synced_bcf_reader.c

/*  synced_bcf_reader.c -- stream through multiple VCF files.

    Copyright (C) 2012-2014 Genome Research Ltd.

    Author: Petr Danecek <pd3@sanger.ac.uk>

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.  */

#include <config.h>

#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <limits.h>
#include <errno.h>
#include <ctype.h>
#include <sys/stat.h>
#include "htslib/synced_bcf_reader.h"
#include "htslib/kseq.h"
#include "htslib/khash_str2int.h"
#include "htslib/bgzf.h"

#define MAX_CSI_COOR 0x7fffffff     // maximum indexable coordinate of .csi

typedef struct
{
    uint32_t start, end;
}
region1_t;

typedef struct _region_t
{
    region1_t *regs;
    int nregs, mregs, creg;
}
region_t;

static void _regions_add(bcf_sr_regions_t *reg, const char *chr, int start, int end);
static bcf_sr_regions_t *_regions_init_string(const char *str);
static int _regions_match_alleles(bcf_sr_regions_t *reg, int als_idx, bcf1_t *rec);

char *bcf_sr_strerror(int errnum)
{
    switch (errnum)
    {
        case open_failed: 
            return strerror(errno); break;
        case not_bgzf:
            return "not compressed with bgzip"; break;
        case idx_load_failed:
            return "could not load index"; break;
        case file_type_error:
            return "unknown file type"; break;
        case api_usage_error:
            return "API usage error"; break;
        case header_error:
            return "could not parse header"; break;
        case no_eof:
            return "no BGZF EOF marker; file may be truncated"; break;
        default: return ""; 
    }
}

static int *init_filters(bcf_hdr_t *hdr, const char *filters, int *nfilters)
{
    kstring_t str = {0,0,0};
    const char *tmp = filters, *prev = filters;
    int nout = 0, *out = NULL;
    while ( 1 )
    {
        if ( *tmp==',' || !*tmp )
        {
            out = (int*) realloc(out, (nout+1)*sizeof(int));
            if ( tmp-prev==1 && *prev=='.' )
            {
                out[nout] = -1;
                nout++;
            }
            else
            {
                str.l = 0;
                kputsn(prev, tmp-prev, &str);
                out[nout] = bcf_hdr_id2int(hdr, BCF_DT_ID, str.s);
                if ( out[nout]>=0 ) nout++;
            }
            if ( !*tmp ) break;
            prev = tmp+1;
        }
        tmp++;
    }
    if ( str.m ) free(str.s);
    *nfilters = nout;
    return out;
}

int bcf_sr_set_regions(bcf_srs_t *readers, const char *regions, int is_file)
{
    assert( !readers->regions );
    if ( readers->nreaders )
    {
        fprintf(stderr,"[%s:%d %s] Error: bcf_sr_set_regions() must be called before bcf_sr_add_reader()\n", __FILE__,__LINE__,__FUNCTION__);
        return -1;
    }
    readers->regions = bcf_sr_regions_init(regions,is_file,0,1,-2);
    if ( !readers->regions ) return -1;
    readers->explicit_regs = 1;
    readers->require_index = 1;
    return 0;
}
int bcf_sr_set_targets(bcf_srs_t *readers, const char *targets, int is_file, int alleles)
{
    assert( !readers->targets );
    if ( targets[0]=='^' )
    {
        readers->targets_exclude = 1;
        targets++;
    }
    readers->targets = bcf_sr_regions_init(targets,is_file,0,1,-2);
    if ( !readers->targets ) return -1;
    readers->targets_als = alleles;
    return 0;
}

int bcf_sr_add_reader(bcf_srs_t *files, const char *fname)
{
    htsFile* file_ptr = hts_open(fname, "r");
    if ( ! file_ptr ) {
        files->errnum = open_failed;
        return 0;
    }

    files->has_line = (int*) realloc(files->has_line, sizeof(int)*(files->nreaders+1));
    files->has_line[files->nreaders] = 0;
    files->readers  = (bcf_sr_t*) realloc(files->readers, sizeof(bcf_sr_t)*(files->nreaders+1));
    bcf_sr_t *reader = &files->readers[files->nreaders++];
    memset(reader,0,sizeof(bcf_sr_t));

    reader->file = file_ptr;

    files->errnum = 0;

    if ( reader->file->format.compression==bgzf )
    {
        BGZF *bgzf = hts_get_bgzfp(reader->file);
        if ( bgzf && bgzf_check_EOF(bgzf) == 0 ) {
            files->errnum = no_eof;
            fprintf(stderr,"[%s] Warning: no BGZF EOF marker; file may be truncated.\n", fname);
        }
    }

    if ( files->require_index )
    {
        if ( reader->file->format.format==vcf )
        {
            if ( reader->file->format.compression!=bgzf )
            {
                files->errnum = not_bgzf;
                return 0;
            }

            reader->tbx_idx = tbx_index_load(fname);
            if ( !reader->tbx_idx )
            {
                files->errnum = idx_load_failed;
                return 0;
            }

            reader->header = bcf_hdr_read(reader->file);
        }
        else if ( reader->file->format.format==bcf )
        {
            if ( reader->file->format.compression!=bgzf )
            {
                files->errnum = not_bgzf;
                return 0;
            }

            reader->header = bcf_hdr_read(reader->file);

