path: root/src/inlines.c

#include <stdlib.h>
#include <stdio.h>
#include <stdbool.h>
#include <ctype.h>
#include <string.h>

#include "stmd.h"
#include "uthash.h"
#include "scanners.h"

typedef struct Subject {
  const gh_buf   *buffer;
  int            pos;
  reference**    reference_map;
  int            label_nestlevel;
} subject;

reference* lookup_reference(reference** refmap, chunk *label);
reference* make_reference(chunk *label, chunk *url, chunk *title);

static unsigned char *clean_url(chunk *url);
static unsigned char *clean_title(chunk *title);

inline static unsigned char *chunk_to_cstr(chunk *c);
inline static void chunk_free(chunk *c);
inline static void chunk_trim(chunk *c);

inline static chunk chunk_literal(const char *data);
inline static chunk chunk_buf_detach(gh_buf *buf);
inline static chunk chunk_buf(const gh_buf *buf, int pos, int len);

static inl *parse_chunk_inlines(chunk *chunk, reference** refmap);
static inl *parse_inlines_while(subject* subj, int (*f)(subject*));
static int parse_inline(subject* subj, inl ** last);

extern void free_reference(reference *ref) {
    free(ref->label);
    free(ref->url);
    free(ref->title);
    free(ref);
}

extern void free_reference_map(reference **refmap) {
    /* free the hash table contents */
    reference *s;
    reference *tmp;
    if (refmap != NULL) {
        HASH_ITER(hh, *refmap, s, tmp) {
            HASH_DEL(*refmap, s);
            free_reference(s);
        }
        free(refmap);
    }
}

// normalize reference:  collapse internal whitespace to single space,
// remove leading/trailing whitespace, case fold
static unsigned char *normalize_reference(chunk *ref)
{
    gh_buf normalized = GH_BUF_INIT;
    int r, w;

    utf8proc_case_fold(&normalized, ref->data, ref->len);
    gh_buf_trim(&normalized);

    for (r = 0, w = 0; r < normalized.size; ++r) {
        if (r && gh_buf_at(&normalized, r - 1) == ' ') {
            while (gh_buf_at(&normalized, r) == ' ')
                r++;
        }

        normalized.ptr[w++] = normalized.ptr[r];
    }

    return gh_buf_detach(&normalized);
}

// Returns reference if refmap contains a reference with matching
// label, otherwise NULL.
extern reference* lookup_reference(reference** refmap, chunk *label)
{
    reference *ref = NULL;
    unsigned char *norm = normalize_reference(label);
    if (refmap != NULL) {
        HASH_FIND_STR(*refmap, (char*)norm, ref);
    }
    free(label);
    return ref;
}

extern reference* make_reference(chunk *label, chunk *url, chunk *title)
{
    reference *ref;
    ref = malloc(sizeof(reference));
    ref->label = normalize_reference(label);
    ref->url = clean_url(url);
    ref->title = clean_title(title);
    return ref;
}

extern void add_reference(reference** refmap, reference* ref)
{
    reference * t = NULL;
    HASH_FIND(hh, *refmap, (char*)ref->label, (unsigned)strlen(ref->label), t);

    if (t == NULL) {
        HASH_ADD_KEYPTR(hh, *refmap, (char*)ref->label, (unsigned)strlen(ref->label), ref);
    } else {
        free_reference(ref);  // we free this now since it won't be in the refmap
    }
}

// Create an inline with a linkable string value.
inline static inl* make_linkable(int t, inl* label, chunk url, chunk title)
{
    inl* e = (inl*) malloc(sizeof(inl));
    e->tag = t;
    e->content.linkable.label = label;
    e->content.linkable.url   = chunk_to_cstr(&url);
    e->content.linkable.title = url.len ? chunk_to_cstr(&title) : NULL;
    e->next = NULL;
    return e;
}

inline static inl* make_inlines(int t, inl* contents)
{
    inl* e = (inl*) malloc(sizeof(inl));
    e->tag = t;
    e->content.inlines = contents;
    e->next = NULL;
    return e;
}

