bam_index.c

#include <ctype.h>
#include <assert.h>
#include "bam.h"
#include "khash.h"
#include "ksort.h"
#include "bam_endian.h"
#ifdef _USE_KNETFILE
#include "knetfile.h"
#endif

/*!
  @header

  Alignment indexing. Before indexing, BAM must be sorted based on the
  leftmost coordinate of alignments. In indexing, BAM uses two indices:
  a UCSC binning index and a simple linear index. The binning index is
  efficient for alignments spanning long distance, while the auxiliary
  linear index helps to reduce unnecessary seek calls especially for
  short alignments.

  The UCSC binning scheme was suggested by Richard Durbin and Lincoln
  Stein and is explained by Kent et al. (2002). In this scheme, each bin
  represents a contiguous genomic region which can be fully contained in
  another bin; each alignment is associated with a bin which represents
  the smallest region containing the entire alignment. The binning
  scheme is essentially another representation of R-tree. A distinct bin
  uniquely corresponds to a distinct internal node in a R-tree. Bin A is
  a child of Bin B if region A is contained in B.

  In BAM, each bin may span 2^29, 2^26, 2^23, 2^20, 2^17 or 2^14 bp. Bin
  0 spans a 512Mbp region, bins 1-8 span 64Mbp, 9-72 8Mbp, 73-584 1Mbp,
  585-4680 128Kbp and bins 4681-37449 span 16Kbp regions. If we want to
  find the alignments overlapped with a region [rbeg,rend), we need to
  calculate the list of bins that may be overlapped the region and test
  the alignments in the bins to confirm the overlaps. If the specified
  region is short, typically only a few alignments in six bins need to
  be retrieved. The overlapping alignments can be quickly fetched.

 */

#define BAM_MIN_CHUNK_GAP 32768
// 1<<14 is the size of minimum bin.
#define BAM_LIDX_SHIFT    14

#define BAM_MAX_BIN 37450 // =(8^6-1)/7+1

typedef struct {
	uint64_t u, v;
} pair64_t;

#define pair64_lt(a,b) ((a).u < (b).u)
KSORT_INIT(off, pair64_t, pair64_lt)

typedef struct {
	uint32_t m, n;
	pair64_t *list;
} bam_binlist_t;

typedef struct {
	int32_t n, m;
	uint64_t *offset;
} bam_lidx_t;

KHASH_MAP_INIT_INT(i, bam_binlist_t)

struct __bam_index_t {
	int32_t n;
	uint64_t n_no_coor; // unmapped reads without coordinate
	khash_t(i) **index;
	bam_lidx_t *index2;
};

// requirement: len <= LEN_MASK
static inline void insert_offset(khash_t(i) *h, int bin, uint64_t beg, uint64_t end)
{
	khint_t k;
	bam_binlist_t *l;
	int ret;
	k = kh_put(i, h, bin, &ret);
	l = &kh_value(h, k);
	if (ret) { // not present
		l->m = 1; l->n = 0;
		l->list = (pair64_t*)calloc(l->m, 16);
	}
	if (l->n == l->m) {
		l->m <<= 1;
		l->list = (pair64_t*)realloc(l->list, l->m * 16);
	}
	l->list[l->n].u = beg; l->list[l->n++].v = end;
}

static inline void insert_offset2(bam_lidx_t *index2, bam1_t *b, uint64_t offset)
{
	int i, beg, end;
	beg = b->core.pos >> BAM_LIDX_SHIFT;
	end = (bam_calend(&b->core, bam1_cigar(b)) - 1) >> BAM_LIDX_SHIFT;
	if (index2->m < end + 1) {
		int old_m = index2->m;
		index2->m = end + 1;
		kroundup32(index2->m);
		index2->offset = (uint64_t*)realloc(index2->offset, index2->m * 8);
		memset(index2->offset + old_m, 0, 8 * (index2->m - old_m));
	}
	if (beg == end) {
		if (index2->offset[beg] == 0) index2->offset[beg] = offset;
	} else {
		for (i = beg; i <= end; ++i)
			if (index2->offset[i] == 0) index2->offset[i] = offset;
	}
	index2->n = end + 1;
}

