#include <zlib.h>
#include <stdio.h>
#include <ctype.h>
#include <stdlib.h>
#include <string.h>
#include <stdarg.h>
#include "kstring.h"
#include "gfa1/gfa.h"

#include "kseq.h"
KSTREAM_INIT2(, gzFile, gzread, 65536)

#include "khash.h"
KHASH_MAP_INIT_STR(seg, uint32_t)
typedef khash_t(seg) seghash_t;

#include "ksort.h"
#define gfa_arc_key(a) ((a).v_lv)
KRADIX_SORT_INIT(arc, gfa_arc_t, gfa_arc_key, 8)

int gfa_verbose = 2;

/********************
 * Tag manipulation *
 ********************/

int gfa_aux_parse(char *s, uint8_t **data, int *max)
{
	char *q, *p;
	kstring_t str;
	if (s == 0) return 0;
	str.l = 0, str.m = *max, str.s = (char*)*data;
	if (*s == '\t') ++s;
	for (p = q = s;; ++p) {
		if (*p == 0 || *p == '\t') {
			int c = *p;
			*p = 0;
			if (p - q >= 5 && q[2] == ':' && q[4] == ':' && (q[3] == 'A' || q[3] == 'i' || q[3] == 'f' || q[3] == 'Z' || q[3] == 'B')) {
				int type = q[3];
				kputsn_(q, 2, &str);
				q += 5;
				if (type == 'A') {
					kputc_('A', &str);
					kputc_(*q, &str);
				} else if (type == 'i') {
					int32_t x;
					x = strtol(q, &q, 10);
					kputc_(type, &str); kputsn_((char*)&x, 4, &str);
				} else if (type == 'f') {
					float x;
					x = strtod(q, &q);
					kputc_('f', &str); kputsn_(&x, 4, &str);
				} else if (type == 'Z') {
					kputc_('Z', &str); kputsn_(q, p - q + 1, &str); // note that this include the trailing NULL
				} else if (type == 'B') {
					type = *q++; // q points to the first ',' following the typing byte
					if (p - q >= 2 && (type == 'c' || type == 'C' || type == 's' || type == 'S' || type == 'i' || type == 'I' || type != 'f')) {
						int32_t n;
						char *r;
						for (r = q, n = 0; *r; ++r)
							if (*r == ',') ++n;
						kputc_('B', &str); kputc_(type, &str); kputsn_(&n, 4, &str);
						// TODO: to evaluate which is faster: a) aligned array and then memmove(); b) unaligned array; c) kputsn_()
						if (type == 'c')      while (q + 1 < p) { int8_t   x = strtol(q + 1, &q, 0); kputc_(x, &str); }
						else if (type == 'C') while (q + 1 < p) { uint8_t  x = strtol(q + 1, &q, 0); kputc_(x, &str); }
						else if (type == 's') while (q + 1 < p) { int16_t  x = strtol(q + 1, &q, 0); kputsn_(&x, 2, &str); }
						else if (type == 'S') while (q + 1 < p) { uint16_t x = strtol(q + 1, &q, 0); kputsn_(&x, 2, &str); }
						else if (type == 'i') while (q + 1 < p) { int32_t  x = strtol(q + 1, &q, 0); kputsn_(&x, 4, &str); }
						else if (type == 'I') while (q + 1 < p) { uint32_t x = strtol(q + 1, &q, 0); kputsn_(&x, 4, &str); }
						else if (type == 'f') while (q + 1 < p) { float    x = strtod(q + 1, &q);    kputsn_(&x, 4, &str); }
					}
				} // should not be here, as we have tested all types
			}
			q = p + 1;
			if (c == 0) break;
		}
	}
	if (str.s) str.s[str.l] = 0;
	*max = str.m, *data = (uint8_t*)str.s;
	return str.l;
}

int gfa_aux_format(int l_aux, const uint8_t *aux, char **t, int *max)
{
	kstring_t str;
	const uint8_t *s = aux;
	str.l = 0, str.s = *t, str.m = *max;
	while (s < aux + l_aux) {
		uint8_t type, key[2];
		key[0] = s[0]; key[1] = s[1];
		s += 2; type = *s++;
		kputc('\t', &str); kputsn((char*)key, 2, &str); kputc(':', &str);
		if (type == 'A') { kputsn("A:", 2, &str); kputc(*s, &str); ++s; }
		else if (type == 'i') { kputsn("i:", 2, &str); kputw(*(int32_t*)s, &str); s += 4; }
		else if (type == 'f') { ksprintf(&str, "f:%g", *(float*)s); s += 4; }
		else if (type == 'Z') { kputc(type, &str); kputc(':', &str); while (*s) kputc(*s++, &str); ++s; }
		else if (type == 'B') {
			uint8_t sub_type = *(s++);
			int32_t i, n;
			memcpy(&n, s, 4);
			s += 4; // no point to the start of the array
			kputsn("B:", 2, &str); kputc(sub_type, &str); // write the typing
			for (i = 0; i < n; ++i) { // FIXME: for better performance, put the loop after "if"
				kputc(',', &str);
				if ('c' == sub_type)      { kputw(*(int8_t*)s, &str); ++s; }
				else if ('C' == sub_type) { kputw(*(uint8_t*)s, &str); ++s; }
				else if ('s' == sub_type) { kputw(*(int16_t*)s, &str); s += 2; }
				else if ('S' == sub_type) { kputw(*(uint16_t*)s, &str); s += 2; }
				else if ('i' == sub_type) { kputw(*(int32_t*)s, &str); s += 4; }
				else if ('I' == sub_type) { kputuw(*(uint32_t*)s, &str); s += 4; }
				else if ('f' == sub_type) { ksprintf(&str, "%g", *(float*)s); s += 4; }
			}
		}
	}
	*t = str.s, *max = str.m;
	return str.l;
}

static inline int gfa_aux_type2size(int x)
{
	if (x == 'C' || x == 'c' || x == 'A') return 1;
	else if (x == 'S' || x == 's') return 2;
	else if (x == 'I' || x == 'i' || x == 'f') return 4;
	else return 0;
}