// Create an inline with a literal string value.
inline static inl* make_literal(int t, chunk s)
{
    inl* e = (inl*) malloc(sizeof(inl));
    e->tag = t;
    e->content.literal = s;
    e->next = NULL;
    return e;
}

// Create an inline with no value.
inline static inl* make_simple(int t)
{
    inl* e = (inl*) malloc(sizeof(inl));
    e->tag = t;
    e->next = NULL;
    return e;
}

// Macros for creating various kinds of inlines.
#define make_str(s) make_literal(INL_STRING, s)
#define make_code(s) make_literal(INL_CODE, s)
#define make_raw_html(s) make_literal(INL_RAW_HTML, s)
#define make_entity(s) make_literal(INL_ENTITY, s)
#define make_linebreak() make_simple(INL_LINEBREAK)
#define make_softbreak() make_simple(INL_SOFTBREAK)
#define make_link(label, url, title) make_linkable(INL_LINK, label, url, title)
#define make_emph(contents) make_inlines(INL_EMPH, contents)
#define make_strong(contents) make_inlines(INL_STRONG, contents)

// Free an inline list.
extern void free_inlines(inl* e)
{
    inl * next;
    while (e != NULL) {
        switch (e->tag){
            case INL_STRING:
            case INL_RAW_HTML:
            case INL_CODE:
            case INL_ENTITY:
                chunk_free(&e->content.literal);
                break;
            case INL_LINEBREAK:
            case INL_SOFTBREAK:
                break;
            case INL_LINK:
            case INL_IMAGE:
                free(e->content.linkable.url);
                free(e->content.linkable.title);
                free_inlines(e->content.linkable.label);
                break;
            case INL_EMPH:
            case INL_STRONG:
                free_inlines(e->content.inlines);
                break;
            default:
                break;
        }
        next = e->next;
        free(e);
        e = next;
    }
}

// Append inline list b to the end of inline list a.
// Return pointer to head of new list.
inline static inl* append_inlines(inl* a, inl* b)
{
    if (a == NULL) {  // NULL acts like an empty list
        return b;
    }
    inl* cur = a;
    while (cur->next) {
        cur = cur->next;
    }
    cur->next = b;
    return a;
}

// Make a 'subject' from an input string.
static void init_subject(subject *e, gh_buf *buffer, int input_pos, reference** refmap)
{
    e->buffer = buffer;
    e->pos = input_pos;
    e->label_nestlevel = 0;
    e->reference_map = refmap;
}

inline static int isbacktick(int c)
{
    return (c == '`');
}

inline static void chunk_free(chunk *c)
{
    if (c->alloc)
        free((char *)c->data);

    c->data = NULL;
    c->alloc = 0;
    c->len = 0;
}

inline static void chunk_trim(chunk *c)
{
    while (c->len && isspace(c->data[0])) {
        c->data++;
        c->len--;
    }

    while (c->len > 0) {
        if (!isspace(c->data[c->len - 1]))
            break;

        c->len--;
    }
}

inline static unsigned char *chunk_to_cstr(chunk *c)
{
    unsigned char *str;

    str = malloc(c->len + 1);
    memcpy(str, c->data, c->len);
    str[c->len] = 0;

    return str;
}

inline static chunk chunk_literal(const char *data)
{
    chunk c = {data, strlen(data), 0};
    return c;
}

inline static chunk chunk_buf(const gh_buf *buf, int pos, int len)
{
    chunk c = {buf->ptr + pos, len, 0};
    return c;
}

inline static chunk chunk_buf_detach(gh_buf *buf)
{
    chunk c;

    c.len = buf->size;
    c.data = gh_buf_detach(buf);
    c.alloc = 1;

    return c;
}

// Return the next character in the subject, without advancing.
// Return 0 if at the end of the subject.
#define peek_char(subj) gh_buf_at((subj)->buffer, (subj)->pos)

