summaryrefslogtreecommitdiff
path: root/src/blocks.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/blocks.c')
-rw-r--r--src/blocks.c92
1 files changed, 69 insertions, 23 deletions
diff --git a/src/blocks.c b/src/blocks.c
index 8c3e18c..06f6dcb 100644
--- a/src/blocks.c
+++ b/src/blocks.c
@@ -16,6 +16,8 @@
#include "debug.h"
#define CODE_INDENT 4
+#define TAB_STOP 4
+
#define peek_at(i, n) (i)->data[n]
static inline bool
@@ -70,7 +72,9 @@ cmark_parser *cmark_parser_new(int options)
parser->current = document;
parser->line_number = 0;
parser->offset = 0;
+ parser->column = 0;
parser->first_nonspace = 0;
+ parser->first_nonspace_column = 0;
parser->indent = 0;
parser->blank = false;
parser->curline = line;
@@ -555,16 +559,53 @@ static void chop_trailing_hashtags(cmark_chunk *ch)
static void
S_find_first_nonspace(cmark_parser *parser, cmark_chunk *input)
{
+ char c;
+ int chars_to_tab = TAB_STOP - (parser->column % TAB_STOP);
+
parser->first_nonspace = parser->offset;
- while (peek_at(input, parser->first_nonspace) == ' ') {
- parser->first_nonspace++;
+ parser->first_nonspace_column = parser->column;
+ while ((c = peek_at(input, parser->first_nonspace))) {
+ if (c == ' ') {
+ parser->first_nonspace += 1;
+ parser->first_nonspace_column += 1;
+ chars_to_tab = chars_to_tab - 1;
+ if (chars_to_tab == 0) {
+ chars_to_tab = TAB_STOP;
+ }
+ } else if (c == '\t') {
+ parser->first_nonspace += 1;
+ parser->first_nonspace_column += chars_to_tab;
+ chars_to_tab = TAB_STOP;
+ } else {
+ break;
+ }
}
- parser->indent = parser->first_nonspace - parser->offset;
+ parser->indent = parser->first_nonspace_column - parser->column;
parser->blank = S_is_line_end_char(peek_at(input, parser->first_nonspace));
}
static void
+S_advance_offset(cmark_parser *parser, cmark_chunk *input, bufsize_t count, bool columns)
+{
+ char c;
+ int chars_to_tab;
+ while (count > 0 && (c = peek_at(input, parser->offset))) {
+ if (c == '\t') {
+ chars_to_tab = 4 - (parser->column % TAB_STOP);
+ parser->column += chars_to_tab;
+ parser->offset += 1;
+ count -= (columns ? chars_to_tab : 1);
+ } else {
+ parser->offset += 1;
+ parser->column += 1; // assume ascii; block starts are ascii
+ count -= 1;
+ }
+ }
+}
+
+
+static void
S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t bytes)
{
cmark_node* last_matched_container;
@@ -578,8 +619,9 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
cmark_chunk input;
bool maybe_lazy;
- utf8proc_detab(parser->curline, buffer, bytes);
+ cmark_strbuf_put(parser->curline, buffer, bytes);
parser->offset = 0;
+ parser->column = 0;
parser->blank = false;
input.data = parser->curline->ptr;
@@ -601,7 +643,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
if (container->type == NODE_BLOCK_QUOTE) {
matched = parser->indent <= 3 && peek_at(&input, parser->first_nonspace) == '>';
if (matched) {
- parser->offset = parser->first_nonspace + 1;
+ S_advance_offset(parser, &input, parser->indent + 1, true);
if (peek_at(&input, parser->offset) == ' ')
parser->offset++;
} else {
@@ -609,13 +651,14 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
}
} else if (container->type == NODE_ITEM) {
-
if (parser->indent >= container->as.list.marker_offset +
container->as.list.padding) {
- parser->offset += container->as.list.marker_offset +
- container->as.list.padding;
+ S_advance_offset(parser, &input,
+ container->as.list.marker_offset +
+ container->as.list.padding, true);
} else if (parser->blank) {
- parser->offset = parser->first_nonspace;
+ S_advance_offset(parser, &input,
+ parser->first_nonspace - parser->offset, false);
} else {
all_matched = false;
}
@@ -624,9 +667,11 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
if (!container->as.code.fenced) { // indented
if (parser->indent >= CODE_INDENT) {
- parser->offset += CODE_INDENT;
+ S_advance_offset(parser, &input, CODE_INDENT, true);
} else if (parser->blank) {
- parser->offset = parser->first_nonspace;
+ S_advance_offset(parser, &input,
+ parser->first_nonspace - parser->offset,
