/*  vcfnorm.c -- Left-align and normalize indels.

    Copyright (C) 2013-2023 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 <stdio.h>
#include <strings.h>
#include <unistd.h>
#include <getopt.h>
#include <assert.h>
#include <ctype.h>
#include <string.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <inttypes.h>
#include <htslib/vcf.h>
#include <htslib/synced_bcf_reader.h>
#include <htslib/faidx.h>
#include <htslib/khash_str2int.h>
#include "bcftools.h"
#include "rbuf.h"
#include "abuf.h"
#include "gff.h"
#include "regidx.h"

#define CHECK_REF_EXIT 1
#define CHECK_REF_WARN 2
#define CHECK_REF_SKIP 4
#define CHECK_REF_FIX  8

#define MROWS_SPLIT 1
#define MROWS_MERGE  2

// for -m+, mapping from allele indexes of a single input record
// to allele indexes of output record
typedef struct
{
    int nals, mals, *map;
}
map_t;

// primitive comparison of two records' alleles via hashes; normalized alleles assumed
typedef struct
{
    int n;  // number of alleles
    char *ref, *alt;
    void *hash;
}
cmpals1_t;

typedef struct
{
    cmpals1_t *cmpals;
    int ncmpals, mcmpals;
}
cmpals_t;

typedef struct
{
    char *tseq, *seq;
    int mseq;
    bcf1_t **lines, **tmp_lines, **alines, **blines, *mrow_out;
    int ntmp_lines, mtmp_lines, nalines, malines, nblines, mblines;
    map_t *maps;     // mrow map for each buffered record
    char **als;
    int mmaps, nals, mals;
    uint8_t *tmp_arr1, *tmp_arr2, *diploid;
    int32_t *int32_arr;
    int ntmp_arr1, ntmp_arr2, nint32_arr;
    kstring_t *tmp_str;
    kstring_t *tmp_als, *tmp_del, tmp_kstr;
    int ntmp_als, ntmp_del;
    rbuf_t rbuf;
    int buf_win;            // maximum distance between two records to consider
    int aln_win;            // the realignment window size (maximum repeat size)
    bcf_srs_t *files;       // using the synced reader only for -r option
    bcf_hdr_t *hdr, *out_hdr;
    cmpals_t cmpals_in, cmpals_out;
    faidx_t *fai;
    struct { int tot, set, swap; } nref;
    char **argv, *output_fname, *ref_fname, *vcf_fname, *region, *targets;
    int argc, rmdup, output_type, n_threads, check_ref, strict_filter, do_indels, clevel;
    int nchanged, nskipped, nsplit, njoined, ntotal, mrows_op, mrows_collapse, parsimonious;
    int record_cmd_line, force, force_warned, keep_sum_ad;
    abuf_t *abuf;
    abuf_opt_t atomize;
    int use_star_allele, ma_use_ref_allele;
    char *old_rec_tag;
    htsFile *out;
    char *index_fn;
    int write_index;
    int right_align;
    char *gff_fname;
    gff_t *gff;
    regidx_t *idx_tscript;
    regitr_t *itr_tscript;
}
args_t;

static inline int replace_iupac_codes(char *seq, int nseq)
{
    // Replace ambiguity codes with N for now, it awaits to be seen what the VCF spec codifies in the end
    int i, n = 0;
    for (i=0; i<nseq; i++)
    {
        char c = toupper(seq[i]);
        if ( c!='A' && c!='C' && c!='G' && c!='T' && c!='N' ) { seq[i] = 'N'; n++; }
    }
    return n;
}
static inline int has_non_acgtn(char *seq, int nseq)
{
    char *end = seq + nseq;
    while ( *seq && seq<end )
    {
        char c = toupper(*seq);
        if ( c!='A' && c!='C' && c!='G' && c!='T' && c!='N' ) return 1;
        seq++;
    }
    return 0;
}

static void seq_to_upper(char *seq, int len)
{
    int i;
    if ( len )
        for (i=0; i<len; i++) seq[i] = nt_to_upper(seq[i]);
    else
        for (i=0; seq[i]; i++) seq[i] = nt_to_upper(seq[i]);
}

static void fix_ref(args_t *args, bcf1_t *line)
{
    bcf_unpack(line, BCF_UN_STR);
    int reflen = strlen(line->d.allele[0]);
    int i,j, maxlen = reflen, len;
    for (i=1; i<line->n_allele; i++)
    {
        int len = strlen(line->d.allele[i]);
        if ( maxlen < len ) maxlen = len;
    }

    char *ref = faidx_fetch_seq(args->fai, (char*)bcf_seqname(args->hdr,line), line->pos, line->pos+maxlen-1, &len);
    if ( !ref ) error("faidx_fetch_seq failed at %s:%"PRId64"\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1);
    replace_iupac_codes(ref,len);

    args->nref.tot++;

    // is the REF different? If not, we are done
    if ( !strncasecmp(line->d.allele[0],ref,reflen) ) { free(ref); return; }

    // is the REF allele missing?
    if ( reflen==1 && line->d.allele[0][0]=='.' )
    {
        line->d.allele[0][0] = ref[0];
        args->nref.set++;
        free(ref);
        bcf_update_alleles(args->out_hdr,line,(const char**)line->d.allele,line->n_allele);
        return;
    }

    // does REF or ALT contain non-standard bases?
    int has_non_acgtn = 0;
    for (i=0; i<line->n_allele; i++)
    {
        if ( line->d.allele[i][0]=='<' ) continue;
        has_non_acgtn += replace_iupac_codes(line->d.allele[i],strlen(line->d.allele[i]));
    }
    if ( has_non_acgtn )
    {
        args->nref.set++;
        bcf_update_alleles(args->out_hdr,line,(const char**)line->d.allele,line->n_allele);
        if ( !strncasecmp(line->d.allele[0],ref,reflen) ) { free(ref); return; }
    }

    // does the REF allele contain N's ?
    int fix = 0;
    for (i=0; i<reflen; i++)
    {
        if ( line->d.allele[0][i]!='N' ) continue;
        if ( ref[i]=='N' ) continue;
        line->d.allele[0][i] = ref[i];
        fix++;
        for (j=1; j<line->n_allele; j++)
        {
            int len = strlen(line->d.allele[j]);
            if ( len <= i || line->d.allele[j][i]!='N' ) continue;
            line->d.allele[j][i] = ref[i];
            fix++;
        }
    }
    if ( fix )
    {
        args->nref.set++;
        bcf_update_alleles(args->out_hdr,line,(const char**)line->d.allele,line->n_allele);
        if ( !strncasecmp(line->d.allele[0],ref,reflen) ) { free(ref); return; }
    }