            reader->bcf_idx = bcf_index_load(fname);
            if ( !reader->bcf_idx )
            {
                files->errnum = idx_load_failed;
                return 0;
            }
        }
        else
        {
            files->errnum = file_type_error;
            return 0;
        }
    }
    else
    {
        if ( reader->file->format.format==bcf || reader->file->format.format==vcf )
        {
            reader->header = bcf_hdr_read(reader->file);
        }
        else
        {
            files->errnum = file_type_error;
            return 0;
        }
        files->streaming = 1;
    }
    if ( files->streaming && files->nreaders>1 )
    {
        files->errnum = api_usage_error;
        fprintf(stderr,"[%s:%d %s] Error: %d readers, yet require_index not set\n", __FILE__,__LINE__,__FUNCTION__,files->nreaders);
        return 0;
    }
    if ( files->streaming && files->regions )
    {
        files->errnum = api_usage_error;
        fprintf(stderr,"[%s:%d %s] Error: cannot tabix-jump in streaming mode\n", __FILE__,__LINE__,__FUNCTION__);
        return 0;
    }
    if ( !reader->header )
    {
        files->errnum = header_error;
        return 0;
    }

    reader->fname = strdup(fname);
    if ( files->apply_filters )
        reader->filter_ids = init_filters(reader->header, files->apply_filters, &reader->nfilter_ids);

    // Update list of chromosomes
    if ( !files->explicit_regs && !files->streaming )
    {
        int n,i;
        const char **names = reader->tbx_idx ? tbx_seqnames(reader->tbx_idx, &n) : bcf_hdr_seqnames(reader->header, &n);
        for (i=0; i<n; i++)
        {
            if ( !files->regions )
                files->regions = _regions_init_string(names[i]);
            else
                _regions_add(files->regions, names[i], -1, -1);
        }
        free(names);
    }

    return 1;
}

bcf_srs_t *bcf_sr_init(void)
{
    bcf_srs_t *files = (bcf_srs_t*) calloc(1,sizeof(bcf_srs_t));
    return files;
}

static void bcf_sr_destroy1(bcf_sr_t *reader)
{
    free(reader->fname);
    if ( reader->tbx_idx ) tbx_destroy(reader->tbx_idx);
    if ( reader->bcf_idx ) hts_idx_destroy(reader->bcf_idx);
    bcf_hdr_destroy(reader->header);
    hts_close(reader->file);
    if ( reader->itr ) tbx_itr_destroy(reader->itr);
    int j;
    for (j=0; j<reader->mbuffer; j++)
        bcf_destroy1(reader->buffer[j]);
    free(reader->buffer);
    free(reader->samples);
    free(reader->filter_ids);
}
void bcf_sr_destroy(bcf_srs_t *files)
{
    int i;
    for (i=0; i<files->nreaders; i++)
        bcf_sr_destroy1(&files->readers[i]);
    free(files->has_line);
    free(files->readers);
    for (i=0; i<files->n_smpl; i++) free(files->samples[i]);
    free(files->samples);
    if (files->targets) bcf_sr_regions_destroy(files->targets);
    if (files->regions) bcf_sr_regions_destroy(files->regions);
    if ( files->tmps.m ) free(files->tmps.s);
    free(files);
}

void bcf_sr_remove_reader(bcf_srs_t *files, int i)
{
    assert( !files->samples );  // not ready for this yet
    bcf_sr_destroy1(&files->readers[i]);
    if ( i+1 < files->nreaders )
    {
        memmove(&files->readers[i], &files->readers[i+1], (files->nreaders-i-1)*sizeof(bcf_sr_t));
        memmove(&files->has_line[i], &files->has_line[i+1], (files->nreaders-i-1)*sizeof(int));
    }
    files->nreaders--;
}


/*
   Removes duplicate records from the buffer. The meaning of "duplicate" is
   controlled by the $collapse variable, which can cause that from multiple
   <indel|snp|any> lines only the first is considered and the rest is ignored.
   The removal is done by setting the redundant lines' positions to -1 and
   moving these lines at the end of the buffer.
 */
static void collapse_buffer(bcf_srs_t *files, bcf_sr_t *reader)
{
    int irec,jrec, has_snp=0, has_indel=0, has_any=0;
    for (irec=1; irec<=reader->nbuffer; irec++)
    {
        bcf1_t *line = reader->buffer[irec];
        if ( line->pos != reader->buffer[1]->pos ) break;
        if ( files->collapse&COLLAPSE_ANY )
        {
            if ( !has_any ) has_any = 1;
            else line->pos = -1;
        }
        int line_type = bcf_get_variant_types(line);
        if ( files->collapse&COLLAPSE_SNPS && line_type&(VCF_SNP|VCF_MNP) )
        {
            if ( !has_snp ) has_snp = 1;
            else line->pos = -1;
        }
        if ( files->collapse&COLLAPSE_INDELS && line_type&VCF_INDEL )
        {
            if ( !has_indel ) has_indel = 1;
            else line->pos = -1;
        }
    }
    bcf1_t *tmp;
    irec = jrec = 1;
    while ( irec<=reader->nbuffer && jrec<=reader->nbuffer )
    {
        if ( reader->buffer[irec]->pos != -1 ) { irec++; continue; }
        if ( jrec<=irec ) jrec = irec+1;
        while ( jrec<=reader->nbuffer && reader->buffer[jrec]->pos==-1 ) jrec++;
        if ( jrec<=reader->nbuffer )
        {
            tmp = reader->buffer[irec]; reader->buffer[irec] = reader->buffer[jrec]; reader->buffer[jrec] = tmp;
        }
    }
    reader->nbuffer = irec - 1;
}

void debug_buffer(FILE *fp, bcf_sr_t *reader)
{
    int j;
    for (j=0; j<=reader->nbuffer; j++)
    {
        bcf1_t *line = reader->buffer[j];
        fprintf(fp,"\t%p\t%s%s\t%s:%d\t%s ", line,reader->fname,j==0?"*":" ",reader->header->id[BCF_DT_CTG][line->rid].key,line->pos+1,line->n_allele?line->d.allele[0]:"");
        int k;
        for (k=1; k<line->n_allele; k++) fprintf(fp," %s", line->d.allele[k]);
        fprintf(fp,"\n");
    }
}