// Return true if there are more characters in the subject.
inline static int is_eof(subject* subj)
{
    return (subj->pos >= gh_buf_len(subj->buffer));
}

// Advance the subject.  Doesn't check for eof.
#define advance(subj) (subj)->pos += 1

// Take characters while a predicate holds, and return a string.
inline static chunk take_while(subject* subj, int (*f)(int))
{
    unsigned char c;
    int startpos = subj->pos;
    int len = 0;

    while ((c = peek_char(subj)) && (*f)(c)) {
        advance(subj);
        len++;
    }

    return chunk_buf(subj->buffer, startpos, len);
}

// Try to process a backtick code span that began with a
// span of ticks of length openticklength length (already
// parsed).  Return 0 if you don't find matching closing
// backticks, otherwise return the position in the subject
// after the closing backticks.
static int scan_to_closing_backticks(subject* subj, int openticklength)
{
    // read non backticks
    char c;
    while ((c = peek_char(subj)) && c != '`') {
        advance(subj);
    }
    if (is_eof(subj)) {
        return 0;  // did not find closing ticks, return 0
    }
    int numticks = 0;
    while (peek_char(subj) == '`') {
        advance(subj);
        numticks++;
    }
    if (numticks != openticklength){
        return(scan_to_closing_backticks(subj, openticklength));
    }
    return (subj->pos);
}

// Destructively modify string, collapsing consecutive
// space and newline characters into a single space.
static void normalize_whitespace(gh_buf *s)
{
    /* TODO */
#if 0
    bool last_char_was_space = false;
    int pos = 0;
    char c;
    while ((c = gh_buf_at(s, pos))) {
        switch (c) {
            case ' ':
                if (last_char_was_space) {
                    bdelete(s, pos, 1);
                } else {
                    pos++;
                }
                last_char_was_space = true;
                break;
            case '\n':
                if (last_char_was_space) {
                    bdelete(s, pos, 1);
                } else {
                    bdelete(s, pos, 1);
                    binsertch(s, pos, 1, ' ');
                    pos++;
                }
                last_char_was_space = true;
                break;
            default:
                pos++;
                last_char_was_space = false;
        }
    }
#endif
}

// Parse backtick code section or raw backticks, return an inline.
// Assumes that the subject has a backtick at the current position.
static inl* handle_backticks(subject *subj)
{
    chunk openticks = take_while(subj, isbacktick);
    int startpos = subj->pos;
    int endpos = scan_to_closing_backticks(subj, openticks.len);

    if (endpos == 0) { // not found
        subj->pos = startpos; // rewind
        return make_str(openticks);
    } else {
        gh_buf buf = GH_BUF_INIT;

        gh_buf_set(&buf, subj->buffer->ptr + startpos, endpos - startpos - openticks.len);
        gh_buf_trim(&buf);
        normalize_whitespace(&buf);

        return make_code(chunk_buf_detach(&buf));
    }
}

// Scan ***, **, or * and return number scanned, or 0.
// Don't advance position.
static int scan_delims(subject* subj, char c, bool * can_open, bool * can_close)
{
    int numdelims = 0;
    char char_before, char_after;
    int startpos = subj->pos;

    char_before = subj->pos == 0 ? '\n' : gh_buf_at(subj->buffer, subj->pos - 1);
    while (peek_char(subj) == c) {
        numdelims++;
        advance(subj);
    }
    char_after = peek_char(subj);
    *can_open = numdelims > 0 && numdelims <= 3 && !isspace(char_after);
    *can_close = numdelims > 0 && numdelims <= 3 && !isspace(char_before);
    if (c == '_') {
        *can_open = *can_open && !isalnum(char_before);
        *can_close = *can_close && !isalnum(char_after);
    }
    subj->pos = startpos;
    return numdelims;
}