#define __skip_tag(s) do { \
		int type = toupper(*(s)); \
		++(s); \
		if (type == 'Z') { while (*(s)) ++(s); ++(s); } \
		else if (type == 'B') (s) += 5 + gfa_aux_type2size(*(s)) * (*(int32_t*)((s)+1)); \
		else (s) += gfa_aux_type2size(type); \
	} while(0)

uint8_t *gfa_aux_get(int l_data, const uint8_t *data, const char tag[2])
{
	const uint8_t *s = data;
	int y = tag[0]<<8 | tag[1];
	while (s < data + l_data) {
		int x = (int)s[0]<<8 | s[1];
		s += 2;
		if (x == y) return (uint8_t*)s;
		__skip_tag(s);
	}
	return 0;
}

// s MUST BE returned by gfa_aux_get()
int gfa_aux_del(int l_data, uint8_t *data, uint8_t *s)
{
	uint8_t *p;
	p = s - 2;
	__skip_tag(s);
	memmove(p, s, l_data - (s - data));
	return l_data - (s - p);
}

/******************
 * Basic routines *
 ******************/

gfa_t *gfa_init(void)
{
	gfa_t *g;
	g = (gfa_t*)calloc(1, sizeof(gfa_t));
	g->h_names = kh_init(seg);
	return g;
}

void gfa_destroy(gfa_t *g)
{
	uint32_t i, j;
	uint64_t k;
	if (g == 0) return;
	kh_destroy(seg, (seghash_t*)g->h_names);
	for (i = 0; i < g->n_seg; ++i) {
		gfa_seg_t *s = &g->seg[i];
		free(s->name);
		free(s->aux.aux);
		for (j = 0; j < s->utg.n; ++j)
			free(s->utg.name[j]);
		free(s->utg.name);
		free(s->utg.a);
	}
	for (k = 0; k < g->n_arc; ++k)
		free(g->arc_aux[k].aux);
	free(g->idx); free(g->seg); free(g->arc); free(g->arc_aux);
	free(g);
}

int32_t gfa_add_seg(gfa_t *g, const char *name)
{
	khint_t k;
	int absent;
	seghash_t *h = (seghash_t*)g->h_names;
	k = kh_put(seg, h, name, &absent);
	if (absent) {
		gfa_seg_t *s;
		if (g->n_seg == g->m_seg) {
			uint32_t old_m = g->m_seg;
			g->m_seg = g->m_seg? g->m_seg<<1 : 16;
			g->seg = (gfa_seg_t*)realloc(g->seg, g->m_seg * sizeof(gfa_seg_t));
			memset(&g->seg[old_m], 0, (g->m_seg - old_m) * sizeof(gfa_seg_t));
		}
		s = &g->seg[g->n_seg++];
		kh_key(h, k) = s->name = strdup(name);
		s->del = s->len = 0;
		kh_val(h, k) = g->n_seg - 1;
	}
	return kh_val(h, k);
}

uint64_t gfa_add_arc1(gfa_t *g, uint32_t v, uint32_t w, int32_t ov, int32_t ow, int64_t link_id, int comp)
{
	gfa_arc_t *a;
	if (g->m_arc == g->n_arc) {
		uint64_t old_m = g->m_arc;
		g->m_arc = g->m_arc? g->m_arc<<1 : 16;
		g->arc = (gfa_arc_t*)realloc(g->arc, g->m_arc * sizeof(gfa_arc_t));
		memset(&g->arc[old_m], 0, (g->m_arc - old_m) * sizeof(gfa_arc_t));
		g->arc_aux = (gfa_aux_t*)realloc(g->arc_aux, g->m_arc * sizeof(gfa_aux_t));
		memset(&g->arc_aux[old_m], 0, (g->m_arc - old_m) * sizeof(gfa_aux_t));
	}
	a = &g->arc[g->n_arc++];
	a->v_lv = (uint64_t)v << 32;
	a->w = w, a->ov = ov, a->ow = ow, a->lw = 0;
	a->link_id = link_id >= 0? link_id : g->n_arc - 1;
	a->del = 0;
	a->comp = comp;
	return a->link_id;
}

void gfa_arc_sort(gfa_t *g)
{
	radix_sort_arc(g->arc, g->arc + g->n_arc);
}

uint64_t *gfa_arc_index_core(size_t max_seq, size_t n, const gfa_arc_t *a)
{
	size_t i, last;
	uint64_t *idx;
	idx = (uint64_t*)calloc(max_seq * 2, 8);
	for (i = 1, last = 0; i <= n; ++i)
		if (i == n || gfa_arc_head(a[i-1]) != gfa_arc_head(a[i]))
			idx[gfa_arc_head(a[i-1])] = (uint64_t)last<<32 | (i - last), last = i;
	return idx;
}

void gfa_arc_index(gfa_t *g)
{
	if (g->idx) free(g->idx);
	g->idx = gfa_arc_index_core(g->n_seg, g->n_arc, g->arc);
}