void debug_buffers(FILE *fp, bcf_srs_t *files)
{
    int i;
    for (i=0; i<files->nreaders; i++)
    {
        fprintf(fp, "has_line: %d\t%s\n", bcf_sr_has_line(files,i),files->readers[i].fname);
        debug_buffer(fp, &files->readers[i]);
    }
    fprintf(fp,"\n");
}

static inline int has_filter(bcf_sr_t *reader, bcf1_t *line)
{
    int i, j;
    if ( !line->d.n_flt )
    {
        for (j=0; j<reader->nfilter_ids; j++)
            if ( reader->filter_ids[j]<0 ) return 1;
        return 0;
    }
    for (i=0; i<line->d.n_flt; i++)
    {
        for (j=0; j<reader->nfilter_ids; j++)
            if ( line->d.flt[i]==reader->filter_ids[j] ) return 1;
    }
    return 0;
}

static int _reader_seek(bcf_sr_t *reader, const char *seq, int start, int end)
{
    if ( end>=MAX_CSI_COOR )
    {
        fprintf(stderr,"The coordinate is out of csi index limit: %d\n", end+1);
        exit(1);
    }
    if ( reader->itr )
    {
        hts_itr_destroy(reader->itr);
        reader->itr = NULL;
    }
    reader->nbuffer = 0;
    if ( reader->tbx_idx )
    {
        int tid = tbx_name2id(reader->tbx_idx, seq);
        if ( tid==-1 ) return -1;    // the sequence not present in this file
        reader->itr = tbx_itr_queryi(reader->tbx_idx,tid,start,end+1);
    }
    else
    {
        int tid = bcf_hdr_name2id(reader->header, seq);
        if ( tid==-1 ) return -1;    // the sequence not present in this file
        reader->itr = bcf_itr_queryi(reader->bcf_idx,tid,start,end+1);
    }
    if ( !reader->itr ) fprintf(stderr,"Could not seek: %s:%d-%d\n",seq,start+1,end+1);
    assert(reader->itr);
    return 0;
}

/*
 *  _readers_next_region() - jumps to next region if necessary
 *  Returns 0 on success or -1 when there are no more regions left
 */
static int _readers_next_region(bcf_srs_t *files)
{
    // Need to open new chromosome? Check number of lines in all readers' buffers
    int i, eos = 0;
    for (i=0; i<files->nreaders; i++)
        if ( !files->readers[i].itr && !files->readers[i].nbuffer ) eos++;

    if ( eos!=files->nreaders )
    {
        // Some of the readers still has buffered lines
        return 0;
    }

    // No lines in the buffer, need to open new region or quit
    if ( bcf_sr_regions_next(files->regions)<0 ) return -1;

    for (i=0; i<files->nreaders; i++)
        _reader_seek(&files->readers[i],files->regions->seq_names[files->regions->iseq],files->regions->start,files->regions->end);

    return 0;
}

/*
 *  _reader_fill_buffer() - buffers all records with the same coordinate
 */
static void _reader_fill_buffer(bcf_srs_t *files, bcf_sr_t *reader)
{
    // Return if the buffer is full: the coordinate of the last buffered record differs
    if ( reader->nbuffer && reader->buffer[reader->nbuffer]->pos != reader->buffer[1]->pos ) return;

    // No iterator (sequence not present in this file) and not streaming
    if ( !reader->itr && !files->streaming ) return;

    // Fill the buffer with records starting at the same position
    int i, ret = 0;
    while (1)
    {
        if ( reader->nbuffer+1 >= reader->mbuffer )
        {
            // Increase buffer size
            reader->mbuffer += 8;
            reader->buffer = (bcf1_t**) realloc(reader->buffer, sizeof(bcf1_t*)*reader->mbuffer);
            for (i=8; i>0; i--)     // initialize
            {
                reader->buffer[reader->mbuffer-i] = bcf_init1();
                reader->buffer[reader->mbuffer-i]->max_unpack = files->max_unpack;
                reader->buffer[reader->mbuffer-i]->pos = -1;    // for rare cases when VCF starts from 1
            }
        }
        if ( files->streaming )
        {
            if ( reader->file->format.format==vcf )
            {
                if ( (ret=hts_getline(reader->file, KS_SEP_LINE, &files->tmps)) < 0 ) break;   // no more lines
                int ret = vcf_parse1(&files->tmps, reader->header, reader->buffer[reader->nbuffer+1]);
                if ( ret<0 ) break;
            }
            else if ( reader->file->format.format==bcf )
            {
                if ( (ret=bcf_read1(reader->file, reader->header, reader->buffer[reader->nbuffer+1])) < 0 ) break; // no more lines
            }
            else
            {
                fprintf(stderr,"[%s:%d %s] fixme: not ready for this\n", __FILE__,__LINE__,__FUNCTION__);
                exit(1);
            }
        }
        else if ( reader->tbx_idx )
        {
            if ( (ret=tbx_itr_next(reader->file, reader->tbx_idx, reader->itr, &files->tmps)) < 0 ) break;  // no more lines
            vcf_parse1(&files->tmps, reader->header, reader->buffer[reader->nbuffer+1]);
        }
        else
        {
            if ( (ret=bcf_itr_next(reader->file, reader->itr, reader->buffer[reader->nbuffer+1])) < 0 ) break; // no more lines
            bcf_subset_format(reader->header,reader->buffer[reader->nbuffer+1]);
        }

