;
switch (c) {
- case U'\n': assert(0); break;
- case U'\r': break; // for CRLF line endings
- case U'\t': {
+ case '\n': assert(0); break;
+ case '\r': break; // for CRLF line endings
+ case '\t': {
uint tab_width = settings->tab_width;
do {
- text_render_char(font, &text_state, U' ');
+ text_render_char(font, &text_state, ' ');
++column;
} while (column % tab_width);
} break;