/****************
 * Line parsers *
 ****************/

int gfa_parse_S(gfa_t *g, char *s)
{
	int i, is_ok = 0;
	char *p, *q, *seg = 0, *seq = 0, *rest = 0;
	uint32_t sid, len = 0;
	for (i = 0, p = q = s + 2;; ++p) {
		if (*p == 0 || *p == '\t') {
			int c = *p;
			*p = 0;
			if (i == 0) seg = q;
			else if (i == 1) {
				seq = q[0] == '*'? 0 : strdup(q);
				is_ok = 1, rest = c? p + 1 : 0;
				break;
			}
			++i, q = p + 1;
			if (c == 0) break;
		}
	}
	if (is_ok) { // all mandatory fields read
		int l_aux, m_aux = 0;
		uint8_t *aux = 0;
		gfa_seg_t *s;
		l_aux = gfa_aux_parse(rest, &aux, &m_aux); // parse optional tags
		if (seq == 0) {
			uint8_t *s;
			s = gfa_aux_get(l_aux, aux, "LN");
			if (s && s[0] == 'i')
				len = *(int32_t*)(s+1);
		} else len = strlen(seq);
		sid = gfa_add_seg(g, seg);
		s = &g->seg[sid];
		s->len = len, s->seq = seq;
		s->aux.m_aux = m_aux, s->aux.l_aux = l_aux, s->aux.aux = aux;
	} else return -1;
	return 0;
}

int gfa_parse_L(gfa_t *g, char *s)
{
	int i, oriv = -1, oriw = -1, is_ok = 0;
	char *p, *q, *segv = 0, *segw = 0, *rest = 0;
	int32_t ov = INT32_MAX, ow = INT32_MAX;
	for (i = 0, p = q = s + 2;; ++p) {
		if (*p == 0 || *p == '\t') {
			int c = *p;
			*p = 0;
			if (i == 0) {
				segv = q;
			} else if (i == 1) {
				if (*q != '+' && *q != '-') return -2;
				oriv = (*q != '+');
			} else if (i == 2) {
				segw = q;
			} else if (i == 3) {
				if (*q != '+' && *q != '-') return -2;
				oriw = (*q != '+');
			} else if (i == 4) {
				if (*q == ':') {
					ov = INT32_MAX;
					ow = isdigit(*(q+1))? strtol(q+1, &q, 10) : INT32_MAX;
				} else if (isdigit(*q)) {
					char *r;
					ov = strtol(q, &r, 10);
					if (isupper(*r)) { // CIGAR
						ov = ow = 0;
						do {
							long l;
							l = strtol(q, &q, 10);
							if (*q == 'M' || *q == 'D' || *q == 'N') ov += l;
							if (*q == 'M' || *q == 'I' || *q == 'S') ow += l;
							++q;
						} while (isdigit(*q));
					} else if (*r == ':') { // overlap lengths
						ow = isdigit(*(r+1))? strtol(r+1, &r, 10) : INT32_MAX;
					} else break;
				} else break;
				is_ok = 1, rest = c? p + 1 : 0;
				break;
			}
			++i, q = p + 1;
			if (c == 0) break;
		}
	}
	if (is_ok) {
		uint32_t v, w;
		uint64_t link_id;
		int l_aux, m_aux = 0;
		uint8_t *aux = 0;
		v = gfa_add_seg(g, segv) << 1 | oriv;
		w = gfa_add_seg(g, segw) << 1 | oriw;
		link_id = gfa_add_arc1(g, v, w, ov, ow, -1, 0);
		l_aux = gfa_aux_parse(rest, &aux, &m_aux); // parse optional tags
		if (l_aux) {
			gfa_aux_t *a = &g->arc_aux[link_id];
			uint8_t *s_L1, *s_L2;
			a->l_aux = l_aux, a->m_aux = m_aux, a->aux = aux;
			s_L1 = gfa_aux_get(a->l_aux, a->aux, "L1");
			if (s_L1) {
				if (ov != INT32_MAX && s_L1[0] == 'i')
					g->seg[v>>1].len = g->seg[v>>1].len > ov + *(int32_t*)(s_L1+1)? g->seg[v>>1].len : ov + *(int32_t*)(s_L1+1);
				a->l_aux = gfa_aux_del(a->l_aux, a->aux, s_L1);
			}
			s_L2 = gfa_aux_get(a->l_aux, a->aux, "L2");
			if (s_L2) {
				if (ow != INT32_MAX && s_L2[0] == 'i')
					g->seg[w>>1].len = g->seg[w>>1].len > ow + *(int32_t*)(s_L2+1)? g->seg[w>>1].len : ow + *(int32_t*)(s_L2+1);
				a->l_aux = gfa_aux_del(a->l_aux, a->aux, s_L2);
			}
			if (a->l_aux == 0) {
				free(a->aux);
				a->aux = 0;
			}
		}
	} else return -1;
	return 0;
}

/********************
 * Fix graph issues *
 ********************/

uint32_t gfa_fix_no_seg(gfa_t *g)
{
	uint32_t i, n_err = 0;
	for (i = 0; i < g->n_seg; ++i) {
		gfa_seg_t *s = &g->seg[i];
		if (s->len == 0) {
			++n_err, s->del = 1;
			if (gfa_verbose >= 2)
				fprintf(stderr, "[W] segment '%s' is used on an L-line but not defined on an S-line\n", s->name);
		}
	}
	return n_err;
}

static void gfa_fix_arc_len(gfa_t *g)
{
	uint64_t k;
	for (k = 0; k < g->n_arc; ++k) {
		gfa_arc_t *a = &g->arc[k];
		uint32_t v = gfa_arc_head(*a), w = gfa_arc_tail(*a);
		if (g->seg[v>>1].del || g->seg[w>>1].del) {
			a->del = 1;
		} else {
			a->v_lv |= g->seg[v>>1].len - a->ov;
			if (a->ow != INT32_MAX)
				a->lw = g->seg[w>>1].len - a->ow;
		}
	}
}