    // is it swapped?
    for (i=1; i<line->n_allele; i++)
    {
        int len = strlen(line->d.allele[i]);
        if ( !strncasecmp(line->d.allele[i],ref,len) ) break;
    }

    kstring_t str = {0,0,0};
    if ( i==line->n_allele )    // none of the alternate alleles matches the reference
    {
        args->nref.set++;
        kputsn(ref,reflen,&str);
        for (i=1; i<line->n_allele; i++)
        {
            kputc(',',&str);
            kputs(line->d.allele[i],&str);
        }
        bcf_update_alleles_str(args->out_hdr,line,str.s);
        free(ref);
        free(str.s);
        return;
    }

    // one of the alternate alleles matches the reference, assume it's a simple swap
    kputs(line->d.allele[i],&str);
    for (j=1; j<line->n_allele; j++)
    {
        kputc(',',&str);
        if ( j==i )
            kputs(line->d.allele[0],&str);
        else
            kputs(line->d.allele[j],&str);
    }
    bcf_update_alleles_str(args->out_hdr,line,str.s);
    args->nref.swap++;
    free(ref);
    free(str.s);

    // swap genotypes
    int ntmp = args->ntmp_arr1 / sizeof(int32_t); // reuse tmp_arr declared as uint8_t
    int ngts = bcf_get_genotypes(args->hdr, line, &args->tmp_arr1, &ntmp);
    args->ntmp_arr1 = ntmp * sizeof(int32_t);
    int32_t *gts = (int32_t*) args->tmp_arr1;
    int ni = 0;
    for (j=0; j<ngts; j++)
    {
        if ( gts[j]==bcf_gt_unphased(0) ) { gts[j] = bcf_gt_unphased(i); ni++; }
        else if ( gts[j]==bcf_gt_phased(0) ) { gts[j] = bcf_gt_phased(i); ni++; }
        else if ( gts[j]==bcf_gt_unphased(i) ) gts[j] = bcf_gt_unphased(0);
        else if ( gts[j]==bcf_gt_phased(i) ) gts[j] = bcf_gt_phased(0);
    }
    bcf_update_genotypes(args->out_hdr,line,gts,ngts);

    // update AC
    int nac = bcf_get_info_int32(args->hdr, line, "AC", &args->tmp_arr1, &ntmp);
    args->ntmp_arr1 = ntmp * sizeof(int32_t);
    if ( i <= nac )
    {
        int32_t *ac = (int32_t*)args->tmp_arr1;
        ac[i-1] = ni;
        bcf_update_info_int32(args->out_hdr, line, "AC", ac, nac);
    }
}

static void fix_dup_alt(args_t *args, bcf1_t *line)
{
    // update alleles, create a mapping between old and new indexes
    hts_expand(uint8_t,line->n_allele,args->ntmp_arr1,args->tmp_arr1);
    args->tmp_arr1[0] = 0;  // ref always unchanged

    int i, j, nals = line->n_allele, nals_ori = line->n_allele;
    for (i=1, j=1; i<line->n_allele; i++)
    {
        if ( strcasecmp(line->d.allele[0],line->d.allele[i]) )
        {
            args->tmp_arr1[i] = j++;
            continue;
        }
        args->tmp_arr1[i] = 0;
        nals--;
    }
    for (i=1, j=1; i<line->n_allele; i++)
    {
        if ( !args->tmp_arr1[i] ) continue;
        line->d.allele[j++] = line->d.allele[i];
    }
    bcf_update_alleles(args->out_hdr, line, (const char**)line->d.allele, nals);


    // update genotypes
    int ntmp = args->ntmp_arr2 / sizeof(int32_t); // reuse tmp_arr declared as uint8_t
    int ngts = bcf_get_genotypes(args->hdr, line, &args->tmp_arr2, &ntmp);
    args->ntmp_arr2 = ntmp * sizeof(int32_t);
    int32_t *gts = (int32_t*) args->tmp_arr2;
    int changed = 0;
    for (i=0; i<ngts; i++)
    {
        if ( bcf_gt_is_missing(gts[i]) || gts[i]==bcf_int32_vector_end ) continue;
        int ial = bcf_gt_allele(gts[i]);
        if ( ial<nals_ori && ial==args->tmp_arr1[ial] ) continue;
        int ial_new = ial<nals_ori ? args->tmp_arr1[ial] : 0;
        gts[i] = bcf_gt_is_phased(gts[i]) ? bcf_gt_phased(ial_new) : bcf_gt_unphased(ial_new);
        changed = 1;
    }
    if ( changed ) bcf_update_genotypes(args->out_hdr,line,gts,ngts);
}

static void set_old_rec_tag(args_t *args, bcf1_t *dst, bcf1_t *src, int ialt)
{
    if ( !args->old_rec_tag ) return;

    // only update if the tag is not present already, there can be multiple normalization steps
    int i, id = bcf_hdr_id2int(args->out_hdr, BCF_DT_ID, args->old_rec_tag);
    bcf_unpack(dst, BCF_UN_INFO);
    for (i=0; i<dst->n_info; i++)
    {
        bcf_info_t *inf = &dst->d.info[i];
        if ( inf && inf->key == id ) return;
    }

    args->tmp_kstr.l = 0;
    ksprintf(&args->tmp_kstr,"%s|%"PRIhts_pos"|%s|",bcf_seqname(args->hdr,src),src->pos+1,src->d.allele[0]);
    for (i=1; i<src->n_allele; i++)
    {
        kputs(src->d.allele[i],&args->tmp_kstr);
        if ( i+1<src->n_allele ) kputc(',',&args->tmp_kstr);
    }
    if ( ialt>0 )
    {
        kputc('|',&args->tmp_kstr);
        kputw(ialt,&args->tmp_kstr);
    }
    if ( (bcf_update_info_string(args->out_hdr, dst, args->old_rec_tag, args->tmp_kstr.s))!=0 )
            error("An error occurred while updating INFO/%s\n",args->old_rec_tag);
}

static int is_left_align(args_t *args, bcf1_t *line)
{
    if ( args->right_align ) return 0;
    if ( !args->gff ) return 1;
    const char *chr = bcf_seqname(args->hdr,line);
    if ( !strncasecmp("chr",chr,3) ) chr += 3;  // strip 'chr' prefix, that's what we requested the GFF reader to do
    if ( !regidx_overlap(args->idx_tscript,chr,line->pos,line->pos+line->rlen, args->itr_tscript) ) return 1;