        // apply filter
        if ( !reader->nfilter_ids )
            bcf_unpack(reader->buffer[reader->nbuffer+1], BCF_UN_STR);
        else
        {
            bcf_unpack(reader->buffer[reader->nbuffer+1], BCF_UN_STR|BCF_UN_FLT);
            if ( !has_filter(reader, reader->buffer[reader->nbuffer+1]) ) continue;
        }
        reader->nbuffer++;

        if ( reader->buffer[reader->nbuffer]->pos != reader->buffer[1]->pos ) break;    // the buffer is full
    }
    if ( ret<0 )
    {
        // done for this region
        tbx_itr_destroy(reader->itr);
        reader->itr = NULL;
    }
    if ( files->collapse && reader->nbuffer>=2 && reader->buffer[1]->pos==reader->buffer[2]->pos )
        collapse_buffer(files, reader);
}

/*
 *  _readers_shift_buffer() - removes the first line and all subsequent lines with the same position
 */
static void _reader_shift_buffer(bcf_sr_t *reader)
{
    int i;
    for (i=2; i<=reader->nbuffer; i++)
        if ( reader->buffer[i]->pos!=reader->buffer[1]->pos ) break;
    if ( i<=reader->nbuffer )
    {
        // A record with a different position follows, swap it. Because of the reader's logic,
        // only one such line can be present.
        bcf1_t *tmp = reader->buffer[1]; reader->buffer[1] = reader->buffer[i]; reader->buffer[i] = tmp;
        reader->nbuffer = 1;
    }
    else
        reader->nbuffer = 0;    // no other line
}

/*
 *  _reader_match_alleles() - from multiple buffered lines selects the one which
 *  corresponds best to the template line. The logic is controlled by COLLAPSE_*
 *  Returns 0 on success or -1 when no good matching line is found.
 */
static int _reader_match_alleles(bcf_srs_t *files, bcf_sr_t *reader, bcf1_t *tmpl)
{
    int i, irec = -1;

    // if no template given, use the first available record
    if ( !tmpl )
        irec = 1;
    else
    {
        int tmpl_type = bcf_get_variant_types(tmpl);
        for (i=1; i<=reader->nbuffer; i++)
        {
            bcf1_t *line = reader->buffer[i];
            if ( line->pos != reader->buffer[1]->pos ) break;  // done with this reader

            // Easiest case: matching by position only
            if ( files->collapse&COLLAPSE_ANY ) { irec=i; break; }

            int line_type = bcf_get_variant_types(line);

            // No matter what the alleles are, as long as they are both SNPs
            if ( files->collapse&COLLAPSE_SNPS && tmpl_type&VCF_SNP && line_type&VCF_SNP ) { irec=i; break; }
            // ... or indels
            if ( files->collapse&COLLAPSE_INDELS && tmpl_type&VCF_INDEL && line_type&VCF_INDEL ) { irec=i; break; }

            // More thorough checking: REFs must match
            if ( tmpl->rlen != line->rlen ) continue;  // different length
            if ( !tmpl->d.allele || !line->d.allele ) continue;   // one of the lines is empty, someone is swapped buffered lines?!
            if ( strcmp(tmpl->d.allele[0], line->d.allele[0]) ) continue; // the strings do not match

            int ial,jal;
            if ( files->collapse==COLLAPSE_NONE )
            {
                // Exact match, all alleles must be identical
                if ( tmpl->n_allele!=line->n_allele ) continue;   // different number of alleles, skip

                int nmatch = 1; // REF has been already checked
                for (ial=1; ial<tmpl->n_allele; ial++)
                {
                    for (jal=1; jal<line->n_allele; jal++)
                        if ( !strcmp(tmpl->d.allele[ial], line->d.allele[jal]) ) { nmatch++; break; }
                }
                if ( nmatch==tmpl->n_allele ) { irec=i; break; }    // found: exact match
                continue;
            }

            if ( line->n_allele==1 && tmpl->n_allele==1 ) { irec=i; break; }    // both sites are non-variant

            // COLLAPSE_SOME: at least some ALTs must match
            for (ial=1; ial<tmpl->n_allele; ial++)
            {
                for (jal=1; jal<line->n_allele; jal++)
                    if ( !strcmp(tmpl->d.allele[ial], line->d.allele[jal]) ) { irec=i; break; }
                if ( irec>=1 ) break;
            }
            if ( irec>=1 ) break;
        }
        if ( irec==-1 ) return -1;  // no matching line was found
    }

    // Set the selected line (irec) as active: set it to buffer[0], move the remaining lines forward
    // and put the old bcf1_t record at the end.
    bcf1_t *tmp = reader->buffer[0];
    reader->buffer[0] = reader->buffer[irec];
    for (i=irec+1; i<=reader->nbuffer; i++) reader->buffer[i-1] = reader->buffer[i];
    reader->buffer[ reader->nbuffer ] = tmp;
    reader->nbuffer--;

    return 0;
}

int _reader_next_line(bcf_srs_t *files)
{
    int i, min_pos = INT_MAX;

    // Loop until next suitable line is found or all readers have finished
    while ( 1 )
    {
        // Get all readers ready for the next region.
        if ( files->regions && _readers_next_region(files)<0 ) break;

        // Fill buffers
        const char *chr = NULL;
        for (i=0; i<files->nreaders; i++)
        {
            _reader_fill_buffer(files, &files->readers[i]);

            // Update the minimum coordinate
            if ( !files->readers[i].nbuffer ) continue;
            if ( min_pos > files->readers[i].buffer[1]->pos )
            {
                min_pos = files->readers[i].buffer[1]->pos;
                chr = bcf_seqname(files->readers[i].header, files->readers[i].buffer[1]);
            }
        }
        if ( min_pos==INT_MAX )
        {
            if ( !files->regions ) break;
            continue;
        }