// Parse strong/emph or a fallback.
// Assumes the subject has '_' or '*' at the current position.
static inl* handle_strong_emph(subject* subj, char c)
{
    bool can_open, can_close;
    inl * result = NULL;
    inl ** last = malloc(sizeof(inl *));
    inl * new;
    inl * il;
    inl * first_head = NULL;
    inl * first_close = NULL;
    int first_close_delims = 0;
    int numdelims;

    *last = NULL;

    numdelims = scan_delims(subj, c, &can_open, &can_close);
    subj->pos += numdelims;

    new = make_str(chunk_buf(subj->buffer, subj->pos - numdelims, numdelims));
    *last = new;
    first_head = new;
    result = new;

    if (!can_open || numdelims == 0) {
        goto done;
    }

    switch (numdelims) {
        case 1:
            while (true) {
                numdelims = scan_delims(subj, c, &can_open, &can_close);
                if (numdelims >= 1 && can_close) {
                    subj->pos += 1;
                    first_head->tag = INL_EMPH;
                    chunk_free(&first_head->content.literal);
                    first_head->content.inlines = first_head->next;
                    first_head->next = NULL;
                    goto done;
                } else {
                    if (!parse_inline(subj, last)) {
                        goto done;
                    }
                }
            }
            break;
        case 2:
            while (true) {
                numdelims = scan_delims(subj, c, &can_open, &can_close);
                if (numdelims >= 2 && can_close) {
                    subj->pos += 2;
                    first_head->tag = INL_STRONG;
                    chunk_free(&first_head->content.literal);
                    first_head->content.inlines = first_head->next;
                    first_head->next = NULL;
                    goto done;
                } else {
                    if (!parse_inline(subj, last)) {
                        goto done;
                    }
                }
            }
            break;
        case 3:
            while (true) {
                numdelims = scan_delims(subj, c, &can_open, &can_close);
                if (can_close && numdelims >= 1 && numdelims <= 3 &&
                        numdelims != first_close_delims) {
                    new = make_str(chunk_buf(subj->buffer, subj->pos, numdelims));
                    append_inlines(*last, new);
                    *last = new;
                    if (first_close_delims == 1 && numdelims > 2) {
                        numdelims = 2;
                    } else if (first_close_delims == 2) {
                        numdelims = 1;
                    } else if (numdelims == 3) {
                        // If we opened with ***, we interpret it as ** followed by *
                        // giving us <strong><em>
                        numdelims = 1;
                    }
                    subj->pos += numdelims;
                    if (first_close) {
                        first_head->tag = first_close_delims == 1 ? INL_STRONG : INL_EMPH;
                        chunk_free(&first_head->content.literal);
                        first_head->content.inlines =
                            make_inlines(first_close_delims == 1 ? INL_EMPH : INL_STRONG,
                                    first_head->next);

                        il = first_head->next;
                        while (il->next && il->next != first_close) {
                            il = il->next;
                        }
                        il->next = NULL;

                        first_head->content.inlines->next = first_close->next;

                        il = first_head->content.inlines;
                        while (il->next && il->next != *last) {
                            il = il->next;
                        }
                        il->next = NULL;
                        free_inlines(*last);

                        first_close->next = NULL;
                        free_inlines(first_close);
                        first_head->next = NULL;
                        goto done;
                    } else {
                        first_close = *last;
                        first_close_delims = numdelims;
                    }
                } else {
                    if (!parse_inline(subj, last)) {
                        goto done;
                    }
                }
            }
            break;
        default:
            goto done;
    }

done:
    free(last);
    return result;
}

// Parse backslash-escape or just a backslash, returning an inline.
static inl* handle_backslash(subject *subj)
{
    advance(subj);
    unsigned char nextchar = peek_char(subj);
    if (ispunct(nextchar)) {  // only ascii symbols and newline can be escaped
        advance(subj);
        return make_str(chunk_buf(subj->buffer, subj->pos - 1, 1));
    } else if (nextchar == '\n') {
        advance(subj);
        return make_linebreak();
    } else {
        return make_str(chunk_literal("\\"));
    }
}