static uint32_t gfa_fix_semi_arc(gfa_t *g)
{
	uint32_t n_err = 0, v, n_vtx = gfa_n_vtx(g);
	int i, j;
	for (v = 0; v < n_vtx; ++v) {
		int nv = gfa_arc_n(g, v);
		gfa_arc_t *av = gfa_arc_a(g, v);
		for (i = 0; i < nv; ++i) {
			if (!av[i].del && (av[i].ow == INT32_MAX || av[i].ov == INT32_MAX)) { // overlap length is missing
				uint32_t w = av[i].w^1;
				int is_multi = 0, c, jv = -1, nw = gfa_arc_n(g, w);
				gfa_arc_t *aw = gfa_arc_a(g, w);
				for (j = 0, c = 0; j < nw; ++j)
					if (!aw[j].del && aw[j].w == (v^1)) ++c, jv = j;
				if (c == 1) {
					if (av[i].ov != INT32_MAX && aw[jv].ow != INT32_MAX && av[i].ov != aw[jv].ow) is_multi = 1;
					if (av[i].ow != INT32_MAX && aw[jv].ov != INT32_MAX && av[i].ow != aw[jv].ov) is_multi = 1;
				}
				if (c == 1 && !is_multi) {
					if (aw[jv].ov != INT32_MAX) av[i].ow = aw[jv].ov;
					if (aw[jv].ow != INT32_MAX) av[i].ov = aw[jv].ow;
				} else {
					if (gfa_verbose >= 2)
						fprintf(stderr, "[W] can't infer overlap length for %s%c -> %s%c\n",
								g->seg[v>>1].name, "+-"[v&1], g->seg[w>>1].name, "+-"[(w^1)&1]);
					++n_err;
					av[i].del = 1;
				}
			}
		}
	}
	return n_err;
}

uint32_t gfa_fix_symm(gfa_t *g)
{
	uint32_t n_err = 0, v, n_vtx = gfa_n_vtx(g);
	int i;
	for (v = 0; v < n_vtx; ++v) {
		int nv = gfa_arc_n(g, v);
		gfa_arc_t *av = gfa_arc_a(g, v);
		for (i = 0; i < nv; ++i) {
			int j, nw;
			gfa_arc_t *aw, *avi = &av[i];
			if (avi->del || avi->comp) continue;
			nw = gfa_arc_n(g, avi->w^1);
			aw = gfa_arc_a(g, avi->w^1);
			for (j = 0; j < nw; ++j) {
				gfa_arc_t *awj = &aw[j];
				if (awj->del || awj->comp) continue;
				if (awj->w == (v^1) && awj->ov == avi->ow && awj->ow == avi->ov) { // complement found
					awj->comp = 1;
					awj->link_id = avi->link_id;
					break;
				}
			}
			if (j == nw) {
				gfa_arc_t *arc_old = g->arc;
				gfa_add_arc1(g, avi->w^1, v^1, avi->ow, avi->ov, avi->link_id, 1);
				if (arc_old != g->arc) av = gfa_arc_a(g, v); // g->arc may be reallocated
			}
		}
	}
	if (n_vtx < gfa_n_vtx(g)) {
		gfa_arc_sort(g);
		gfa_arc_index(g);
	}
	return n_err;
}

/****************
 * User-end I/O *
 ****************/

gfa_t *gfa_read(const char *fn)
{
	gzFile fp;
	gfa_t *g;
	kstring_t s = {0,0,0};
	kstream_t *ks;
	int dret;
	uint64_t lineno = 0;

	fp = fn && strcmp(fn, "-")? gzopen(fn, "r") : gzdopen(fileno(stdin), "r");
	if (fp == 0) return 0;
	ks = ks_init(fp);
	g = gfa_init();
	while (ks_getuntil(ks, KS_SEP_LINE, &s, &dret) >= 0) {
		int ret = 0;
		++lineno;
		if (s.l < 3 || s.s[1] != '\t') continue; // empty line
		if (s.s[0] == 'S') ret = gfa_parse_S(g, s.s);
		else if (s.s[0] == 'L') ret = gfa_parse_L(g, s.s);
		if (ret < 0 && gfa_verbose >= 1)
			fprintf(stderr, "[E] invalid %c-line at line %ld (error code %d)\n", s.s[0], (long)lineno, ret);
	}
	free(s.s);
	gfa_fix_no_seg(g);
	gfa_arc_sort(g);
	gfa_arc_index(g);
	gfa_fix_semi_arc(g);
	gfa_fix_symm(g);
	gfa_fix_arc_len(g);
	gfa_cleanup(g);
	ks_destroy(ks);
	gzclose(fp);
	return g;
}

void gfa_print(const gfa_t *g, FILE *fp, int M_only)
{
	uint32_t i;
	uint64_t k;
	for (i = 0; i < g->n_seg; ++i) {
		const gfa_seg_t *s = &g->seg[i];
		if (s->del) continue;
		fprintf(fp, "S\t%s\t", s->name);
		if (s->seq) fputs(s->seq, fp);
		else fputc('*', fp);
		if (s->aux.l_aux == 0 || !gfa_aux_get(s->aux.l_aux, s->aux.aux, "LN"))
			fprintf(fp, "\tLN:i:%d", s->len);
		if (s->aux.l_aux > 0) {
			char *t = 0;
			int max = 0, len;
			len = gfa_aux_format(s->aux.l_aux, s->aux.aux, &t, &max);
			fputs(t, fp);
			free(t);
		}
		fputc('\n', fp);
		if (s->utg.n) {
			uint32_t j, l;
			for (j = l = 0; j < s->utg.n; ++j) {
				const gfa_utg_t *u = &s->utg;
				fprintf(fp, "a\t%s\t%d\t%s\t%c\t%d\n", s->name, l, u->name[j], "+-"[u->a[j]>>32&1], (uint32_t)u->a[j]);
				l += (uint32_t)u->a[j];
			}
		}
	}
	for (k = 0; k < g->n_arc; ++k) {
		const gfa_arc_t *a = &g->arc[k];
		const gfa_aux_t *aux = &g->arc_aux[a->link_id];
		if (a->del || a->comp) continue;
		fprintf(fp, "L\t%s\t%c\t%s\t%c", g->seg[a->v_lv>>33].name, "+-"[a->v_lv>>32&1], g->seg[a->w>>1].name, "+-"[a->w&1]);
		if (M_only) {
			fprintf(fp, "\t%dM", a->ov < a->ow? a->ov : a->ow);
		} else {
			if (a->ov == a->ow) fprintf(fp, "\t%dM", a->ov);
			else fprintf(fp, "\t%d:%d", a->ov, a->ow);
		}
		fprintf(fp, "\tL1:i:%d", gfa_arc_len(*a));
		fprintf(fp, "\tL2:i:%d", a->lw);
		if (aux->l_aux) {
			char *t = 0;
			int max = 0, len;
			len = gfa_aux_format(aux->l_aux, aux->aux, &t, &max);
			if (t) fputs(t, fp);
			free(t);
		}
		fputc('\n', fp);
	}
}