static void merge_chunks(bam_index_t *idx)
{
#if defined(BAM_TRUE_OFFSET) || defined(BAM_VIRTUAL_OFFSET16)
	khash_t(i) *index;
	int i, l, m;
	khint_t k;
	for (i = 0; i < idx->n; ++i) {
		index = idx->index[i];
		for (k = kh_begin(index); k != kh_end(index); ++k) {
			bam_binlist_t *p;
			if (!kh_exist(index, k) || kh_key(index, k) == BAM_MAX_BIN) continue;
			p = &kh_value(index, k);
			m = 0;
			for (l = 1; l < p->n; ++l) {
#ifdef BAM_TRUE_OFFSET
				if (p->list[m].v + BAM_MIN_CHUNK_GAP > p->list[l].u) p->list[m].v = p->list[l].v;
#else
				if (p->list[m].v>>16 == p->list[l].u>>16) p->list[m].v = p->list[l].v;
#endif
				else p->list[++m] = p->list[l];
			} // ~for(l)
			p->n = m + 1;
		} // ~for(k)
	} // ~for(i)
#endif // defined(BAM_TRUE_OFFSET) || defined(BAM_BGZF)
}

static void fill_missing(bam_index_t *idx)
{
	int i, j;
	for (i = 0; i < idx->n; ++i) {
		bam_lidx_t *idx2 = &idx->index2[i];
		for (j = 1; j < idx2->n; ++j)
			if (idx2->offset[j] == 0)
				idx2->offset[j] = idx2->offset[j-1];
	}
}

bam_index_t *bam_index_core(bamFile fp)
{
	bam1_t *b;
	bam_header_t *h;
	int i, ret;
	bam_index_t *idx;
	uint32_t last_bin, save_bin, recalculated_bin;
	int32_t last_coor, last_tid, save_tid;
	bam1_core_t *c;
	uint64_t save_off, last_off, n_mapped, n_unmapped, off_beg, off_end, n_no_coor;

	h = bam_header_read(fp);
	if(h == 0) {
	    fprintf(stderr, "[bam_index_core] Invalid BAM header.");
	    return NULL;
	}

	idx = (bam_index_t*)calloc(1, sizeof(bam_index_t));
	b = (bam1_t*)calloc(1, sizeof(bam1_t));
	c = &b->core;

	idx->n = h->n_targets;
	idx->index = (khash_t(i)**)calloc(idx->n, sizeof(void*));
	for (i = 0; i < idx->n; ++i) idx->index[i] = kh_init(i);
	idx->index2 = (bam_lidx_t*)calloc(idx->n, sizeof(bam_lidx_t));