    // if there are two conflicting overlapping transcripts, go with the default left-alignment
    int has_fwd = 0;
    while ( regitr_overlap(args->itr_tscript) )
    {
        gf_tscript_t *tr = regitr_payload(args->itr_tscript, gf_tscript_t*);
        if ( tr->strand==STRAND_FWD ) has_fwd = 1;
        if ( tr->strand==STRAND_REV ) return 1;
    }
    // either no hit at all (then left-align) or everything was on fwd strand (then right-align)
    return has_fwd ? 0 : 1;
}
static hts_pos_t realign_left(args_t *args, bcf1_t *line)
{
    // trim from right
    char *ref = NULL;
    int i;
    hts_pos_t nref=0, new_pos = line->pos;
    kstring_t *als = args->tmp_als;
    while (1)
    {
        // is the rightmost base identical in all alleles?
        int min_len = als[0].l;
        for (i=1; i<line->n_allele; i++)
        {
            if ( toupper(als[0].s[ als[0].l-1 ]) != toupper(als[i].s[ als[i].l-1 ]) ) break;
            if ( als[i].l < min_len ) min_len = als[i].l;
        }
        if ( i!=line->n_allele ) break; // there are differences, cannot be trimmed
        if ( min_len<=1 && new_pos==0 ) break;

        int pad_from_left = 0;
        for (i=0; i<line->n_allele; i++) // trim all alleles
        {
            als[i].l--;
            if ( !als[i].l ) pad_from_left = 1;
        }
        if ( pad_from_left )
        {
            // extend all alleles to the left by aln_win bases (unless close to the chr start).
            // Extra bases will be trimmed from the left after this loop is done
            int npad = new_pos >= args->aln_win ? args->aln_win : new_pos;
            free(ref);
            ref = faidx_fetch_seq64(args->fai, bcf_seqname(args->hdr,line), new_pos-npad, new_pos-1, &nref);
            if ( !ref ) error("faidx_fetch_seq64 failed at %s:%"PRId64"\n", bcf_seqname(args->hdr,line), (int64_t) new_pos-npad+1);
            replace_iupac_codes(ref,nref);
            for (i=0; i<line->n_allele; i++)
            {
                ks_resize(&als[i], als[i].l + npad);
                if ( als[i].l ) memmove(als[i].s+npad,als[i].s,als[i].l);
                memcpy(als[i].s,ref,npad);
                als[i].l += npad;
            }
            new_pos -= npad;
        }
    }
    free(ref);

    // trim from left
    int ntrim_left = 0;
    while (1)
    {
        // is the first base identical in all alleles?
        int min_len = als[0].l - ntrim_left;
        for (i=1; i<line->n_allele; i++)
        {
            if ( toupper(als[0].s[ntrim_left]) != toupper(als[i].s[ntrim_left]) ) break;
            if ( min_len > als[i].l - ntrim_left ) min_len = als[i].l - ntrim_left;
        }
        if ( i!=line->n_allele || min_len<=1 ) break; // there are differences, cannot be trimmed
        ntrim_left++;
    }
    if ( ntrim_left )
    {
        for (i=0; i<line->n_allele; i++)
        {
            memmove(als[i].s,als[i].s+ntrim_left,als[i].l-ntrim_left);
            als[i].l -= ntrim_left;
        }
        new_pos += ntrim_left;
    }
    return new_pos;
}

static hts_pos_t realign_right(args_t *args, bcf1_t *line)
{
    char *ref = NULL;
    int i;
    hts_pos_t new_pos = line->pos, nref = 0;
    kstring_t *als = args->tmp_als;

    // trim from left
    int ntrim_left = 0, npad_right = line->rlen, has_indel = 0;
    while (1)
    {
        // is the leftmost base identical in all alleles?
        int min_len = als[0].l - ntrim_left;
        for (i=1; i<line->n_allele; i++)
        {
            if ( als[0].l!=als[i].l ) has_indel = 1;
            if ( toupper(als[0].s[ntrim_left]) != toupper(als[i].s[ntrim_left]) ) break;
            if ( min_len > als[i].l - ntrim_left ) min_len = als[i].l - ntrim_left;
        }
        if ( i!=line->n_allele ) break; // there are differences, cannot be trimmed further

        ntrim_left++;
        if ( min_len<=1 ) // pad from the right
        {
            free(ref);
            ref = faidx_fetch_seq64(args->fai, bcf_seqname(args->hdr,line), line->pos + npad_right, line->pos + npad_right + args->aln_win, &nref);
            if ( !ref ) error("faidx_fetch_seq64 failed at %s:%"PRIhts_pos"\n",bcf_seqname(args->hdr,line), new_pos + ntrim_left);
            npad_right += args->aln_win;
            replace_iupac_codes(ref,nref);
            for (i=0; i<line->n_allele; i++) kputs(ref, &als[i]);
        }
    }
    ntrim_left -= has_indel;
    if ( ntrim_left > 0 )
    {
        for (i=0; i<line->n_allele; i++)
        {
            memmove(als[i].s, als[i].s + ntrim_left, als[i].l - ntrim_left);
            als[i].l -= ntrim_left;
        }
        new_pos += ntrim_left;
    }
    free(ref);

    // trim from right
    while (1)
    {
        // is the last base identical in all alleles?
        int min_len = als[0].l;
        for (i=1; i<line->n_allele; i++)
        {
            if ( toupper(als[0].s[ als[0].l-1 ]) != toupper(als[i].s[ als[i].l-1 ]) ) break;
            if ( min_len > als[i].l ) min_len = als[i].l;
        }
        if ( i!=line->n_allele || min_len<=1 ) break; // there are differences, cannot be trimmed more
        for (i=0; i<line->n_allele; i++) { als[i].l--; als[i].s[als[i].l]=0; }
    }
    return new_pos;
}

#define ERR_DUP_ALLELE       -2
#define ERR_REF_MISMATCH     -1
#define ERR_OK                0
#define ERR_SYMBOLIC          1
#define ERR_SPANNING_DELETION 2

static int realign(args_t *args, bcf1_t *line)
{
    bcf_unpack(line, BCF_UN_STR);

    // Sanity check REF
    int i, nref, reflen = strlen(line->d.allele[0]);
    char *ref = faidx_fetch_seq(args->fai, (char*)args->hdr->id[BCF_DT_CTG][line->rid].key, line->pos, line->pos+reflen-1, &nref);
    if ( !ref ) error("faidx_fetch_seq failed at %s:%"PRId64"\n", args->hdr->id[BCF_DT_CTG][line->rid].key, (int64_t) line->pos+1);
    seq_to_upper(ref,0);
    replace_iupac_codes(ref,nref);  // any non-ACGT character in fasta ref is replaced with N