        // Skip this position if not present in targets
        if ( files->targets )
        {
            int ret = bcf_sr_regions_overlap(files->targets, chr, min_pos, min_pos);
            if ( (!files->targets_exclude && ret<0) || (files->targets_exclude && !ret) )
            {
                // Remove all lines with this position from the buffer
                for (i=0; i<files->nreaders; i++)
                    if ( files->readers[i].nbuffer && files->readers[i].buffer[1]->pos==min_pos )
                        _reader_shift_buffer(&files->readers[i]);
                min_pos = INT_MAX;
                continue;
            }
        }

        break;  // done: min_pos is set
    }

    // There can be records with duplicate positions. Set the active line intelligently so that
    // the alleles match.
    int nret = 0;   // number of readers sharing the position
    bcf1_t *first = NULL;   // record which will be used for allele matching
    for (i=0; i<files->nreaders; i++)
    {
        files->has_line[i] = 0;

        // Skip readers with no records at this position
        if ( !files->readers[i].nbuffer || files->readers[i].buffer[1]->pos!=min_pos ) continue;

        // Until now buffer[0] of all reader was empty and the lines started at buffer[1].
        // Now lines which are ready to be output will be moved to buffer[0].
        if ( _reader_match_alleles(files, &files->readers[i], first) < 0 ) continue;
        if ( !first ) first = files->readers[i].buffer[0];

        nret++;
        files->has_line[i] = 1;
    }
    return nret;
}

int bcf_sr_next_line(bcf_srs_t *files)
{
    if ( !files->targets_als )
        return _reader_next_line(files);

    while (1)
    {
        int i, ret = _reader_next_line(files);
        if ( !ret ) return ret;

        for (i=0; i<files->nreaders; i++)
            if ( files->has_line[i] ) break;

        if ( _regions_match_alleles(files->targets, files->targets_als-1, files->readers[i].buffer[0]) ) return ret;

        // Check if there are more duplicate lines in the buffers. If not, return this line as if it
        // matched the targets, even if there is a type mismatch
        for (i=0; i<files->nreaders; i++)
        {
            if ( !files->has_line[i] ) continue;
            if ( files->readers[i].nbuffer==0 || files->readers[i].buffer[1]->pos!=files->readers[i].buffer[0]->pos ) continue;
            break;
        }
        if ( i==files->nreaders ) return ret;   // no more lines left, output even if target alleles are not of the same type
    }
}

static void bcf_sr_seek_start(bcf_srs_t *readers)
{
    bcf_sr_regions_t *reg = readers->regions;
    int i;
    for (i=0; i<reg->nseqs; i++)
        reg->regs[i].creg = -1;
    reg->iseq = 0;
}


int bcf_sr_seek(bcf_srs_t *readers, const char *seq, int pos)
{
    if ( !seq && !pos )
    {
        // seek to start
        bcf_sr_seek_start(readers);
        return 0;
    }

    bcf_sr_regions_overlap(readers->regions, seq, pos, pos);
    int i, nret = 0;
    for (i=0; i<readers->nreaders; i++)
    {
        nret += _reader_seek(&readers->readers[i],seq,pos,MAX_CSI_COOR-1);
    }
    return nret;
}

int bcf_sr_set_samples(bcf_srs_t *files, const char *fname, int is_file)
{
    int i, j, nsmpl, free_smpl = 0;
    char **smpl = NULL;

    void *exclude = (fname[0]=='^') ? khash_str2int_init() : NULL;
    if ( exclude || strcmp("-",fname) ) // "-" stands for all samples
    {
        smpl = hts_readlist(fname, is_file, &nsmpl);
        if ( !smpl )
        {
            fprintf(stderr,"Could not read the file: \"%s\"\n", fname);
            return 0;
        }
        if ( exclude )
        {
            for (i=0; i<nsmpl; i++)
                khash_str2int_inc(exclude, smpl[i]);
        }
        free_smpl = 1;
    }
    if ( !smpl )
    {
        smpl  = files->readers[0].header->samples;   // intersection of all samples
        nsmpl = bcf_hdr_nsamples(files->readers[0].header);
    }

    files->samples = NULL;
    files->n_smpl  = 0;
    for (i=0; i<nsmpl; i++)
    {
        if ( exclude && khash_str2int_has_key(exclude,smpl[i])  ) continue;

        int n_isec = 0;
        for (j=0; j<files->nreaders; j++)
        {
            if ( bcf_hdr_id2int(files->readers[j].header, BCF_DT_SAMPLE, smpl[i])<0 ) break;
            n_isec++;
        }
        if ( n_isec!=files->nreaders )
        {
            fprintf(stderr,"Warning: The sample \"%s\" was not found in %s, skipping\n", smpl[i], files->readers[n_isec].fname);
            continue;
        }

        files->samples = (char**) realloc(files->samples, (files->n_smpl+1)*sizeof(const char*));
        files->samples[files->n_smpl++] = strdup(smpl[i]);
    }

    if ( exclude ) khash_str2int_destroy(exclude);
    if ( free_smpl )
    {
        for (i=0; i<nsmpl; i++) free(smpl[i]);
        free(smpl);
    }

    if ( !files->n_smpl )
    {
        if ( files->nreaders>1 )
            fprintf(stderr,"No samples in common.\n");
        return 0;
    }
    for (i=0; i<files->nreaders; i++)
    {
        bcf_sr_t *reader = &files->readers[i];
        reader->samples  = (int*) malloc(sizeof(int)*files->n_smpl);
        reader->n_smpl   = files->n_smpl;
        for (j=0; j<files->n_smpl; j++)
            reader->samples[j] = bcf_hdr_id2int(reader->header, BCF_DT_SAMPLE, files->samples[j]);
    }
    return 1;
}