	save_bin = save_tid = last_tid = last_bin = 0xffffffffu;
	save_off = last_off = bam_tell(fp); last_coor = 0xffffffffu;
	n_mapped = n_unmapped = n_no_coor = off_end = 0;
	off_beg = off_end = bam_tell(fp);
	while ((ret = bam_read1(fp, b)) >= 0) {
		if (c->tid < 0) ++n_no_coor;
		if (last_tid < c->tid || (last_tid >= 0 && c->tid < 0)) { // change of chromosomes
			last_tid = c->tid;
			last_bin = 0xffffffffu;
		} else if ((uint32_t)last_tid > (uint32_t)c->tid) {
			fprintf(stderr, "[bam_index_core] the alignment is not sorted (%s): %d-th chr > %d-th chr\n",
					bam1_qname(b), last_tid+1, c->tid+1);
			return NULL;
		} else if ((int32_t)c->tid >= 0 && last_coor > c->pos) {
			fprintf(stderr, "[bam_index_core] the alignment is not sorted (%s): %u > %u in %d-th chr\n",
					bam1_qname(b), last_coor, c->pos, c->tid+1);
			return NULL;
		}
		if (c->tid >= 0) {
			if (c->n_cigar) {
				recalculated_bin = bam_reg2bin(c->pos, bam_calend(c, bam1_cigar(b)));
			} else {
				recalculated_bin = bam_reg2bin(c->pos, c->pos + 1);
			}
			if (c->bin != recalculated_bin) {
				fprintf(stderr, "[bam_index_core] read '%s' mapped to '%s' at POS %d to %d has BIN %d but should be %d\n",
					bam1_qname(b), h->target_name[c->tid], c->pos + 1, bam_calend(c, bam1_cigar(b)), c->bin, recalculated_bin);
				fprintf(stderr, "[bam_index_core] Fix it by using BAM->SAM->BAM to force a recalculation of the BIN field\n");
				return NULL;
			}
		}
		if (c->tid >= 0 && !(c->flag & BAM_FUNMAP)) insert_offset2(&idx->index2[b->core.tid], b, last_off);
		if (c->bin != last_bin) { // then possibly write the binning index
			if (save_bin != 0xffffffffu) // save_bin==0xffffffffu only happens to the first record
				insert_offset(idx->index[save_tid], save_bin, save_off, last_off);
			if (last_bin == 0xffffffffu && save_tid != 0xffffffffu) { // write the meta element
				off_end = last_off;
				insert_offset(idx->index[save_tid], BAM_MAX_BIN, off_beg, off_end);
				insert_offset(idx->index[save_tid], BAM_MAX_BIN, n_mapped, n_unmapped);
				n_mapped = n_unmapped = 0;
				off_beg = off_end;
			}
			save_off = last_off;
			save_bin = last_bin = c->bin;
			save_tid = c->tid;
			if (save_tid < 0) break;
		}
		if (bam_tell(fp) <= last_off) {
			fprintf(stderr, "[bam_index_core] bug in BGZF/RAZF: %llx < %llx\n",
					(unsigned long long)bam_tell(fp), (unsigned long long)last_off);
			return NULL;
		}
		if (c->flag & BAM_FUNMAP) ++n_unmapped;
		else ++n_mapped;
		last_off = bam_tell(fp);
		last_coor = b->core.pos;
	}
        bam_header_destroy(h);
	if (save_tid >= 0) {
		insert_offset(idx->index[save_tid], save_bin, save_off, bam_tell(fp));
		insert_offset(idx->index[save_tid], BAM_MAX_BIN, off_beg, bam_tell(fp));
		insert_offset(idx->index[save_tid], BAM_MAX_BIN, n_mapped, n_unmapped);
	}
	merge_chunks(idx);
	fill_missing(idx);
	if (ret >= 0) {
		while ((ret = bam_read1(fp, b)) >= 0) {
			++n_no_coor;
			if (c->tid >= 0 && n_no_coor) {
				fprintf(stderr, "[bam_index_core] the alignment is not sorted: reads without coordinates prior to reads with coordinates.\n");
				return NULL;
			}
		}
	}
	if (ret < -1) fprintf(stderr, "[bam_index_core] truncated file? Continue anyway. (%d)\n", ret);
	free(b->data); free(b);
	idx->n_no_coor = n_no_coor;
	return idx;
}

void bam_index_destroy(bam_index_t *idx)
{
	khint_t k;
	int i;
	if (idx == 0) return;
	for (i = 0; i < idx->n; ++i) {
		khash_t(i) *index = idx->index[i];
		bam_lidx_t *index2 = idx->index2 + i;
		for (k = kh_begin(index); k != kh_end(index); ++k) {
			if (kh_exist(index, k))
				free(kh_value(index, k).list);
		}
		kh_destroy(i, index);
		free(index2->offset);
	}
	free(idx->index); free(idx->index2);
	free(idx);
}