    // does VCF REF contain non-standard bases?
    if ( has_non_acgtn(line->d.allele[0],reflen) )
    {
        if ( args->check_ref==CHECK_REF_EXIT )
            error("Non-ACGTN reference allele at %s:%"PRId64" .. REF_SEQ:'%s' vs VCF:'%s'\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1,ref,line->d.allele[0]);
        if ( args->check_ref & CHECK_REF_WARN )
            fprintf(stderr,"NON_ACGTN_REF\t%s\t%"PRId64"\t%s\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1,line->d.allele[0]);
        free(ref);
        return ERR_REF_MISMATCH;
    }
    if ( strcasecmp(ref,line->d.allele[0]) )
    {
        if ( args->check_ref==CHECK_REF_EXIT )
            error("Reference allele mismatch at %s:%"PRId64" .. REF_SEQ:'%s' vs VCF:'%s'\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1,ref,line->d.allele[0]);
        if ( args->check_ref & CHECK_REF_WARN )
            fprintf(stderr,"REF_MISMATCH\t%s\t%"PRId64"\t%s\t%s\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1,line->d.allele[0],ref);
        free(ref);
        return ERR_REF_MISMATCH;
    }
    free(ref);
    ref = NULL;

    if ( line->n_allele == 1 ) // a REF
    {
        if ( line->rlen > 1 )
        {
            line->d.allele[0][1] = 0;
            bcf_update_alleles(args->out_hdr,line,(const char**)line->d.allele,line->n_allele);
        }
        return ERR_OK;
    }
    if ( bcf_get_variant_types(line)==VCF_BND ) return ERR_SYMBOLIC;   // breakend, not an error

    // make a copy of each allele for trimming
    hts_expand0(kstring_t,line->n_allele,args->ntmp_als,args->tmp_als);
    hts_expand0(kstring_t,line->n_allele,args->ntmp_del,args->tmp_del);
    kstring_t *als = args->tmp_als;
    kstring_t *del = args->tmp_del;
    int symbolic_alts = 1;
    for (i=0; i<line->n_allele; i++)
    {
        del[i].l = 0;
        if ( line->d.allele[i][0]=='<' )
        {
            // symbolic allele, only <DEL.*> will be realigned
            if ( strncmp("<DEL",line->d.allele[i],4) ) return ERR_SYMBOLIC;
            if ( nref < line->rlen )
            {
                free(ref);
                reflen = line->rlen;
                ref = faidx_fetch_seq(args->fai, (char*)args->hdr->id[BCF_DT_CTG][line->rid].key, line->pos, line->pos+reflen-1, &nref);
                if ( !ref ) error("faidx_fetch_seq failed at %s:%"PRId64"\n", args->hdr->id[BCF_DT_CTG][line->rid].key, (int64_t) line->pos+1);
                seq_to_upper(ref,0);
                replace_iupac_codes(ref,nref);  // any non-ACGT character in fasta ref is replaced with N
                als[0].l = 0;
                kputs(ref, &als[0]);
            }
            als[i].l = 0;
            kputsn(als[0].s,1,&als[i]);
            kputs(line->d.allele[i],&del[i]);
            continue;
        }
        if ( i>0 ) symbolic_alts = 0;
        if ( line->d.allele[i][0]=='*' ) return ERR_SPANNING_DELETION;  // spanning deletion
        if ( has_non_acgtn(line->d.allele[i],line->shared.l) )
        {
            if ( args->check_ref==CHECK_REF_EXIT )
                error("Non-ACGTN alternate allele at %s:%"PRId64" .. VCF:'%s'\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1,line->d.allele[i]);
            if ( args->check_ref & CHECK_REF_WARN )
                fprintf(stderr,"NON_ACGTN_ALT\t%s\t%"PRId64"\t%s\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1,line->d.allele[i]);
            return ERR_REF_MISMATCH;
        }

        als[i].l = 0;
        kputs(line->d.allele[i], &als[i]);
        seq_to_upper(als[i].s,0);

        if ( i>0 && als[i].l==als[0].l && !strcasecmp(als[0].s,als[i].s) ) return ERR_DUP_ALLELE;
    }
    free(ref);
    ref = NULL;

    // which direction are we aligning?
    int left_align = is_left_align(args, line);

    hts_pos_t new_pos;
    if ( left_align )
        new_pos = realign_left(args, line);
    else
        new_pos = realign_right(args, line);

    // Have the alleles changed? Consider <DEL> could have expanded the REF allele. In that
    // case it must be trimmed, however the new REF length must reflect the entire length.
    als[0].s[ als[0].l ] = 0;   // for strcmp to work
    int new_reflen = strlen(als[0].s);
    if ( symbolic_alts )
    {
        als[0].l = 1;
        als[0].s[ als[0].l ] = 0;
    }
    if ( new_pos==line->pos && !strcasecmp(line->d.allele[0],als[0].s) ) return ERR_OK;

    set_old_rec_tag(args, line, line, 0);

    // Create new block of alleles and update
    args->tmp_kstr.l = 0;
    for (i=0; i<line->n_allele; i++)
    {
        if (i>0) kputc(',',&args->tmp_kstr);
        if ( del[i].l ) kputs(del[i].s,&args->tmp_kstr);
        else kputsn(als[i].s,als[i].l,&args->tmp_kstr);
    }
    args->tmp_kstr.s[ args->tmp_kstr.l ] = 0;
    bcf_update_alleles_str(args->out_hdr,line,args->tmp_kstr.s);
    args->nchanged++;

    // Update INFO/END if necessary
    if ( (new_pos!=line->pos || reflen!=new_reflen) && bcf_get_info_int32(args->hdr, line, "END", &args->int32_arr, &args->nint32_arr)==1 )
    {
        // bcf_update_alleles_str() messed up rlen because line->pos changed. This will be fixed by bcf_update_info_int32()
        line->pos = new_pos;
        args->int32_arr[0] = line->pos + new_reflen;
        bcf_update_info_int32(args->out_hdr, line, "END", args->int32_arr, 1);
    }
    line->pos = new_pos;