/**********************
 * Graph manipulation *
 **********************/

#include <assert.h>
#include "kvec.h"

typedef struct { size_t n, m; uint64_t *a; } gfa64_v;

void gfa_arc_rm(gfa_t *g)
{
	uint32_t e, n;
	for (e = n = 0; e < g->n_arc; ++e) {
		uint32_t u = g->arc[e].v_lv>>32, v = g->arc[e].w;
		if (!g->arc[e].del && !g->seg[u>>1].del && !g->seg[v>>1].del)
			g->arc[n++] = g->arc[e];
	}
	if (n < g->n_arc) { // arc index is out of sync
		if (g->idx) free(g->idx);
		g->idx = 0;
	}
	g->n_arc = n;
}

void gfa_cleanup(gfa_t *g)
{
	gfa_arc_rm(g);
	if (!g->is_srt) {
		gfa_arc_sort(g);
		g->is_srt = 1;
		if (g->idx) free(g->idx);
		g->idx = 0;
	}
	if (g->idx == 0) gfa_arc_index(g);
}

// delete short arcs
int gfa_arc_del_short(gfa_t *g, float drop_ratio)
{
	uint32_t v, n_vtx = gfa_n_vtx(g), n_short = 0;
	for (v = 0; v < n_vtx; ++v) {
		gfa_arc_t *av = gfa_arc_a(g, v);
		uint32_t i, thres, nv = gfa_arc_n(g, v);
		if (nv < 2) continue;
		thres = (uint32_t)(av[0].ov * drop_ratio + .499);
		for (i = nv - 1; i >= 1 && av[i].ov < thres; --i);
		for (i = i + 1; i < nv; ++i)
			av[i].del = 1, ++n_short;
	}
	if (n_short) {
		gfa_cleanup(g);
		gfa_symm(g);
	}
	if (gfa_verbose >= 3) fprintf(stderr, "[M] removed %d short overlaps\n", n_short);
	return n_short;
}

// delete multi-arcs
static int gfa_arc_del_multi(gfa_t *g)
{
	uint32_t *cnt, n_vtx = gfa_n_vtx(g), n_multi = 0, v;
	cnt = (uint32_t*)calloc(n_vtx, 4);
	for (v = 0; v < n_vtx; ++v) {
		gfa_arc_t *av = gfa_arc_a(g, v);
		int32_t i, nv = gfa_arc_n(g, v);
		if (nv < 2) continue;
		for (i = nv - 1; i >= 0; --i) ++cnt[av[i].w];
		for (i = nv - 1; i >= 0; --i)
			if (--cnt[av[i].w] != 0)
				av[i].del = 1, ++n_multi;
	}
	free(cnt);
	if (n_multi) gfa_cleanup(g);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] removed %d multi-arcs\n", n_multi);
	return n_multi;
}

// remove asymmetric arcs: u->v is present, but v'->u' not
static int gfa_arc_del_asymm(gfa_t *g)
{
	uint32_t e, n_asymm = 0;
	for (e = 0; e < g->n_arc; ++e) {
		uint32_t v = g->arc[e].w^1, u = g->arc[e].v_lv>>32^1;
		uint32_t i, nv = gfa_arc_n(g, v);
		gfa_arc_t *av = gfa_arc_a(g, v);
		for (i = 0; i < nv; ++i)
			if (av[i].w == u) break;
		if (i == nv) g->arc[e].del = 1, ++n_asymm;
	}
	if (n_asymm) gfa_cleanup(g);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] removed %d asymmetric arcs\n", n_asymm);
	return n_asymm;
}

void gfa_symm(gfa_t *g)
{
	gfa_arc_del_multi(g);
	gfa_arc_del_asymm(g);
	g->is_symm = 1;
}

// transitive reduction; see Myers, 2005
int gfa_arc_del_trans(gfa_t *g, int fuzz)
{
	uint8_t *mark;
	uint32_t v, n_vtx = gfa_n_vtx(g), n_reduced = 0;

	mark = (uint8_t*)calloc(n_vtx, 1);
	for (v = 0; v < n_vtx; ++v) {
		uint32_t L, i, nv = gfa_arc_n(g, v);
		gfa_arc_t *av = gfa_arc_a(g, v);
		if (nv == 0) continue; // no hits
		if (g->seg[v>>1].del) {
			for (i = 0; i < nv; ++i) av[i].del = 1, ++n_reduced;
			continue;
		}
		for (i = 0; i < nv; ++i) mark[av[i].w] = 1;
		L = gfa_arc_len(av[nv-1]) + fuzz;
		for (i = 0; i < nv; ++i) {
			uint32_t w = av[i].w;
			uint32_t j, nw = gfa_arc_n(g, w);
			gfa_arc_t *aw = gfa_arc_a(g, w);
			if (mark[av[i].w] != 1) continue;
			for (j = 0; j < nw && gfa_arc_len(aw[j]) + gfa_arc_len(av[i]) <= L; ++j)
				if (mark[aw[j].w]) mark[aw[j].w] = 2;
		}
		#if 0
		for (i = 0; i < nv; ++i) {
			uint32_t w = av[i].w;
			uint32_t j, nw = gfa_arc_n(g, w);
			gfa_arc_t *aw = gfa_arc_a(g, w);
			for (j = 0; j < nw && (j == 0 || gfa_arc_len(aw[j]) < fuzz); ++j)
				if (mark[aw[j].w]) mark[aw[j].v] = 2;
		}
		#endif
		for (i = 0; i < nv; ++i) {
			if (mark[av[i].w] == 2) av[i].del = 1, ++n_reduced;
			mark[av[i].w] = 0;
		}
	}
	free(mark);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] transitively reduced %d arcs\n", n_reduced);
	if (n_reduced) {
		gfa_cleanup(g);
		gfa_symm(g);
	}
	return n_reduced;
}