// Add a new region into a list sorted by start,end. On input the coordinates
// are 1-based, stored 0-based, inclusive.
static void _regions_add(bcf_sr_regions_t *reg, const char *chr, int start, int end)
{
    if ( start==-1 && end==-1 )
    {
        start = 0; end = MAX_CSI_COOR-1;
    }
    else
    {
        start--; end--; // store 0-based coordinates
    }

    if ( !reg->seq_hash )
         reg->seq_hash = khash_str2int_init();

    int iseq;
    if ( khash_str2int_get(reg->seq_hash, chr, &iseq)<0 )
    {
        // the chromosome block does not exist
        iseq = reg->nseqs++;
        reg->seq_names = (char**) realloc(reg->seq_names,sizeof(char*)*reg->nseqs);
        reg->regs = (region_t*) realloc(reg->regs,sizeof(region_t)*reg->nseqs);
        memset(&reg->regs[reg->nseqs-1],0,sizeof(region_t));
        reg->seq_names[iseq] = strdup(chr);
        reg->regs[iseq].creg = -1;
        khash_str2int_set(reg->seq_hash,reg->seq_names[iseq],iseq);
    }

    region_t *creg = &reg->regs[iseq];

    // the regions may not be sorted on input: binary search
    int i, min = 0, max = creg->nregs - 1;
    while ( min<=max )
    {
        i = (max+min)/2;
        if ( start < creg->regs[i].start ) max = i - 1;
        else if ( start > creg->regs[i].start ) min = i + 1;
        else break;
    }
    if ( min>max || creg->regs[i].start!=start || creg->regs[i].end!=end )
    {
        // no such region, insert a new one just after max
        hts_expand(region1_t,creg->nregs+1,creg->mregs,creg->regs);
        if ( ++max < creg->nregs )
            memmove(&creg->regs[max+1],&creg->regs[max],(creg->nregs - max)*sizeof(region1_t));
        creg->regs[max].start = start;
        creg->regs[max].end   = end;
        creg->nregs++;
    }
}

// File name or a list of genomic locations. If file name, NULL is returned.
static bcf_sr_regions_t *_regions_init_string(const char *str)
{
    bcf_sr_regions_t *reg = (bcf_sr_regions_t *) calloc(1, sizeof(bcf_sr_regions_t));
    reg->start = reg->end = -1;
    reg->prev_start = reg->prev_seq = -1;

    kstring_t tmp = {0,0,0};
    const char *sp = str, *ep = str;
    int from, to;
    while ( 1 )
    {
        while ( *ep && *ep!=',' && *ep!=':' ) ep++;
        tmp.l = 0;
        kputsn(sp,ep-sp,&tmp);
        if ( *ep==':' )
        {
            sp = ep+1;
            from = hts_parse_decimal(sp,(char**)&ep,0);
            if ( sp==ep )
            {
                fprintf(stderr,"[%s:%d %s] Could not parse the region(s): %s\n", __FILE__,__LINE__,__FUNCTION__,str);
                free(reg); free(tmp.s); return NULL;
            }
            if ( !*ep || *ep==',' )
            {
                _regions_add(reg, tmp.s, from, from);
                sp = ep;
                continue;
            }
            if ( *ep!='-' )
            {
                fprintf(stderr,"[%s:%d %s] Could not parse the region(s): %s\n", __FILE__,__LINE__,__FUNCTION__,str);
                free(reg); free(tmp.s); return NULL;
            }
            ep++;
            sp = ep;
            to = hts_parse_decimal(sp,(char**)&ep,0);
            if ( *ep && *ep!=',' )
            {
                fprintf(stderr,"[%s:%d %s] Could not parse the region(s): %s\n", __FILE__,__LINE__,__FUNCTION__,str);
                free(reg); free(tmp.s); return NULL;
            }
            if ( sp==ep ) to = MAX_CSI_COOR-1;
            _regions_add(reg, tmp.s, from, to);
            if ( !*ep ) break;
            sp = ep;
        }
        else
        {
            if ( tmp.l ) _regions_add(reg, tmp.s, -1, -1);
            if ( !*ep ) break;
            sp = ++ep;
        }
    }
    free(tmp.s);
    return reg;
}

// ichr,ifrom,ito are 0-based;
// returns -1 on error, 0 if the line is a comment line, 1 on success
static int _regions_parse_line(char *line, int ichr,int ifrom,int ito, char **chr,char **chr_end,int *from,int *to)
{
    *chr_end = NULL;

    if ( line[0]=='#' ) return 0;

    int k,l;    // index of the start and end column of the tab-delimited file
    if ( ifrom <= ito )
        k = ifrom, l = ito;
    else
        l = ifrom, k = ito;

    int i;
    char *se = line, *ss = NULL; // start and end
    char *tmp;
    for (i=0; i<=k && *se; i++)
    {
        ss = i==0 ? se++ : ++se;
        while (*se && *se!='\t') se++;
    }
    if ( i<=k ) return -1;
    if ( k==l )
    {
        *from = *to = hts_parse_decimal(ss, &tmp, 0);
        if ( tmp==ss ) return -1;
    }
    else
    {
        if ( k==ifrom )
            *from = hts_parse_decimal(ss, &tmp, 0);
        else
            *to = hts_parse_decimal(ss, &tmp, 0);
        if ( ss==tmp ) return -1;

        for (i=k; i<l && *se; i++)
        {
            ss = ++se;
            while (*se && *se!='\t') se++;
        }
        if ( i<l ) return -1;
        if ( k==ifrom )
            *to = hts_parse_decimal(ss, &tmp, 0);
        else
            *from = hts_parse_decimal(ss, &tmp, 0);
        if ( ss==tmp ) return -1;
    }