    return ERR_OK;
}

static void split_info_numeric(args_t *args, bcf1_t *src, bcf_info_t *info, int ialt, bcf1_t *dst)
{
    #define BRANCH_NUMERIC(type,type_t,is_vector_end,is_missing) \
    { \
        const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,info->key); \
        int ntmp = args->ntmp_arr1 / sizeof(type_t); \
        int ret = bcf_get_info_##type(args->hdr,src,tag,&args->tmp_arr1,&ntmp); \
        args->ntmp_arr1 = ntmp * sizeof(type_t); \
        assert( ret>0 ); \
        type_t *vals = (type_t*) args->tmp_arr1; \
        int len = bcf_hdr_id2length(args->hdr,BCF_HL_INFO,info->key); \
        if ( len==BCF_VL_A ) \
        { \
            if ( ret!=src->n_allele-1 ) \
            { \
                if ( args->force && !args->force_warned ) \
                { \
                    fprintf(stderr, \
                        "Warning: wrong number of fields in INFO/%s at %s:%"PRId64", expected %d, found %d\n" \
                        "         (This warning is printed only once.)\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele-1,ret); \
                    args->force_warned = 1; \
                } \
                if ( args->force ) \
                { \
                    bcf_update_info_##type(args->out_hdr,dst,tag,NULL,0); \
                    return; \
                } \
                error("Error: wrong number of fields in INFO/%s at %s:%"PRId64", expected %d, found %d\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele-1,ret); \
            } \
            bcf_update_info_##type(args->out_hdr,dst,tag,vals+ialt,1); \
        } \
        else if ( len==BCF_VL_R ) \
        { \
            if ( ret!=src->n_allele ) \
            { \
                if ( args->force && !args->force_warned ) \
                { \
                    fprintf(stderr, \
                        "Warning: wrong number of fields in INFO/%s at %s:%"PRId64", expected %d, found %d\n" \
                        "         (This warning is printed only once.)\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele,ret); \
                    args->force_warned = 1; \
                } \
                if ( args->force ) \
                { \
                    bcf_update_info_##type(args->out_hdr,dst,tag,NULL,0); \
                    return; \
                } \
                error("Error: wrong number of fields in INFO/%s at %s:%"PRId64", expected %d, found %d\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele,ret); \
            } \
            if ( args->keep_sum_ad >= 0 && args->keep_sum_ad==info->key ) \
            { \
                int j; \
                for (j=1; j<info->len; j++) \
                    if ( j!=ialt+1 && !(is_missing) && !(is_vector_end) ) vals[0] += vals[j]; \
                vals[1] = vals[ialt+1]; \
            } \
            else \
            { \
                if ( ialt!=0 ) vals[1] = vals[ialt+1]; \
            } \
            bcf_update_info_##type(args->out_hdr,dst,tag,vals,2); \
        } \
        else if ( len==BCF_VL_G ) \
        { \
            if ( ret!=src->n_allele*(src->n_allele+1)/2 ) \
            { \
                if ( args->force && !args->force_warned ) \
                { \
                    fprintf(stderr, \
                        "Warning: wrong number of fields in INFO/%s at %s:%"PRId64", expected %d, found %d\n" \
                        "         (This warning is printed only once.)\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele*(src->n_allele+1)/2,ret); \
                    args->force_warned = 1; \
                } \
                if ( args->force ) \
                { \
                    bcf_update_info_##type(args->out_hdr,dst,tag,NULL,0); \
                    return; \
                } \
                error("Error: wrong number of fields in INFO/%s at %s:%"PRId64", expected %d, found %d\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele*(src->n_allele+1)/2,ret); \
            } \
            if ( ialt!=0 ) \
            { \
                vals[1] = vals[bcf_alleles2gt(0,ialt+1)]; \
                vals[2] = vals[bcf_alleles2gt(ialt+1,ialt+1)]; \
            } \
            bcf_update_info_##type(args->out_hdr,dst,tag,vals,3); \
        } \
        else \
            bcf_update_info_##type(args->out_hdr,dst,tag,vals,ret); \
    }
    switch (bcf_hdr_id2type(args->hdr,BCF_HL_INFO,info->key))
    {
        case BCF_HT_INT:  BRANCH_NUMERIC(int32, int32_t, vals[j]==bcf_int32_vector_end, vals[j]==bcf_int32_missing); break;
        case BCF_HT_REAL: BRANCH_NUMERIC(float, float, bcf_float_is_vector_end(vals[j]), bcf_float_is_missing(vals[j])); break;
    }
    #undef BRANCH_NUMERIC
}
// Find n-th field in a comma-separated list and move it to dst.
// The memory areas may overlap.
#define STR_MOVE_NTH(dst,src,end,nth,len) \
{ \
    char *ss = src, *se = src; \
    int j = 0; \
    while ( *se && se<(end) ) \
    { \
        if ( *se==',' ) \
        { \
            if ( j==nth ) break; \
            j++; \
            ss = se+1; \
        } \
        se++; \
    } \
    if ( j==nth ) \
    { \
        int n = se - ss; \
        memmove((dst),ss,n); \
        src = se; \
        len += n; \
    } \
    else len = -1; \
}
static void split_info_string(args_t *args, bcf1_t *src, bcf_info_t *info, int ialt, bcf1_t *dst)
{
    const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,info->key);
    int ret = bcf_get_info_string(args->hdr,src,tag,&args->tmp_arr1,&args->ntmp_arr1);
    assert( ret>0 );

    kstring_t str;
    str.m = args->ntmp_arr1;
    str.l = ret;
    str.s = (char*) args->tmp_arr1;

    int len = bcf_hdr_id2length(args->hdr,BCF_HL_INFO,info->key);
    if ( len==BCF_VL_A )
    {
        char *tmp = str.s;
        int len = 0;
        STR_MOVE_NTH(str.s,tmp,str.s+str.l,ialt,len);
        if ( len<0 ) return;   // wrong number of fields: skip
        str.s[len] = 0;
        bcf_update_info_string(args->out_hdr,dst,tag,str.s);
    }
    else if ( len==BCF_VL_R )
    {
        char *tmp = str.s;
        int len = 0;
        STR_MOVE_NTH(str.s,tmp,str.s+str.l,0,len);
        str.s[len]=','; tmp++; len++;
        STR_MOVE_NTH(&str.s[len],tmp,str.s+str.l,ialt,len);
        if ( len<0 ) return;   // wrong number of fields: skip
        str.s[len] = 0;
        bcf_update_info_string(args->out_hdr,dst,tag,str.s);
    }
    else if ( len==BCF_VL_G )
    {
        int i0a = bcf_alleles2gt(0,ialt+1), iaa = bcf_alleles2gt(ialt+1,ialt+1);
        char *tmp = str.s;
        int len = 0;
        STR_MOVE_NTH(str.s,tmp,str.s+str.l,0,len);
        str.s[len]=','; tmp++; len++;
        STR_MOVE_NTH(&str.s[len],tmp,str.s+str.l,i0a-1,len);
        if ( len<0 ) return;   // wrong number of fields: skip
        str.s[len]=','; tmp++; len++;
        STR_MOVE_NTH(&str.s[len],tmp,str.s+str.l,iaa-i0a-1,len);
        if ( len<0 ) return;   // wrong number of fields: skip
        str.s[len] = 0;
        bcf_update_info_string(args->out_hdr,dst,tag,str.s);
    }
    else
        bcf_update_info_string(args->out_hdr,dst,tag,str.s);
}
static void split_info_flag(args_t *args, bcf1_t *src, bcf_info_t *info, int ialt, bcf1_t *dst)
{
    const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,info->key);
    int ret = bcf_get_info_flag(args->hdr,src,tag,&args->tmp_arr1,&args->ntmp_arr1);
    bcf_update_info_flag(args->out_hdr,dst,tag,NULL,ret);
}

static void split_format_genotype(args_t *args, bcf1_t *src, bcf_fmt_t *fmt, int ialt, bcf1_t *dst)
{
    int ntmp = args->ntmp_arr1 / 4;
    int ngts = bcf_get_genotypes(args->hdr,src,&args->tmp_arr1,&ntmp);
    args->ntmp_arr1 = ntmp * 4;
    assert( ngts >0 );