// Parse an entity or a regular "&" string.
// Assumes the subject has an '&' character at the current position.
static inl* handle_entity(subject* subj)
{
    int match;
    inl *result;
    match = scan_entity(subj->buffer, subj->pos);
    if (match) {
        result = make_entity(chunk_buf(subj->buffer, subj->pos, match));
        subj->pos += match;
    } else {
        advance(subj);
        result = make_str(chunk_literal("&"));
    }
    return result;
}

// Like make_str, but parses entities.
// Returns an inline sequence consisting of str and entity elements.
static inl *make_str_with_entities(chunk *content)
{
    inl * result = NULL;
    inl * new;
    int searchpos;
    char c;
    subject subj;
    gh_buf content_buf = GH_BUF_INIT;

    gh_buf_set(&content_buf, content->data, content->len);
    init_subject(&subj, &content_buf, 0, NULL);

    while ((c = peek_char(&subj))) {
        switch (c) {
            case '&':
                new = handle_entity(&subj);
                break;
            default:
                searchpos = gh_buf_strchr(subj.buffer, '&', subj.pos);
                if (searchpos < 0) {
                    searchpos = gh_buf_len(subj.buffer);
                }

                new = make_str(chunk_buf(subj.buffer, subj.pos, searchpos - subj.pos));
                subj.pos = searchpos;
        }
        result = append_inlines(result, new);
    }

    gh_buf_free(&content_buf);
    return result;
}

// Destructively unescape a string: remove backslashes before punctuation chars.
extern void unescape_buffer(gh_buf *buf)
{
    int r, w;

    for (r = 0, w = 0; r < buf->size; ++r) {
        if (buf->ptr[r] == '\\' && ispunct(buf->ptr[r + 1]))
            continue;

        buf->ptr[w++] = buf->ptr[r];
    }

    gh_buf_truncate(buf, w);
}

// Clean a URL: remove surrounding whitespace and surrounding <>,
// and remove \ that escape punctuation.
static unsigned char *clean_url(chunk *url)
{
    gh_buf buf = GH_BUF_INIT;

    chunk_trim(url);

    if (url->data[0] == '<' && url->data[url->len - 1] == '>') {
        gh_buf_set(&buf, url->data + 1, url->len - 2);
    } else {
        gh_buf_set(&buf, url->data, url->len);
    }

    unescape_buffer(&buf);
    return gh_buf_detach(&buf);
}

// Clean a title: remove surrounding quotes and remove \ that escape punctuation.
static unsigned char *clean_title(chunk *title)
{
    gh_buf buf = GH_BUF_INIT;
    unsigned char first = title->data[0];
    unsigned char last = title->data[title->len - 1];

    // remove surrounding quotes if any:
    if ((first == '\'' && last == '\'') ||
        (first == '(' && last == ')') ||
        (first == '"' && last == '"')) {
        gh_buf_set(&buf, title->data + 1, title->len - 2);
    } else {
        gh_buf_set(&buf, title->data, title->len);
    }

    unescape_buffer(&buf);
    return gh_buf_detach(&buf);
}

// Parse an autolink or HTML tag.
// Assumes the subject has a '<' character at the current position.
static inl* handle_pointy_brace(subject* subj)
{
    int matchlen = 0;
    chunk contents;

    advance(subj);  // advance past first <

    // first try to match a URL autolink
    matchlen = scan_autolink_uri(subj->buffer, subj->pos);
    if (matchlen > 0) {
        contents = chunk_buf(subj->buffer, subj->pos, matchlen - 1);
        subj->pos += matchlen;

        return make_link(
            make_str_with_entities(&contents),
            contents,
            chunk_literal("")
        );
    }

    // next try to match an email autolink
    matchlen = scan_autolink_email(subj->buffer, subj->pos);
    if (matchlen > 0) {
        gh_buf mail_url = GH_BUF_INIT;

        contents = chunk_buf(subj->buffer, subj->pos, matchlen - 1);
        subj->pos += matchlen;

        gh_buf_puts(&mail_url, "mailto:");
        gh_buf_put(&mail_url, contents.data, contents.len);

        return make_link(
                make_str_with_entities(&contents),
                chunk_buf_detach(&mail_url),
                chunk_literal("")
        );
    }