    ss = se = line;
    for (i=0; i<=ichr && *se; i++)
    {
        if ( i>0 ) ss = ++se;
        while (*se && *se!='\t') se++;
    }
    if ( i<=ichr ) return -1;
    *chr_end = se;
    *chr = ss;
    return 1;
}

bcf_sr_regions_t *bcf_sr_regions_init(const char *regions, int is_file, int ichr, int ifrom, int ito)
{
    bcf_sr_regions_t *reg;
    if ( !is_file ) return _regions_init_string(regions);

    reg = (bcf_sr_regions_t *) calloc(1, sizeof(bcf_sr_regions_t));
    reg->start = reg->end = -1;
    reg->prev_start = reg->prev_seq = -1;

    reg->file = hts_open(regions, "rb");
    if ( !reg->file )
    {
        fprintf(stderr,"[%s:%d %s] Could not open file: %s\n", __FILE__,__LINE__,__FUNCTION__,regions);
        free(reg);
        return NULL;
    }

    reg->tbx = tbx_index_load(regions);
    if ( !reg->tbx )
    {
        int len = strlen(regions);
        int is_bed  = strcasecmp(".bed",regions+len-4) ? 0 : 1;
        if ( !is_bed && !strcasecmp(".bed.gz",regions+len-7) ) is_bed = 1;

        if ( reg->file->format.format==vcf ) ito = 1;

        // read the whole file, tabix index is not present
        while ( hts_getline(reg->file, KS_SEP_LINE, &reg->line) > 0 )
        {
            char *chr, *chr_end;
            int from, to, ret;
            ret = _regions_parse_line(reg->line.s, ichr,ifrom,abs(ito), &chr,&chr_end,&from,&to);
            if ( ret < 0 )
            {
                if ( ito<0 )
                    ret = _regions_parse_line(reg->line.s, ichr,ifrom,ifrom, &chr,&chr_end,&from,&to);
                if ( ret<0 )
                {
                    fprintf(stderr,"[%s:%d] Could not parse the file %s, using the columns %d,%d[,%d]\n", __FILE__,__LINE__,regions,ichr+1,ifrom+1,ito+1);
                    hts_close(reg->file); reg->file = NULL; free(reg);
                    return NULL;
                }
            }
            if ( !ret ) continue;
            if ( is_bed ) from++;
            *chr_end = 0;
            _regions_add(reg, chr, from, to);
            *chr_end = '\t';
        }
        hts_close(reg->file); reg->file = NULL;
        if ( !reg->nseqs ) { free(reg); return NULL; }
        return reg;
    }

    reg->seq_names = (char**) tbx_seqnames(reg->tbx, &reg->nseqs);
    if ( !reg->seq_hash )
        reg->seq_hash = khash_str2int_init();
    int i;
    for (i=0; i<reg->nseqs; i++)
    {
        khash_str2int_set(reg->seq_hash,reg->seq_names[i],i);
    }
    reg->fname  = strdup(regions);
    reg->is_bin = 1;
    return reg;
}

void bcf_sr_regions_destroy(bcf_sr_regions_t *reg)
{
    int i;
    free(reg->fname);
    if ( reg->itr ) tbx_itr_destroy(reg->itr);
    if ( reg->tbx ) tbx_destroy(reg->tbx);
    if ( reg->file ) hts_close(reg->file);
    if ( reg->als ) free(reg->als);
    if ( reg->als_str.s ) free(reg->als_str.s);
    free(reg->line.s);
    if ( reg->regs )
    {
         // free only in-memory names, tbx names are const
        for (i=0; i<reg->nseqs; i++)
        {
            free(reg->seq_names[i]);
            free(reg->regs[i].regs);
        }
    }
    free(reg->regs);
    free(reg->seq_names);
    khash_str2int_destroy(reg->seq_hash);
    free(reg);
}

int bcf_sr_regions_seek(bcf_sr_regions_t *reg, const char *seq)
{
    reg->iseq = reg->start = reg->end = -1;
    if ( khash_str2int_get(reg->seq_hash, seq, &reg->iseq) < 0 ) return -1;  // sequence seq not in regions

    // using in-memory regions
    if ( reg->regs )
    {
        reg->regs[reg->iseq].creg = -1;
        return 0;
    }

    // reading regions from tabix
    if ( reg->itr ) tbx_itr_destroy(reg->itr);
    reg->itr = tbx_itr_querys(reg->tbx, seq);
    if ( reg->itr ) return 0;

    return -1;
}

int bcf_sr_regions_next(bcf_sr_regions_t *reg)
{
    if ( reg->iseq<0 ) return -1;
    reg->start = reg->end = -1;
    reg->nals = 0;

    // using in-memory regions
    if ( reg->regs )
    {
        while ( reg->iseq < reg->nseqs )
        {
            reg->regs[reg->iseq].creg++;
            if ( reg->regs[reg->iseq].creg < reg->regs[reg->iseq].nregs ) break;
            reg->iseq++;
        }
        if ( reg->iseq >= reg->nseqs ) { reg->iseq = -1; return -1; } // no more regions left
        region1_t *creg = &reg->regs[reg->iseq].regs[reg->regs[reg->iseq].creg];
        reg->start = creg->start;
        reg->end   = creg->end;
        return 0;
    }