    int32_t *gt = (int32_t*) args->tmp_arr1;
    int i, j, nsmpl = bcf_hdr_nsamples(args->hdr);
    ngts /= nsmpl;
    for (i=0; i<nsmpl; i++)
    {
        for (j=0; j<ngts; j++)
        {
            if ( gt[j]==bcf_int32_vector_end ) break;
            if ( bcf_gt_is_missing(gt[j]) ) continue; // missing allele: leave as is
            if ( (ialt==0 || args->ma_use_ref_allele) && bcf_gt_allele(gt[j])==0 ) continue; // ref && `--multi-overlaps 0`: leave as is
            if ( bcf_gt_allele(gt[j])==ialt+1 )
                gt[j] = bcf_gt_unphased(1) | bcf_gt_is_phased(gt[j]); // set to first ALT
            else if ( args->ma_use_ref_allele )
                gt[j] = bcf_gt_unphased(0) | bcf_gt_is_phased(gt[j]); // set to REF
            else
                gt[j] = bcf_gt_missing | bcf_gt_is_phased(gt[j]);     // set to missing
        }
        gt += ngts;
    }
    bcf_update_genotypes(args->out_hdr,dst,args->tmp_arr1,ngts*nsmpl);
}
static void split_format_numeric(args_t *args, bcf1_t *src, bcf_fmt_t *fmt, int ialt, bcf1_t *dst)
{
    #define BRANCH_NUMERIC(type,type_t,is_vector_end,is_missing,set_vector_end,set_missing) \
    { \
        const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,fmt->id); \
        int ntmp = args->ntmp_arr1 / sizeof(type_t); \
        int nvals = bcf_get_format_##type(args->hdr,src,tag,&args->tmp_arr1,&ntmp); \
        args->ntmp_arr1 = ntmp * sizeof(type_t); \
        assert( nvals>0 ); \
        type_t *vals = (type_t *) args->tmp_arr1; \
        int len = bcf_hdr_id2length(args->hdr,BCF_HL_FMT,fmt->id); \
        int i,j, nsmpl = bcf_hdr_nsamples(args->hdr); \
        if ( nvals==nsmpl ) /* all values are missing */ \
        { \
            bcf_update_format_##type(args->out_hdr,dst,tag,vals,nsmpl); \
            return; \
        } \
        if ( len==BCF_VL_A ) \
        { \
            if ( nvals!=(src->n_allele-1)*nsmpl ) \
            { \
                if ( args->force && !args->force_warned ) \
                { \
                    fprintf(stderr, \
                        "Warning: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d, found %d. Removing the field.\n" \
                        "         (This warning is printed only once.)\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,(src->n_allele-1)*nsmpl,nvals); \
                    args->force_warned = 1; \
                } \
                if ( args->force ) \
                { \
                    bcf_update_format_##type(args->out_hdr,dst,tag,NULL,0); \
                    return; \
                } \
                error("Error: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d, found %d\n", \
                    tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,(src->n_allele-1)*nsmpl,nvals); \
            } \
            nvals /= nsmpl; \
            type_t *src_vals = vals, *dst_vals = vals; \
            for (i=0; i<nsmpl; i++) \
            { \
                int idst = 0; \
                int isrc = ialt; \
                if ( is_missing || is_vector_end ) set_missing; \
                else dst_vals[idst] = src_vals[isrc]; \
                dst_vals += 1; \
                src_vals += nvals; \
            } \
            bcf_update_format_##type(args->out_hdr,dst,tag,vals,nsmpl); \
        } \
        else if ( len==BCF_VL_R ) \
        { \
            if ( nvals!=src->n_allele*nsmpl ) \
            { \
                if ( args->force && !args->force_warned ) \
                { \
                    fprintf(stderr, \
                        "Warning: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d, found %d. Removing the field.\n" \
                        "         (This warning is printed only once.)\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,(src->n_allele-1)*nsmpl,nvals); \
                    args->force_warned = 1; \
                } \
                if ( args->force ) \
                { \
                    bcf_update_format_##type(args->out_hdr,dst,tag,NULL,0); \
                    return; \
                } \
                error("Error: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d, found %d\n", \
                    tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele*nsmpl,nvals); \
            } \
            nvals /= nsmpl; \
            type_t *src_vals = vals, *dst_vals = vals; \
            if ( args->keep_sum_ad >= 0 && args->keep_sum_ad==fmt->id ) \
            { \
                for (i=0; i<nsmpl; i++) \
                { \
                    dst_vals[0] = src_vals[0]; \
                    for (j=1; j<nvals; j++) \
                    { \
                        int isrc = j; \
                        if ( j!=ialt+1 && !(is_missing) && !(is_vector_end) ) dst_vals[0] += src_vals[j]; \
                    } \
                    int isrc = ialt + 1; \
                    int idst = 1; \
                    if ( is_vector_end ) set_missing; \
                    else dst_vals[idst] = src_vals[isrc]; \
                    dst_vals += 2; \
                    src_vals += nvals; \
                } \
            } \
            else \
            { \
                for (i=0; i<nsmpl; i++) \
                { \
                    dst_vals[0] = src_vals[0]; \
                    int isrc = ialt + 1; \
                    int idst = 1; \
                    if ( is_vector_end ) set_missing; \
                    else dst_vals[idst] = src_vals[isrc]; \
                    dst_vals += 2; \
                    src_vals += nvals; \
                } \
            } \
            bcf_update_format_##type(args->out_hdr,dst,tag,vals,nsmpl*2); \
        } \
        else if ( len==BCF_VL_G ) \
        { \
            if ( nvals!=src->n_allele*(src->n_allele+1)/2*nsmpl && nvals!=src->n_allele*nsmpl ) \
            { \
                if ( args->force && !args->force_warned ) \
                { \
                    fprintf(stderr, \
                        "Warning: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d, found %d. Removing the field.\n" \
                        "         (This warning is printed only once.)\n", \
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,(src->n_allele-1)*nsmpl,nvals); \
                    args->force_warned = 1; \
                } \
                if ( args->force ) \
                { \
                    bcf_update_format_##type(args->out_hdr,dst,tag,NULL,0); \
                    return; \
                } \
                error("Error at %s:%"PRId64", the tag %s has wrong number of fields\n", bcf_seqname(args->hdr,src),(int64_t) src->pos+1,bcf_hdr_int2id(args->hdr,BCF_DT_ID,fmt->id)); \
            } \
            nvals /= nsmpl; \
            int all_haploid = nvals==src->n_allele ? 1 : 0; \
            type_t *src_vals = vals, *dst_vals = vals; \
            for (i=0; i<nsmpl; i++) \
            { \
                int haploid = all_haploid; \
                if ( !haploid ) \
                { \
                    int j; \
                    for (j=0; j<nvals; j++) \
                    { \
                        int isrc = j; \
                        if ( is_vector_end ) break; \
                    } \
                    if ( j!=nvals ) haploid = 1; \
                } \
                dst_vals[0] = src_vals[0]; \
                if ( haploid ) \
                { \
                    dst_vals[1] = src_vals[ialt+1]; \
                    if ( !all_haploid ) { int idst = 2; set_vector_end; } \
                } \
                else \
                { \
                    dst_vals[1] = src_vals[bcf_alleles2gt(0,ialt+1)]; \
                    dst_vals[2] = src_vals[bcf_alleles2gt(ialt+1,ialt+1)]; \
                } \
                dst_vals += all_haploid ? 2 : 3; \
                src_vals += nvals; \
            } \
            bcf_update_format_##type(args->out_hdr,dst,tag,vals,all_haploid ? nsmpl*2 : nsmpl*3); \
        } \
        else \
            bcf_update_format_##type(args->out_hdr,dst,tag,vals,nvals); \
    }
    switch (bcf_hdr_id2type(args->hdr,BCF_HL_FMT,fmt->id))
    {
        case BCF_HT_INT:  BRANCH_NUMERIC(int32, int32_t, src_vals[isrc]==bcf_int32_vector_end, src_vals[isrc]==bcf_int32_missing, dst_vals[idst]=bcf_int32_vector_end, dst_vals[idst]=bcf_int32_missing); break;
        case BCF_HT_REAL: BRANCH_NUMERIC(float, float, bcf_float_is_vector_end(src_vals[isrc]), bcf_float_is_missing(src_vals[isrc]), bcf_float_set_vector_end(dst_vals[idst]), bcf_float_set_missing(src_vals[idst])); break;
    }
    #undef BRANCH_NUMERIC
}
static void squeeze_format_char(char *str, int src_blen, int dst_blen, int n)
{
    int i, isrc = 0, idst = 0;
    for (i=0; i<n; i++)
    {
        memmove(str+idst,str+isrc,dst_blen);
        idst += dst_blen;
        isrc += src_blen;
    }
}
static void split_format_string(args_t *args, bcf1_t *src, bcf_fmt_t *fmt, int ialt, bcf1_t *dst)
{
    const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,fmt->id);
    int ret = bcf_get_format_char(args->hdr,src,tag,&args->tmp_arr1,&args->ntmp_arr1);
    if ( !ret ) return; // all values can be empty, leave out the tag, no need to panic
    assert( ret>0 );