    // finally, try to match an html tag
    matchlen = scan_html_tag(subj->buffer, subj->pos);
    if (matchlen > 0) {
        contents = chunk_buf(subj->buffer, subj->pos - 1, matchlen + 1);
        subj->pos += matchlen;
        return make_raw_html(contents);
    }

    // if nothing matches, just return the opening <:
    return make_str(chunk_literal("<"));
}

// Parse a link label.  Returns 1 if successful.
// Unless raw_label is null, it is set to point to the raw contents of the [].
// Assumes the subject has a '[' character at the current position.
// Returns 0 and does not advance if no matching ] is found.
// Note the precedence:  code backticks have precedence over label bracket
// markers, which have precedence over *, _, and other inline formatting
// markers. So, 2 below contains a link while 1 does not:
// 1. [a link `with a ](/url)` character
// 2. [a link *with emphasized ](/url) text*
static int link_label(subject* subj, chunk *raw_label)
{
    int nestlevel = 0;
    inl* tmp = NULL;
    int startpos = subj->pos;

    if (subj->label_nestlevel) {
        // if we've already checked to the end of the subject
        // for a label, even with a different starting [, we
        // know we won't find one here and we can just return.
        // Note:  nestlevel 1 would be: [foo [bar]
        // nestlevel 2 would be: [foo [bar [baz]
        subj->label_nestlevel--;
        return 0;
    }

    advance(subj);  // advance past [
    char c;
    while ((c = peek_char(subj)) && (c != ']' || nestlevel > 0)) {
        switch (c) {
            case '`':
                tmp = handle_backticks(subj);
                free_inlines(tmp);
                break;
            case '<':
                tmp = handle_pointy_brace(subj);
                free_inlines(tmp);
                break;
            case '[':  // nested []
                nestlevel++;
                advance(subj);
                break;
            case ']':  // nested []
                nestlevel--;
                advance(subj);
                break;
            case '\\':
                advance(subj);
                if (ispunct(peek_char(subj))) {
                    advance(subj);
                }
                break;
            default:
                advance(subj);
        }
    }
    if (c == ']') {
        *raw_label = chunk_buf(
            subj->buffer,
            startpos + 1,
            subj->pos - (startpos + 1)
        );

        subj->label_nestlevel = 0;
        advance(subj);  // advance past ]
        return 1;
    } else {
        if (c == 0) {
            subj->label_nestlevel = nestlevel;
        }
        subj->pos = startpos; // rewind
        return 0;
    }
}

// Parse a link or the link portion of an image, or return a fallback.
static inl* handle_left_bracket(subject* subj)
{
    inl *lab = NULL;
    inl *result = NULL;
    reference *ref;
    int n;
    int sps;
    int found_label;
    int endlabel, starturl, endurl, starttitle, endtitle, endall;

    chunk rawlabel;
    chunk url, title;

    found_label = link_label(subj, &rawlabel);
    endlabel = subj->pos;

    if (found_label) {
        if (peek_char(subj) == '(' &&
                ((sps = scan_spacechars(subj->buffer, subj->pos + 1)) > -1) &&
                ((n = scan_link_url(subj->buffer, subj->pos + 1 + sps)) > -1)) {

            // try to parse an explicit link:
            starturl = subj->pos + 1 + sps; // after (
            endurl = starturl + n;
            starttitle = endurl + scan_spacechars(subj->buffer, endurl);