    // reading from tabix
    char *chr, *chr_end;
    int ichr = 0, ifrom = 1, ito = 2, is_bed = 0, from, to;
    if ( reg->tbx )
    {
        ichr   = reg->tbx->conf.sc-1;
        ifrom  = reg->tbx->conf.bc-1;
        ito    = reg->tbx->conf.ec-1;
        if ( ito<0 ) ito = ifrom;
        is_bed = reg->tbx->conf.preset==TBX_UCSC ? 1 : 0;
    }

    int ret = 0;
    while ( !ret )
    {
        if ( reg->itr )
        {
            // tabix index present, reading a chromosome block
            ret = tbx_itr_next(reg->file, reg->tbx, reg->itr, &reg->line);
            if ( ret<0 ) { reg->iseq = -1; return -1; }
        }
        else
        {
            if ( reg->is_bin )
            {
                // Waited for seek which never came. Reopen in text mode and stream
                // through the regions, otherwise hts_getline would fail
                hts_close(reg->file);
                reg->file = hts_open(reg->fname, "r");
                if ( !reg->file )
                {
                    fprintf(stderr,"[%s:%d %s] Could not open file: %s\n", __FILE__,__LINE__,__FUNCTION__,reg->fname);
                    reg->file = NULL;
                    bcf_sr_regions_destroy(reg);
                    return -1;
                }
                reg->is_bin = 0;
            }

            // tabix index absent, reading the whole file
            ret = hts_getline(reg->file, KS_SEP_LINE, &reg->line);
            if ( ret<0 ) { reg->iseq = -1; return -1; }
        }
        ret = _regions_parse_line(reg->line.s, ichr,ifrom,ito, &chr,&chr_end,&from,&to);
        if ( ret<0 )
        {
            fprintf(stderr,"[%s:%d] Could not parse the file %s, using the columns %d,%d,%d\n", __FILE__,__LINE__,reg->fname,ichr+1,ifrom+1,ito+1);
            return -1;
        }
    }
    if ( is_bed ) from++;

    *chr_end = 0;
    if ( khash_str2int_get(reg->seq_hash, chr, &reg->iseq)<0 )
    {
        fprintf(stderr,"Broken tabix index? The sequence \"%s\" not in dictionary [%s]\n", chr,reg->line.s);
        exit(1);
    }
    *chr_end = '\t';

    reg->start = from - 1;
    reg->end   = to - 1;
    return 0;
}

static int _regions_match_alleles(bcf_sr_regions_t *reg, int als_idx, bcf1_t *rec)
{
    int i = 0, max_len = 0;
    if ( !reg->nals )
    {
        char *ss = reg->line.s;
        while ( i<als_idx && *ss )
        {
            if ( *ss=='\t' ) i++;
            ss++;
        }
        char *se = ss;
        reg->nals = 1;
        while ( *se && *se!='\t' )
        {
            if ( *se==',' ) reg->nals++;
            se++;
        }
        ks_resize(&reg->als_str, se-ss+1+reg->nals);
        reg->als_str.l = 0;
        hts_expand(char*,reg->nals,reg->mals,reg->als);
        reg->nals = 0;

        se = ss;
        while ( *(++se) )
        {
            if ( *se=='\t' ) break;
            if ( *se!=',' ) continue;
            reg->als[reg->nals] = &reg->als_str.s[reg->als_str.l];
            kputsn(ss,se-ss,&reg->als_str);
            if ( &reg->als_str.s[reg->als_str.l] - reg->als[reg->nals] > max_len ) max_len = &reg->als_str.s[reg->als_str.l] - reg->als[reg->nals];
            reg->als_str.l++;
            reg->nals++;
            ss = ++se;
        }
        reg->als[reg->nals] = &reg->als_str.s[reg->als_str.l];
        kputsn(ss,se-ss,&reg->als_str);
        if ( &reg->als_str.s[reg->als_str.l] - reg->als[reg->nals] > max_len ) max_len = &reg->als_str.s[reg->als_str.l] - reg->als[reg->nals];
        reg->nals++;
        reg->als_type = max_len > 1 ? VCF_INDEL : VCF_SNP;  // this is a simplified check, see vcf.c:bcf_set_variant_types
    }
    int type = bcf_get_variant_types(rec);
    if ( reg->als_type & VCF_INDEL )
        return type & VCF_INDEL ? 1 : 0;
    return !(type & VCF_INDEL) ? 1 : 0;
}

int bcf_sr_regions_overlap(bcf_sr_regions_t *reg, const char *seq, int start, int end)
{
    int iseq;
    if ( khash_str2int_get(reg->seq_hash, seq, &iseq)<0 ) return -1;    // no such sequence

    if ( reg->prev_seq==-1 || iseq!=reg->prev_seq || reg->prev_start > start ) // new chromosome or after a seek
    {
        // flush regions left on previous chromosome
        if ( reg->missed_reg_handler && reg->prev_seq!=-1 && reg->iseq!=-1 )
            bcf_sr_regions_flush(reg);

        bcf_sr_regions_seek(reg, seq);
        reg->start = reg->end = -1;
    }
    if ( reg->prev_seq==iseq && reg->iseq!=iseq ) return -2;    // no more regions on this chromosome
    reg->prev_seq = reg->iseq;
    reg->prev_start = start;

    while ( iseq==reg->iseq && reg->end < start )
    {
        if ( bcf_sr_regions_next(reg) < 0 ) return -2;  // no more regions left
        if ( reg->iseq != iseq ) return -1; // does not overlap any regions
        if ( reg->missed_reg_handler && reg->end < start ) reg->missed_reg_handler(reg, reg->missed_reg_data);
    }
    if ( reg->start <= end ) return 0;    // region overlap
    return -1;  // no overlap
}

void bcf_sr_regions_flush(bcf_sr_regions_t *reg)
{
    if ( !reg->missed_reg_handler || reg->prev_seq==-1 ) return;
    while ( !bcf_sr_regions_next(reg) ) reg->missed_reg_handler(reg, reg->missed_reg_data);
    return;
}