    kstring_t str;
    str.m = args->ntmp_arr1;
    str.l = ret;
    str.s = (char*) args->tmp_arr1;

    int nsmpl = bcf_hdr_nsamples(args->hdr);
    int len = bcf_hdr_id2length(args->hdr,BCF_HL_FMT,fmt->id);
    if ( len==BCF_VL_A )
    {
        int i, blen = ret/nsmpl, maxlen = 0;
        char *ptr = str.s;
        for (i=0; i<nsmpl; i++)
        {
            char *tmp = ptr;
            int len = 0;
            STR_MOVE_NTH(tmp,tmp,ptr+blen,ialt,len);
            if ( len<0 ) return;   // wrong number of fields: skip
            if ( maxlen < len ) maxlen = len;
            ptr += blen;
        }
        if ( maxlen<blen ) squeeze_format_char(str.s,blen,maxlen,nsmpl);
        bcf_update_format_char(args->out_hdr,dst,tag,str.s,nsmpl*maxlen);
    }
    else if ( len==BCF_VL_R )
    {
        int i, blen = ret/nsmpl, maxlen = 0;
        char *ptr = str.s;
        for (i=0; i<nsmpl; i++)
        {
            char *tmp = ptr;
            int len = 0;
            STR_MOVE_NTH(ptr,tmp,ptr+blen,0,len);
            ptr[len]=','; tmp++; len++;
            STR_MOVE_NTH(&ptr[len],tmp,ptr+blen,ialt,len);
            if ( len<0 ) return;   // wrong number of fields: skip
            if ( maxlen < len ) maxlen = len;
            ptr += blen;
        }
        if ( maxlen<blen ) squeeze_format_char(str.s,blen,maxlen,nsmpl);
        bcf_update_format_char(args->out_hdr,dst,tag,str.s,nsmpl*maxlen);
    }
    else if ( len==BCF_VL_G )
    {
        int i, blen = ret/nsmpl, maxlen = 0, i0a = bcf_alleles2gt(0,ialt+1), iaa = bcf_alleles2gt(ialt+1,ialt+1);
        char *ptr = str.s;
        for (i=0; i<nsmpl; i++)
        {
            char *se = ptr, *sx = ptr+blen;
            int nfields = 1;
            while ( *se && se<sx )
            {
                if ( *se==',' ) nfields++;
                se++;
            }
            if ( nfields==1 && se-ptr==1 && *ptr=='.' ) continue;   // missing value
            if ( nfields!=src->n_allele*(src->n_allele+1)/2 && nfields!=src->n_allele )
            {
                if ( args->force && !args->force_warned )
                {
                    fprintf(stderr,
                            "Warning: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d or %d, found %d. Removing the field.\n"
                            "         (This warning is printed only once.)\n",
                            tag,bcf_seqname(args->hdr,src),(int64_t)src->pos+1,src->n_allele*(src->n_allele+1)/2,src->n_allele,nfields);
                    args->force_warned = 1;
                }
                if ( args->force )
                {
                    bcf_update_format_char(args->out_hdr,dst,tag,NULL,0);
                    return;
                }
                error("Error: wrong number of fields in FMT/%s at %s:%"PRId64", expected %d or %d, found %d\n",
                        tag,bcf_seqname(args->hdr,src),(int64_t) src->pos+1,src->n_allele*(src->n_allele+1)/2,src->n_allele,nfields);
            }

            int len = 0;
            if ( nfields==src->n_allele )   // haploid
            {
                char *tmp = ptr;
                STR_MOVE_NTH(&ptr[len],tmp,ptr+blen,0,len);
                ptr[len]=','; tmp++; len++;
                STR_MOVE_NTH(&ptr[len],tmp,ptr+blen,ialt,len);
                if ( len<0 ) return;   // wrong number of fields: skip
            }
            else    // diploid
            {
                char *tmp = ptr;
                STR_MOVE_NTH(&ptr[len],tmp,ptr+blen,0,len);
                ptr[len]=','; tmp++; len++;
                STR_MOVE_NTH(&ptr[len],tmp,ptr+blen,i0a-1,len);
                if ( len<0 ) return;   // wrong number of fields: skip
                ptr[len]=','; tmp++; len++;
                STR_MOVE_NTH(&ptr[len],tmp,ptr+blen,iaa-i0a-1,len);
                if ( len<0 ) return;   // wrong number of fields: skip
            }
            if ( maxlen < len ) maxlen = len;
            ptr += blen;
        }
        if ( maxlen<blen ) squeeze_format_char(str.s,blen,maxlen,nsmpl);
        bcf_update_format_char(args->out_hdr,dst,tag,str.s,nsmpl*maxlen);
    }
    else
        bcf_update_format_char(args->out_hdr,dst,tag,str.s,str.l);
}

static void split_multiallelic_to_biallelics(args_t *args, bcf1_t *line)
{
    int i;

    bcf_unpack(line, BCF_UN_ALL);