            // ensure there are spaces btw url and title
            endtitle = (starttitle == endurl) ? starttitle :
                starttitle + scan_link_title(subj->buffer, starttitle);

            endall = endtitle + scan_spacechars(subj->buffer, endtitle);

            if (gh_buf_at(subj->buffer, endall) == ')') {
                subj->pos = endall + 1;

                url = chunk_buf(subj->buffer, starturl, endurl - starturl);
                title = chunk_buf(subj->buffer, starttitle, endtitle - starttitle);
                lab = parse_chunk_inlines(&rawlabel, NULL);

                return make_link(lab, url, title);
            } else {
                // if we get here, we matched a label but didn't get further:
                subj->pos = endlabel;
                lab = parse_chunk_inlines(&rawlabel, subj->reference_map);
                result = append_inlines(make_str(chunk_literal("[")),
                        append_inlines(lab,
                            make_str(chunk_literal("]"))));
                return result;
            }
        } else {
            chunk rawlabel_tmp;
            chunk reflabel;

            // Check for reference link.
            // First, see if there's another label:
            subj->pos = subj->pos + scan_spacechars(subj->buffer, endlabel);
            reflabel = rawlabel;

            // if followed by a nonempty link label, we change reflabel to it:
            if (peek_char(subj) == '[' && link_label(subj, &rawlabel_tmp)) {
                if (rawlabel_tmp.len > 0)
                    reflabel = rawlabel_tmp;
            } else {
                subj->pos = endlabel;
            }

            // lookup rawlabel in subject->reference_map:
            ref = lookup_reference(subj->reference_map, &reflabel);
            if (ref != NULL) { // found
                lab = parse_chunk_inlines(&rawlabel, NULL);
                result = make_link(lab, chunk_literal(ref->url), chunk_literal(ref->title));
            } else {
                subj->pos = endlabel;
                lab = parse_chunk_inlines(&rawlabel, subj->reference_map);
                result = append_inlines(make_str(chunk_literal("[")),
                        append_inlines(lab, make_str(chunk_literal("]"))));
            }
            return result;
        }
    }
    // If we fall through to here, it means we didn't match a link:
    advance(subj);  // advance past [
    return make_str(chunk_literal("["));
}

// Parse a hard or soft linebreak, returning an inline.
// Assumes the subject has a newline at the current position.
static inl* handle_newline(subject *subj)
{
    int nlpos = subj->pos;
    // skip over newline
    advance(subj);
    // skip spaces at beginning of line
    while (peek_char(subj) == ' ') {
        advance(subj);
    }
    if (nlpos > 1 &&
            gh_buf_at(subj->buffer, nlpos - 1) == ' ' &&
            gh_buf_at(subj->buffer, nlpos - 2) == ' ') {
        return make_linebreak();
    } else {
        return make_softbreak();
    }
}

inline static int not_eof(subject* subj)
{
    return !is_eof(subj);
}

// Parse inlines while a predicate is satisfied.  Return inlines.
extern inl* parse_inlines_while(subject* subj, int (*f)(subject*))
{
    inl* result = NULL;
    inl** last = &result;
    while ((*f)(subj) && parse_inline(subj, last)) {
    }
    return result;
}

inl *parse_chunk_inlines(chunk *chunk, reference** refmap)
{
    inl *result;
    subject subj;
    gh_buf full_chunk = GH_BUF_INIT;

    gh_buf_set(&full_chunk, chunk->data, chunk->len);
    init_subject(&subj, &full_chunk, 0, refmap);
    result = parse_inlines_while(&subj, not_eof);

    gh_buf_free(&full_chunk);
    return result;
}

static int find_special_char(subject *subj)
{
    int n = subj->pos + 1;
    int size = (int)gh_buf_len(subj->buffer);

    while (n < size) {
        if (strchr("\n\\`&_*[]<!", gh_buf_at(subj->buffer, n)))
            return n;
    }