void bam_index_save(const bam_index_t *idx, FILE *fp)
{
	int32_t i, size;
	khint_t k;
	fwrite("BAI\1", 1, 4, fp);
	if (bam_is_be) {
		uint32_t x = idx->n;
		fwrite(bam_swap_endian_4p(&x), 4, 1, fp);
	} else fwrite(&idx->n, 4, 1, fp);
	for (i = 0; i < idx->n; ++i) {
		khash_t(i) *index = idx->index[i];
		bam_lidx_t *index2 = idx->index2 + i;
		// write binning index
		size = kh_size(index);
		if (bam_is_be) { // big endian
			uint32_t x = size;
			fwrite(bam_swap_endian_4p(&x), 4, 1, fp);
		} else fwrite(&size, 4, 1, fp);
		for (k = kh_begin(index); k != kh_end(index); ++k) {
			if (kh_exist(index, k)) {
				bam_binlist_t *p = &kh_value(index, k);
				if (bam_is_be) { // big endian
					uint32_t x;
					x = kh_key(index, k); fwrite(bam_swap_endian_4p(&x), 4, 1, fp);
					x = p->n; fwrite(bam_swap_endian_4p(&x), 4, 1, fp);
					for (x = 0; (int)x < p->n; ++x) {
						bam_swap_endian_8p(&p->list[x].u);
						bam_swap_endian_8p(&p->list[x].v);
					}
					fwrite(p->list, 16, p->n, fp);
					for (x = 0; (int)x < p->n; ++x) {
						bam_swap_endian_8p(&p->list[x].u);
						bam_swap_endian_8p(&p->list[x].v);
					}
				} else {
					fwrite(&kh_key(index, k), 4, 1, fp);
					fwrite(&p->n, 4, 1, fp);
					fwrite(p->list, 16, p->n, fp);
				}
			}
		}
		// write linear index (index2)
		if (bam_is_be) {
			int x = index2->n;
			fwrite(bam_swap_endian_4p(&x), 4, 1, fp);
		} else fwrite(&index2->n, 4, 1, fp);
		if (bam_is_be) { // big endian
			int x;
			for (x = 0; (int)x < index2->n; ++x)
				bam_swap_endian_8p(&index2->offset[x]);
			fwrite(index2->offset, 8, index2->n, fp);
			for (x = 0; (int)x < index2->n; ++x)
				bam_swap_endian_8p(&index2->offset[x]);
		} else fwrite(index2->offset, 8, index2->n, fp);
	}
	{ // write the number of reads coor-less records.
		uint64_t x = idx->n_no_coor;
		if (bam_is_be) bam_swap_endian_8p(&x);
		fwrite(&x, 8, 1, fp);
	}
	fflush(fp);
}

static bam_index_t *bam_index_load_core(FILE *fp)
{
	int i;
	char magic[4];
	bam_index_t *idx;
	if (fp == 0) {
		fprintf(stderr, "[bam_index_load_core] fail to load index.\n");
		return 0;
	}
	fread(magic, 1, 4, fp);
	if (strncmp(magic, "BAI\1", 4)) {
		fprintf(stderr, "[bam_index_load] wrong magic number.\n");
		fclose(fp);
		return 0;
	}
	idx = (bam_index_t*)calloc(1, sizeof(bam_index_t));	
	fread(&idx->n, 4, 1, fp);
	if (bam_is_be) bam_swap_endian_4p(&idx->n);
	idx->index = (khash_t(i)**)calloc(idx->n, sizeof(void*));
	idx->index2 = (bam_lidx_t*)calloc(idx->n, sizeof(bam_lidx_t));
	for (i = 0; i < idx->n; ++i) {
		khash_t(i) *index;
		bam_lidx_t *index2 = idx->index2 + i;
		uint32_t key, size;
		khint_t k;
		int j, ret;
		bam_binlist_t *p;
		index = idx->index[i] = kh_init(i);
		// load binning index
		fread(&size, 4, 1, fp);
		if (bam_is_be) bam_swap_endian_4p(&size);
		for (j = 0; j < (int)size; ++j) {
			fread(&key, 4, 1, fp);
			if (bam_is_be) bam_swap_endian_4p(&key);
			k = kh_put(i, index, key, &ret);
			p = &kh_value(index, k);
			fread(&p->n, 4, 1, fp);
			if (bam_is_be) bam_swap_endian_4p(&p->n);
			p->m = p->n;
			p->list = (pair64_t*)malloc(p->m * 16);
			fread(p->list, 16, p->n, fp);
			if (bam_is_be) {
				int x;
				for (x = 0; x < p->n; ++x) {
					bam_swap_endian_8p(&p->list[x].u);
					bam_swap_endian_8p(&p->list[x].v);
				}
			}
		}
		// load linear index
		fread(&index2->n, 4, 1, fp);
		if (bam_is_be) bam_swap_endian_4p(&index2->n);
		index2->m = index2->n;
		index2->offset = (uint64_t*)calloc(index2->m, 8);
		fread(index2->offset, index2->n, 8, fp);
		if (bam_is_be)
			for (j = 0; j < index2->n; ++j) bam_swap_endian_8p(&index2->offset[j]);
	}
	if (fread(&idx->n_no_coor, 8, 1, fp) == 0) idx->n_no_coor = 0;
	if (bam_is_be) bam_swap_endian_8p(&idx->n_no_coor);
	return idx;
}

bam_index_t *bam_index_load_local(const char *_fn)
{
	FILE *fp;
	char *fnidx, *fn;