/**********************************
 * Filter short potential unitigs *
 **********************************/

#define GFA_ET_MERGEABLE 0
#define GFA_ET_TIP       1
#define GFA_ET_MULTI_OUT 2
#define GFA_ET_MULTI_NEI 3

static inline int gfa_is_utg_end(const gfa_t *g, uint32_t v, uint64_t *lw)
{
	uint32_t w, nv, nw, nw0, nv0 = gfa_arc_n(g, v^1);
	int i, i0 = -1;
	gfa_arc_t *aw, *av = gfa_arc_a(g, v^1);
	for (i = nv = 0; i < nv0; ++i)
		if (!av[i].del) i0 = i, ++nv;
	if (nv == 0) return GFA_ET_TIP; // tip
	if (nv > 1) return GFA_ET_MULTI_OUT; // multiple outgoing arcs
	if (lw) *lw = av[i0].v_lv<<32 | av[i0].w;
	w = av[i0].w ^ 1;
	nw0 = gfa_arc_n(g, w);
	aw = gfa_arc_a(g, w);
	for (i = nw = 0; i < nw0; ++i)
		if (!aw[i].del) ++nw;
	if (nw != 1) return GFA_ET_MULTI_NEI;
	return GFA_ET_MERGEABLE;
}

int gfa_extend(const gfa_t *g, uint32_t v, int max_ext, gfa64_v *a)
{
	int ret;
	uint64_t lw;
	a->n = 0;
	kv_push(uint64_t, *a, v);
	do {
		ret = gfa_is_utg_end(g, v^1, &lw);
		if (ret != 0) break;
		kv_push(uint64_t, *a, lw);
		v = (uint32_t)lw;
	} while (--max_ext > 0);
	return ret;
}

int gfa_cut_tip(gfa_t *g, int max_ext)
{
	gfa64_v a = {0,0,0};
	uint32_t n_vtx = gfa_n_vtx(g), v, i, cnt = 0;
	for (v = 0; v < n_vtx; ++v) {
		if (g->seg[v>>1].del) continue;
		if (gfa_is_utg_end(g, v, 0) != GFA_ET_TIP) continue; // not a tip
		if (gfa_extend(g, v, max_ext, &a) == GFA_ET_MERGEABLE) continue; // not a short unitig
		for (i = 0; i < a.n; ++i)
			gfa_seg_del(g, (uint32_t)a.a[i]>>1);
		++cnt;
	}
	free(a.a);
	if (cnt > 0) gfa_cleanup(g);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] cut %d tips\n", cnt);
	return cnt;
}

int gfa_cut_internal(gfa_t *g, int max_ext)
{
	gfa64_v a = {0,0,0};
	uint32_t n_vtx = gfa_n_vtx(g), v, i, cnt = 0;
	for (v = 0; v < n_vtx; ++v) {
		if (g->seg[v>>1].del) continue;
		if (gfa_is_utg_end(g, v, 0) != GFA_ET_MULTI_NEI) continue;
		if (gfa_extend(g, v, max_ext, &a) != GFA_ET_MULTI_NEI) continue;
		for (i = 0; i < a.n; ++i)
			gfa_seg_del(g, (uint32_t)a.a[i]>>1);
		++cnt;
	}
	free(a.a);
	if (cnt > 0) gfa_cleanup(g);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] cut %d internal sequences\n", cnt);
	return cnt;
}

int gfa_cut_biloop(gfa_t *g, int max_ext)
{
	gfa64_v a = {0,0,0};
	uint32_t n_vtx = gfa_n_vtx(g), v, i, cnt = 0;
	for (v = 0; v < n_vtx; ++v) {
		uint32_t nv, nw, w = UINT32_MAX, x, ov = 0, ox = 0;
		gfa_arc_t *av, *aw;
		if (g->seg[v>>1].del) continue;
		if (gfa_is_utg_end(g, v, 0) != GFA_ET_MULTI_NEI) continue;
		if (gfa_extend(g, v, max_ext, &a) != GFA_ET_MULTI_OUT) continue;
		x = (uint32_t)a.a[a.n - 1] ^ 1;
		nv = gfa_arc_n(g, v ^ 1), av = gfa_arc_a(g, v ^ 1);
		for (i = 0; i < nv; ++i)
			if (!av[i].del) w = av[i].w ^ 1;
		assert(w != UINT32_MAX);
		nw = gfa_arc_n(g, w), aw = gfa_arc_a(g, w);
		for (i = 0; i < nw; ++i) { // we are looking for: v->...->x', w->v and w->x
			if (aw[i].del) continue;
			if (aw[i].w == x) ox = aw[i].ov;
			if (aw[i].w == v) ov = aw[i].ov;
		}
		if (ov == 0 && ox == 0) continue;
		if (ov > ox) {
			gfa_arc_del(g, w, x, 1);
			gfa_arc_del(g, x^1, w^1, 1);
			++cnt;
		}
	}
	free(a.a);
	if (cnt > 0) gfa_cleanup(g);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] cut %d small bi-loops\n", cnt);
	return cnt;
}

/******************
 * Bubble popping *
 ******************/

typedef struct {
	uint32_t p; // the optimal parent vertex
	uint32_t d; // the shortest distance from the initial vertex
	uint32_t c; // max count of reads
	uint32_t r:31, s:1; // r: the number of remaining incoming arc; s: state
} binfo_t;

typedef struct {
	binfo_t *a;
	kvec_t(uint32_t) S; // set of vertices without parents
	kvec_t(uint32_t) T; // set of tips
	kvec_t(uint32_t) b; // visited vertices
	kvec_t(uint32_t) e; // visited edges/arcs
} buf_t;