+ false);
} else {
all_matched = false;
}
@@ -642,7 +687,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
// closing fence - and since we're at
// the end of a line, we can return:
all_matched = false;
- parser->offset += matched;
+ S_advance_offset(parser, &input, matched, false);
parser->current = finalize(parser, container);
goto finished;
} else {
@@ -650,7 +695,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
i = container->as.code.fence_offset;
while (i > 0 &&
peek_at(&input, parser->offset) == ' ') {
- parser->offset++;
+ S_advance_offset(parser, &input, 1, false);
i--;
}
}
@@ -697,15 +742,16 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
if (!indented && peek_at(&input, parser->first_nonspace) == '>') {
- parser->offset = parser->first_nonspace + 1;
+ S_advance_offset(parser, &input, parser->first_nonspace + 1 - parser->offset, false);
// optional following character
if (peek_at(&input, parser->offset) == ' ')
- parser->offset++;
+ S_advance_offset(parser, &input, 1, false);
container = add_child(parser, container, NODE_BLOCK_QUOTE, parser->offset + 1);
} else if (!indented && (matched = scan_atx_header_start(&input, parser->first_nonspace))) {
- parser->offset = parser->first_nonspace + matched;
+ S_advance_offset(parser, &input,
+ parser->first_nonspace + matched - parser->offset, false);
container = add_child(parser, container, NODE_HEADER, parser->offset + 1);
bufsize_t hashpos = cmark_chunk_strchr(&input, '#', parser->first_nonspace);
@@ -726,7 +772,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
container->as.code.fence_length = matched;
container->as.code.fence_offset = parser->first_nonspace - parser->offset;
container->as.code.info = cmark_chunk_literal("");
- parser->offset = parser->first_nonspace + matched;
+ S_advance_offset(parser, &input, parser->first_nonspace + matched - parser->offset, false);
} else if (!indented && (matched = scan_html_block_tag(&input, parser->first_nonspace))) {
@@ -743,7 +789,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
container->type = NODE_HEADER;
container->as.header.level = lev;
container->as.header.setext = true;
- parser->offset = input.len - 1;
+ S_advance_offset(parser, &input, input.len - 1 - parser->offset, false);
} else if (!indented &&
!(container->type == NODE_PARAGRAPH &&
@@ -753,7 +799,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
// it's only now that we know the line is not part of a setext header:
container = add_child(parser, container, NODE_HRULE, parser->first_nonspace + 1);
container = finalize(parser, container);
- parser->offset = input.len - 1;
+ S_advance_offset(parser, &input, input.len - 1 - parser->offset, false);
} else if ((matched = parse_list_marker(&input, parser->first_nonspace, &data)) &&
(!indented || container->type == NODE_LIST)) {
@@ -761,7 +807,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
// spaces indent, as long as the list container is still open.
// compute padding:
- parser->offset = parser->first_nonspace + matched;
+ S_advance_offset(parser, &input, parser->first_nonspace + matched - parser->offset, false);
i = 0;
while (i <= 5 && peek_at(&input, parser->offset + i) == ' ') {
i++;
@@ -771,11 +817,11 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
S_is_line_end_char(peek_at(&input, parser->offset))) {
data->padding = matched + 1;
if (i > 0) {
- parser->offset += 1;
+ S_advance_offset(parser, &input, 1, false);
}
} else {
data->padding = matched + i;
- parser->offset += i;
+ S_advance_offset(parser, &input, i, true);
}
// check container; if it's a list, see if this list item
@@ -799,7 +845,7 @@ S_process_line(cmark_parser *parser, const unsigned char *buffer, bufsize_t byte
free(data);
} else if (indented && !maybe_lazy && !parser->blank) {
- parser->offset += CODE_INDENT;
+ S_advance_offset(parser, &input, CODE_INDENT, true);
container = add_child(parser, container, NODE_CODE_BLOCK, parser->offset + 1);
container->as.code.fenced = false;
container->as.code.fence_char = 0;