	if (strstr(_fn, "ftp://") == _fn || strstr(_fn, "http://") == _fn) {
		const char *p;
		int l = strlen(_fn);
		for (p = _fn + l - 1; p >= _fn; --p)
			if (*p == '/') break;
		fn = strdup(p + 1);
	} else fn = strdup(_fn);
	fnidx = (char*)calloc(strlen(fn) + 5, 1);
	strcpy(fnidx, fn); strcat(fnidx, ".bai");
	fp = fopen(fnidx, "rb");
	if (fp == 0) { // try "{base}.bai"
		char *s = strstr(fn, "bam");
		if (s == fn + strlen(fn) - 3) {
			strcpy(fnidx, fn);
			fnidx[strlen(fn)-1] = 'i';
			fp = fopen(fnidx, "rb");
		}
	}
	free(fnidx); free(fn);
	if (fp) {
		bam_index_t *idx = bam_index_load_core(fp);
		fclose(fp);
		return idx;
	} else return 0;
}

#ifdef _USE_KNETFILE
static void download_from_remote(const char *url)
{
	const int buf_size = 1 * 1024 * 1024;
	char *fn;
	FILE *fp;
	uint8_t *buf;
	knetFile *fp_remote;
	int l;
	if (strstr(url, "ftp://") != url && strstr(url, "http://") != url) return;
	l = strlen(url);
	for (fn = (char*)url + l - 1; fn >= url; --fn)
		if (*fn == '/') break;
	++fn; // fn now points to the file name
	fp_remote = knet_open(url, "r");
	if (fp_remote == 0) {
		fprintf(stderr, "[download_from_remote] fail to open remote file.\n");
		return;
	}
	if ((fp = fopen(fn, "wb")) == 0) {
		fprintf(stderr, "[download_from_remote] fail to create file in the working directory.\n");
		knet_close(fp_remote);
		return;
	}
	buf = (uint8_t*)calloc(buf_size, 1);
	while ((l = knet_read(fp_remote, buf, buf_size)) != 0)
		fwrite(buf, 1, l, fp);
	free(buf);
	fclose(fp);
	knet_close(fp_remote);
}
#else
static void download_from_remote(const char *url)
{
	return;
}
#endif

bam_index_t *bam_index_load(const char *fn)
{
	bam_index_t *idx;
	idx = bam_index_load_local(fn);
	if (idx == 0 && (strstr(fn, "ftp://") == fn || strstr(fn, "http://") == fn)) {
		char *fnidx = calloc(strlen(fn) + 5, 1);
		strcat(strcpy(fnidx, fn), ".bai");
		fprintf(stderr, "[bam_index_load] attempting to download the remote index file.\n");
		download_from_remote(fnidx);
        free(fnidx);
		idx = bam_index_load_local(fn);
	}
	if (idx == 0) fprintf(stderr, "[bam_index_load] fail to load BAM index.\n");
	return idx;
}

int bam_index_build2(const char *fn, const char *_fnidx)
{
	char *fnidx;
	FILE *fpidx;
	bamFile fp;
	bam_index_t *idx;
	if ((fp = bam_open(fn, "r")) == 0) {
		fprintf(stderr, "[bam_index_build2] fail to open the BAM file.\n");
		return -1;
	}
	idx = bam_index_core(fp);
	bam_close(fp);
	if(idx == 0) {
		fprintf(stderr, "[bam_index_build2] fail to index the BAM file.\n");
		return -1;
	}
	if (_fnidx == 0) {
		fnidx = (char*)calloc(strlen(fn) + 5, 1);
		strcpy(fnidx, fn); strcat(fnidx, ".bai");
	} else fnidx = strdup(_fnidx);
	fpidx = fopen(fnidx, "wb");
	if (fpidx == 0) {
		fprintf(stderr, "[bam_index_build2] fail to create the index file.\n");
		free(fnidx);
        bam_index_destroy(idx);
		return -1;
	}
	bam_index_save(idx, fpidx);
	bam_index_destroy(idx);
	fclose(fpidx);
	free(fnidx);
	return 0;
}

int bam_index_build(const char *fn)
{
	return bam_index_build2(fn, 0);
}

int bam_index(int argc, char *argv[])
{
	if (argc < 2) {
		fprintf(stderr, "Usage: samtools index <in.bam> [out.index]\n");
		return 1;
	}
	if (argc >= 3) bam_index_build2(argv[1], argv[2]);
	else bam_index_build(argv[1]);
	return 0;
}