// count the number of outgoing arcs, excluding reduced arcs
static inline int count_out(const gfa_t *g, uint32_t v)
{
	uint32_t i, n, nv = gfa_arc_n(g, v);
	const gfa_arc_t *av = gfa_arc_a(g, v);
	for (i = n = 0; i < nv; ++i)
		if (!av[i].del) ++n;
	return n;
}

// in a resolved bubble, mark unused vertices and arcs as "reduced"
static void gfa_bub_backtrack(gfa_t *g, uint32_t v0, buf_t *b)
{
	uint32_t i, v;
	assert(b->S.n == 1);
	for (i = 0; i < b->b.n; ++i)
		g->seg[b->b.a[i]>>1].del = 1;
	for (i = 0; i < b->e.n; ++i) {
		gfa_arc_t *a = &g->arc[b->e.a[i]];
		a->del = 1;
		gfa_arc_del(g, a->w^1, a->v_lv>>32^1, 1);
	}
	v = b->S.a[0];
	do {
		uint32_t u = b->a[v].p; // u->v
		g->seg[v>>1].del = 0;
		gfa_arc_del(g, u, v, 0);
		gfa_arc_del(g, v^1, u^1, 0);
		v = u;
	} while (v != v0);
}

// pop bubbles from vertex v0; the graph MJUST BE symmetric: if u->v present, v'->u' must be present as well
static uint64_t gfa_bub_pop1(gfa_t *g, uint32_t v0, int max_dist, buf_t *b)
{
	uint32_t i, n_pending = 0;
	uint64_t n_pop = 0;
	if (g->seg[v0>>1].del) return 0; // already deleted
	if ((uint32_t)g->idx[v0] < 2) return 0; // no bubbles
	b->S.n = b->T.n = b->b.n = b->e.n = 0;
	b->a[v0].c = b->a[v0].d = 0;
	kv_push(uint32_t, b->S, v0);
	do {
		uint32_t v = kv_pop(b->S), d = b->a[v].d, c = b->a[v].c;
		uint32_t nv = gfa_arc_n(g, v);
		gfa_arc_t *av = gfa_arc_a(g, v);
		assert(nv > 0);
		for (i = 0; i < nv; ++i) { // loop through v's neighbors
			uint32_t w = av[i].w, l = (uint32_t)av[i].v_lv; // u->w with length l
			binfo_t *t = &b->a[w];
			if (w == v0) goto pop_reset;
			if (av[i].del) continue;
			kv_push(uint32_t, b->e, (g->idx[v]>>32) + i);
			if (d + l > max_dist) break; // too far
			if (t->s == 0) { // this vertex has never been visited
				kv_push(uint32_t, b->b, w); // save it for revert
				t->p = v, t->s = 1, t->d = d + l;
				t->r = count_out(g, w^1);
				++n_pending;
			} else { // visited before
				if (c + 1 > t->c || (c + 1 == t->c && d + l > t->d)) t->p = v;
				if (c + 1 > t->c) t->c = c + 1;
				if (d + l < t->d) t->d = d + l; // update dist
			}
			assert(t->r > 0);
			if (--(t->r) == 0) {
				uint32_t x = gfa_arc_n(g, w);
				if (x) kv_push(uint32_t, b->S, w);
				else kv_push(uint32_t, b->T, w); // a tip
				--n_pending;
			}
		}
		if (i < nv || b->S.n == 0) goto pop_reset;
	} while (b->S.n > 1 || n_pending);
	gfa_bub_backtrack(g, v0, b);
	n_pop = 1 | (uint64_t)b->T.n<<32;
pop_reset:
	for (i = 0; i < b->b.n; ++i) { // clear the states of visited vertices
		binfo_t *t = &b->a[b->b.a[i]];
		t->s = t->c = t->d = 0;
	}
	return n_pop;
}

// pop bubbles
int gfa_pop_bubble(gfa_t *g, int max_dist)
{
	uint32_t v, n_vtx = gfa_n_vtx(g);
	uint64_t n_pop = 0;
	buf_t b;
	if (!g->is_symm) gfa_symm(g);
	memset(&b, 0, sizeof(buf_t));
	b.a = (binfo_t*)calloc(n_vtx, sizeof(binfo_t));
	for (v = 0; v < n_vtx; ++v) {
		uint32_t i, n_arc = 0, nv = gfa_arc_n(g, v);
		gfa_arc_t *av = gfa_arc_a(g, v);
		if (nv < 2 || g->seg[v>>1].del) continue;
		for (i = 0; i < nv; ++i) // gfa_bub_pop1() may delete some edges/arcs
			if (!av[i].del) ++n_arc;
		if (n_arc > 1)
			n_pop += gfa_bub_pop1(g, v, max_dist, &b);
	}
	free(b.a); free(b.S.a); free(b.T.a); free(b.b.a); free(b.e.a);
	if (n_pop) gfa_cleanup(g);
	if (gfa_verbose >= 3) fprintf(stderr, "[M] popped %d bubbles and trimmed %d tips\n", (uint32_t)n_pop, (uint32_t)(n_pop>>32));
	return n_pop;
}