    return -1;
}

// Parse an inline, advancing subject, and add it to last element.
// Adjust tail to point to new last element of list.
// Return 0 if no inline can be parsed, 1 otherwise.
static int parse_inline(subject* subj, inl ** last)
{
    inl* new = NULL;
    chunk contents;
    unsigned char c;
    int endpos;
    c = peek_char(subj);
    if (c == 0) {
        return 0;
    }
    switch(c){
        case '\n':
            new = handle_newline(subj);
            break;
        case '`':
            new = handle_backticks(subj);
            break;
        case '\\':
            new = handle_backslash(subj);
            break;
        case '&':
            new = handle_entity(subj);
            break;
        case '<':
            new = handle_pointy_brace(subj);
            break;
        case '_':
            if (subj->pos > 0 && (isalnum(gh_buf_at(subj->buffer, subj->pos - 1)) ||
                        gh_buf_at(subj->buffer, subj->pos - 1) == '_')) {
                goto text_literal;
            }

            new = handle_strong_emph(subj, '_');
            break;
        case '*':
            new = handle_strong_emph(subj, '*');
            break;
        case '[':
            new = handle_left_bracket(subj);
            break;
        case '!':
            advance(subj);
            if (peek_char(subj) == '[') {
                new = handle_left_bracket(subj);
                if (new != NULL && new->tag == INL_LINK) {
                    new->tag = INL_IMAGE;
                } else {
                    new = append_inlines(make_str(chunk_literal("!")), new);
                }
            } else {
                new = make_str(chunk_literal("!"));
            }
            break;
        default:
        text_literal:
            endpos = find_special_char(subj);
            if (endpos < 0) {
                endpos = gh_buf_len(subj->buffer);
            }

            contents = chunk_buf(subj->buffer, subj->pos, endpos - subj->pos);
            subj->pos = endpos;

            // if we're at a newline, strip trailing spaces.
            if (peek_char(subj) == '\n') {
                chunk_trim(&contents);
            }

            new = make_str(contents);
    }
    if (*last == NULL) {
        *last = new;
    } else {
        append_inlines(*last, new);
    }
    return 1;
}

extern inl* parse_inlines(gh_buf *input, int input_pos, reference** refmap)
{
    subject subj;
    init_subject(&subj, input, input_pos, refmap);
    return parse_inlines_while(&subj, not_eof);
}

// Parse zero or more space characters, including at most one newline.
void spnl(subject* subj)
{
    bool seen_newline = false;
    while (peek_char(subj) == ' ' ||
            (!seen_newline &&
             (seen_newline = peek_char(subj) == '\n'))) {
        advance(subj);
    }
}

// Parse reference.  Assumes string begins with '[' character.
// Modify refmap if a reference is encountered.
// Return 0 if no reference found, otherwise position of subject
// after reference is parsed.
extern int parse_reference(gh_buf *input, int input_pos, reference** refmap)
{
    subject subj;

    chunk lab;
    chunk url;
    chunk title;

    int matchlen = 0;
    int beforetitle;
    reference * new = NULL;

    init_subject(&subj, input, input_pos, NULL);

    // parse label:
    if (!link_label(&subj, &lab))
        return 0;

    // colon:
    if (peek_char(&subj) == ':') {
        advance(&subj);
    } else {
        return 0;
    }

    // parse link url:
    spnl(&subj);
    matchlen = scan_link_url(subj.buffer, subj.pos);
    if (matchlen) {
        url = chunk_buf(subj.buffer, subj.pos, matchlen);
        subj.pos += matchlen;
    } else {
        return 0;
    }

    // parse optional link_title
    beforetitle = subj.pos;
    spnl(&subj);
    matchlen = scan_link_title(subj.buffer, subj.pos);
    if (matchlen) {
        title = chunk_buf(subj.buffer, subj.pos, matchlen);
        subj.pos += matchlen;
    } else {
        subj.pos = beforetitle;
        title = chunk_literal("");
    }
    // parse final spaces and newline:
    while (peek_char(&subj) == ' ') {
        advance(&subj);
    }
    if (peek_char(&subj) == '\n') {
        advance(&subj);
    } else if (peek_char(&subj) != 0) {
        return 0;
    }
    // insert reference into refmap
    new = make_reference(&lab, &url, &title);
    add_reference(refmap, new);

    return subj.pos;
}