int bam_idxstats(int argc, char *argv[])
{
	bam_index_t *idx;
	bam_header_t *header;
	bamFile fp;
	int i;
	if (argc < 2) {
		fprintf(stderr, "Usage: samtools idxstats <in.bam>\n");
		return 1;
	}
	fp = bam_open(argv[1], "r");
	if (fp == 0) { fprintf(stderr, "[%s] fail to open BAM.\n", __func__); return 1; }
	header = bam_header_read(fp);
	bam_close(fp);
	idx = bam_index_load(argv[1]);
	if (idx == 0) { fprintf(stderr, "[%s] fail to load the index.\n", __func__); return 1; }
	for (i = 0; i < idx->n; ++i) {
		khint_t k;
		khash_t(i) *h = idx->index[i];
		printf("%s\t%d", header->target_name[i], header->target_len[i]);
		k = kh_get(i, h, BAM_MAX_BIN);
		if (k != kh_end(h))
			printf("\t%llu\t%llu", (long long)kh_val(h, k).list[1].u, (long long)kh_val(h, k).list[1].v);
		else printf("\t0\t0");
		putchar('\n');
	}
	printf("*\t0\t0\t%llu\n", (long long)idx->n_no_coor);
	bam_header_destroy(header);
	bam_index_destroy(idx);
	return 0;
}

static inline int reg2bins(uint32_t beg, uint32_t end, uint16_t list[BAM_MAX_BIN])
{
	int i = 0, k;
	if (beg >= end) return 0;
	if (end >= 1u<<29) end = 1u<<29;
	--end;
	list[i++] = 0;
	for (k =    1 + (beg>>26); k <=    1 + (end>>26); ++k) list[i++] = k;
	for (k =    9 + (beg>>23); k <=    9 + (end>>23); ++k) list[i++] = k;
	for (k =   73 + (beg>>20); k <=   73 + (end>>20); ++k) list[i++] = k;
	for (k =  585 + (beg>>17); k <=  585 + (end>>17); ++k) list[i++] = k;
	for (k = 4681 + (beg>>14); k <= 4681 + (end>>14); ++k) list[i++] = k;
	return i;
}

static inline int is_overlap(uint32_t beg, uint32_t end, const bam1_t *b)
{
	uint32_t rbeg = b->core.pos;
	uint32_t rend = b->core.n_cigar? bam_calend(&b->core, bam1_cigar(b)) : b->core.pos + 1;
	return (rend > beg && rbeg < end);
}

struct __bam_iter_t {
	int from_first; // read from the first record; no random access
	int tid, beg, end, n_off, i, finished;
	uint64_t curr_off;
	pair64_t *off;
};

// bam_fetch helper function retrieves 
bam_iter_t bam_iter_query(const bam_index_t *idx, int tid, int beg, int end)
{
	uint16_t *bins;
	int i, n_bins, n_off;
	pair64_t *off;
	khint_t k;
	khash_t(i) *index;
	uint64_t min_off;
	bam_iter_t iter = 0;