/************
 * subgraph *
 ************/

int32_t gfa_name2id(const gfa_t *g, const char *name)
{
	seghash_t *h = (seghash_t*)g->h_names;
	khint_t k;
	k = kh_get(seg, h, name);
	return k == kh_end(h)? -1 : kh_val(h, k);
}

void gfa_sub(gfa_t *g, int n, char *const* seg, int step)
{
	int32_t i;
	kvec_t(uint64_t) stack = {0,0,0};
	for (i = 0; i < n; ++i) {
		int32_t s;
		s = gfa_name2id(g, seg[i]);
		if (s >= 0) {
			kv_push(uint64_t, stack, (uint64_t)(s<<1|0)<<32);
			kv_push(uint64_t, stack, (uint64_t)(s<<1|1)<<32);
		}
	}
	for (i = 0; i < g->n_seg; ++i) // mark all segments to be deleted
		g->seg[i].del = 1;
	while (stack.n) {
		uint64_t x = kv_pop(stack);
		uint32_t v = x>>32, r = (uint32_t)x;
		if (g->seg[v>>1].del == 0) continue;
		g->seg[v>>1].del = 0;
		if (r < step) {
			uint32_t nv = gfa_arc_n(g, v);
			gfa_arc_t *av = gfa_arc_a(g, v);
			for (i = 0; i < nv; ++i)
				if (g->seg[av[i].w>>1].del)
					kv_push(uint64_t, stack, (uint64_t)av[i].w<<32 | (r + 1));
		}
	}
	free(stack.a);
	gfa_arc_rm(g);
}

/****************
 * Unitig graph *
 ****************/

#include "kdq.h"
KDQ_INIT(uint64_t)

#define arc_cnt(g, v) ((uint32_t)(g)->idx[(v)])
#define arc_first(g, v) ((g)->arc[(g)->idx[(v)]>>32])

gfa_t *gfa_ug_gen(const gfa_t *g)
{
	int32_t *mark;
	uint32_t i, v, n_vtx = gfa_n_vtx(g);
	kdq_t(uint64_t) *q;
	gfa_t *ug;

	ug = gfa_init();
	mark = (int32_t*)calloc(n_vtx, 4);

	q = kdq_init(uint64_t);
	for (v = 0; v < n_vtx; ++v) {
		uint32_t w, x, l, start, end, len, tmp, len_r;
		char utg_name[11];
		gfa_seg_t *u;
		gfa_arc_t *a;

		if (g->seg[v>>1].del || arc_cnt(g, v) == 0 || mark[v]) continue;
		mark[v] = 1;
		q->count = 0, start = v, end = v^1, len = len_r = 0;
		// forward
		w = v;
		while (1) {
			if (arc_cnt(g, w) != 1) break;
			x = arc_first(g, w).w; // w->x
			if (arc_cnt(g, x^1) != 1) break;
			mark[x] = mark[w^1] = 1;
			a = &arc_first(g, w);
			l = gfa_arc_len(*a);
			kdq_push(uint64_t, q, (uint64_t)w<<32 | l);
			end = x^1, len += l, len_r += a->lw;
			w = x;
			if (x == v) break;
		}
		if (start != (end^1) || kdq_size(q) == 0) { // linear unitig
			l = g->seg[end>>1].len;
			kdq_push(uint64_t, q, (uint64_t)(end^1)<<32 | l);
			len += l;
		} else { // circular unitig
			start = end = UINT32_MAX;
			goto add_unitig; // then it is not necessary to do the backward
		}
		// backward
		x = v;
		while (1) { // similar to forward but not the same
			if (arc_cnt(g, x^1) != 1) break;
			w = arc_first(g, x^1).w ^ 1; // w->x
			if (arc_cnt(g, w) != 1) break;
			mark[x] = mark[w^1] = 1;
			a = &arc_first(g, w);
			l = gfa_arc_len(*a);
			kdq_unshift(uint64_t, q, (uint64_t)w<<32 | l);
			start = w, len += l, len_r += a->lw;
			x = w;
		}
		len_r += g->seg[start>>1].len;
add_unitig:
		if (start != UINT32_MAX) mark[start] = mark[end] = 1;
		sprintf(utg_name, "utg%.7d", ug->n_seg + 1);
		tmp = gfa_add_seg(ug, utg_name);
		u = &ug->seg[tmp];
		u->seq = 0, u->len = len;
		u->utg.start = start, u->utg.end = end, u->utg.n = kdq_size(q), u->circ = (start == UINT32_MAX);
		u->utg.m = u->utg.n;
		kv_roundup32(u->utg.m);
		u->utg.a = (uint64_t*)malloc(8 * u->utg.m);
		u->utg.name = (char**)malloc(sizeof(char*) * u->utg.m);
		u->utg.len2 = len_r;
		for (i = 0; i < kdq_size(q); ++i) {
			u->utg.a[i] = kdq_at(q, i);
			u->utg.name[i] = strdup(g->seg[u->utg.a[i]>>33].name);
		}
	}
	kdq_destroy(uint64_t, q);

	// add arcs between unitigs; reusing mark for a different purpose
	for (v = 0; v < n_vtx; ++v) mark[v] = -1;
	for (i = 0; i < ug->n_seg; ++i) {
		if (ug->seg[i].circ) continue;
		mark[ug->seg[i].utg.start] = i<<1 | 0;
		mark[ug->seg[i].utg.end] = i<<1 | 1;
	}
	for (i = 0; i < g->n_arc; ++i) {
		gfa_arc_t *p = &g->arc[i];
		if (p->del) continue;
		if (mark[p->v_lv>>32^1] >= 0 && mark[p->w] >= 0) {
			gfa_seg_t *s1 = &ug->seg[mark[p->v_lv>>32^1]>>1];
			gfa_seg_t *s2 = &ug->seg[mark[p->w]>>1];
			int ov = p->ov;
			if (ov >= s1->len || ov >= s2->len)
				ov = (s1->len < s2->len? s1->len : s2->len) - 1;
			gfa_add_arc1(ug, mark[p->v_lv>>32^1]^1, mark[p->w], ov, INT32_MAX, -1, 0);
		}
	}
	free(mark);
	gfa_arc_sort(ug);
	gfa_arc_index(ug);
	gfa_fix_semi_arc(ug);
	gfa_fix_arc_len(ug);
	gfa_cleanup(ug);
	return ug;
}