	if (beg < 0) beg = 0;
	if (end < beg) return 0;
	// initialize iter
	iter = calloc(1, sizeof(struct __bam_iter_t));
	iter->tid = tid, iter->beg = beg, iter->end = end; iter->i = -1;
	//
	bins = (uint16_t*)calloc(BAM_MAX_BIN, 2);
	n_bins = reg2bins(beg, end, bins);
	index = idx->index[tid];
	if (idx->index2[tid].n > 0) {
		min_off = (beg>>BAM_LIDX_SHIFT >= idx->index2[tid].n)? idx->index2[tid].offset[idx->index2[tid].n-1]
			: idx->index2[tid].offset[beg>>BAM_LIDX_SHIFT];
		if (min_off == 0) { // improvement for index files built by tabix prior to 0.1.4
			int n = beg>>BAM_LIDX_SHIFT;
			if (n > idx->index2[tid].n) n = idx->index2[tid].n;
			for (i = n - 1; i >= 0; --i)
				if (idx->index2[tid].offset[i] != 0) break;
			if (i >= 0) min_off = idx->index2[tid].offset[i];
		}
	} else min_off = 0; // tabix 0.1.2 may produce such index files
	for (i = n_off = 0; i < n_bins; ++i) {
		if ((k = kh_get(i, index, bins[i])) != kh_end(index))
			n_off += kh_value(index, k).n;
	}
	if (n_off == 0) {
		free(bins); return iter;
	}
	off = (pair64_t*)calloc(n_off, 16);
	for (i = n_off = 0; i < n_bins; ++i) {
		if ((k = kh_get(i, index, bins[i])) != kh_end(index)) {
			int j;
			bam_binlist_t *p = &kh_value(index, k);
			for (j = 0; j < p->n; ++j)
				if (p->list[j].v > min_off) off[n_off++] = p->list[j];
		}
	}
	free(bins);
	if (n_off == 0) {
		free(off); return iter;
	}
	{
		bam1_t *b = (bam1_t*)calloc(1, sizeof(bam1_t));
		int l;
		ks_introsort(off, n_off, off);
		// resolve completely contained adjacent blocks
		for (i = 1, l = 0; i < n_off; ++i)
			if (off[l].v < off[i].v)
				off[++l] = off[i];
		n_off = l + 1;
		// resolve overlaps between adjacent blocks; this may happen due to the merge in indexing
		for (i = 1; i < n_off; ++i)
			if (off[i-1].v >= off[i].u) off[i-1].v = off[i].u;
		{ // merge adjacent blocks
#if defined(BAM_TRUE_OFFSET) || defined(BAM_VIRTUAL_OFFSET16)
			for (i = 1, l = 0; i < n_off; ++i) {
#ifdef BAM_TRUE_OFFSET
				if (off[l].v + BAM_MIN_CHUNK_GAP > off[i].u) off[l].v = off[i].v;
#else
				if (off[l].v>>16 == off[i].u>>16) off[l].v = off[i].v;
#endif
				else off[++l] = off[i];
			}
			n_off = l + 1;
#endif
		}
		bam_destroy1(b);
	}
	iter->n_off = n_off; iter->off = off;
	return iter;
}

pair64_t *get_chunk_coordinates(const bam_index_t *idx, int tid, int beg, int end, int *cnt_off)
{ // for pysam compatibility
	bam_iter_t iter;
	pair64_t *off;
	iter = bam_iter_query(idx, tid, beg, end);
	off = iter->off; *cnt_off = iter->n_off;
	free(iter);
	return off;
}

void bam_iter_destroy(bam_iter_t iter)
{
	if (iter) { free(iter->off); free(iter); }
}

int bam_iter_read(bamFile fp, bam_iter_t iter, bam1_t *b)
{
	int ret;
	if (iter && iter->finished) return -1;
	if (iter == 0 || iter->from_first) {
		ret = bam_read1(fp, b);
		if (ret < 0 && iter) iter->finished = 1;
		return ret;
	}
	if (iter->off == 0) return -1;
	for (;;) {
		if (iter->curr_off == 0 || iter->curr_off >= iter->off[iter->i].v) { // then jump to the next chunk
			if (iter->i == iter->n_off - 1) { ret = -1; break; } // no more chunks
			if (iter->i >= 0) assert(iter->curr_off == iter->off[iter->i].v); // otherwise bug
			if (iter->i < 0 || iter->off[iter->i].v != iter->off[iter->i+1].u) { // not adjacent chunks; then seek
				bam_seek(fp, iter->off[iter->i+1].u, SEEK_SET);
				iter->curr_off = bam_tell(fp);
			}
			++iter->i;
		}
		if ((ret = bam_read1(fp, b)) >= 0) {
			iter->curr_off = bam_tell(fp);
			if (b->core.tid != iter->tid || b->core.pos >= iter->end) { // no need to proceed
				ret = bam_validate1(NULL, b)? -1 : -5; // determine whether end of region or error
				break;
			}
			else if (is_overlap(iter->beg, iter->end, b)) return ret;
		} else break; // end of file or error
	}
	iter->finished = 1;
	return ret;
}

int bam_fetch(bamFile fp, const bam_index_t *idx, int tid, int beg, int end, void *data, bam_fetch_f func)
{
	int ret;
	bam_iter_t iter;
	bam1_t *b;
	b = bam_init1();
	iter = bam_iter_query(idx, tid, beg, end);
	while ((ret = bam_iter_read(fp, iter, b)) >= 0) func(b, data);
	bam_iter_destroy(iter);
	bam_destroy1(b);
	return (ret == -1)? 0 : ret;
}