    // Init the target biallelic lines
    args->ntmp_lines = line->n_allele-1;
    if ( args->mtmp_lines < args->ntmp_lines )
    {
        args->tmp_lines = (bcf1_t **)realloc(args->tmp_lines,sizeof(bcf1_t*)*args->ntmp_lines);
        for (i=args->mtmp_lines; i<args->ntmp_lines; i++)
            args->tmp_lines[i] = NULL;
        args->mtmp_lines = args->ntmp_lines;
    }
    kstring_t tmp = {0,0,0};
    kputs(line->d.allele[0], &tmp);
    kputc(',', &tmp);
    int rlen  = tmp.l;
    int gt_id = bcf_hdr_id2int(args->hdr,BCF_DT_ID,"GT");
    for (i=0; i<args->ntmp_lines; i++)  // for each ALT allele
    {
        if ( !args->tmp_lines[i] ) args->tmp_lines[i] = bcf_init1();
        bcf1_t *dst = args->tmp_lines[i];
        bcf_clear(dst);

        dst->rid  = line->rid;
        dst->pos  = line->pos;
        dst->qual = line->qual;

        // Not quite sure how to handle IDs, they can be assigned to a specific
        // ALT.  For now we leave the ID unchanged for all.
        bcf_update_id(args->out_hdr, dst, line->d.id ? line->d.id : ".");

        tmp.l = rlen;
        kputs(line->d.allele[i+1],&tmp);
        bcf_update_alleles_str(args->out_hdr,dst,tmp.s);

        if ( line->d.n_flt ) bcf_update_filter(args->hdr, dst, line->d.flt, line->d.n_flt);

        int j;
        for (j=0; j<line->n_info; j++)
        {
            bcf_info_t *info = &line->d.info[j];
            int type = bcf_hdr_id2type(args->hdr,BCF_HL_INFO,info->key);
            if ( type==BCF_HT_INT || type==BCF_HT_REAL ) split_info_numeric(args, line, info, i, dst);
            else if ( type==BCF_HT_FLAG ) split_info_flag(args, line, info, i, dst);
            else split_info_string(args, line, info, i, dst);
        }
        set_old_rec_tag(args, dst, line, i + 1); // 1-based indexes

        dst->n_sample = line->n_sample;
        for (j=0; j<line->n_fmt; j++)
        {
            bcf_fmt_t *fmt = &line->d.fmt[j];
            int type = bcf_hdr_id2type(args->hdr,BCF_HL_FMT,fmt->id);
            if ( fmt->id==gt_id ) split_format_genotype(args, line, fmt, i, dst);
            else if ( type==BCF_HT_INT || type==BCF_HT_REAL ) split_format_numeric(args, line, fmt, i, dst);
            else split_format_string(args, line, fmt, i, dst);
        }
    }
    free(tmp.s);
}

// Enlarge FORMAT array containing nsmpl samples each with nals_ori values
// to accommodate nvals values for each sample, filling the gaps with missing
// values. Works also for INFO arrays, with nsmpl set to 1.
#define ENLARGE_ARRAY(type_t,set_missing,arr,narr_bytes,nsmpl,nvals_ori,nvals) \
{ \
    int nbytes_new = (nsmpl)*(nvals)*sizeof(type_t); \
    hts_expand(uint8_t,nbytes_new,narr_bytes,arr); \
    int ismpl, k; \
    for (ismpl=nsmpl-1; ismpl>=0; ismpl--) \
    { \
        type_t *dst_ptr = ((type_t*)arr) + ismpl*(nvals); \
        type_t *src_ptr = ((type_t*)arr) + ismpl*nvals_ori; \
        memmove(dst_ptr,src_ptr,sizeof(type_t)*nvals_ori); \
        for (k=nvals_ori; k<nvals; k++) set_missing; \
    } \
}
static void merge_info_numeric(args_t *args, bcf1_t **lines, int nlines, bcf_info_t *info, bcf1_t *dst)
{
    #define BRANCH_NUMERIC(type,type_t,set_missing,is_vector_end) \
    { \
        const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,info->key); \
        int ntmp = args->ntmp_arr1 / sizeof(type_t); \
        int nvals_ori = bcf_get_info_##type(args->hdr,lines[0],tag,&args->tmp_arr1,&ntmp); \
        args->ntmp_arr1 = ntmp * sizeof(type_t); \
        assert( nvals_ori>0 ); \
        type_t *vals = (type_t*) args->tmp_arr1, *vals2; \
        int i,k,len = bcf_hdr_id2length(args->hdr,BCF_HL_INFO,info->key);  \
        if ( len==BCF_VL_A ) \
        { \
            if (nvals_ori!=lines[0]->n_allele - 1) \
                error("vcfnorm: number of fields in first record at position %s:%"PRId64" for INFO tag %s not as expected [found: %d vs expected:%d]\n", bcf_seqname(args->hdr,lines[0]),(int64_t) lines[0]->pos+1, tag, nvals_ori, lines[0]->n_allele-1); \
            int nvals = dst->n_allele - 1; \
            ENLARGE_ARRAY(type_t,set_missing,args->tmp_arr1,args->ntmp_arr1,1,nvals_ori,nvals); \
            vals = (type_t*) args->tmp_arr1; \
            for (i=1; i<nlines; i++) \
            { \
                int ntmp2 = args->ntmp_arr2 / sizeof(type_t); \
                int nvals2 = bcf_get_info_##type(args->hdr,lines[i],tag,&args->tmp_arr2,&ntmp2); \
                if (nvals2<0) continue; /* info tag does not exist in this record, skip */ \
                args->ntmp_arr2 = ntmp2 * sizeof(type_t); \
                if (nvals2!=lines[i]->n_allele-1) \
                    error("vcfnorm: could not merge INFO tag %s at position %s:%"PRId64"\n", tag, bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1); \
                vals2 = (type_t*) args->tmp_arr2; \
                for (k=0; k<nvals2; k++) \
                { \
                    if ( is_vector_end ) break; \
                    vals[ args->maps[i].map[k+1] - 1 ] = vals2[k]; \
                } \
            } \
            bcf_update_info_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals); \
        } \
        else if ( len==BCF_VL_R ) \
        { \
            if (nvals_ori!=lines[0]->n_allele) \
                error("vcfnorm: number of fields in first record at position %s:%"PRId64" for INFO tag %s not as expected [found: %d vs expected:%d]\n", bcf_seqname(args->hdr,lines[0]),(int64_t) lines[0]->pos+1, tag, nvals_ori, lines[0]->n_allele); \
            int nvals = dst->n_allele; \
            ENLARGE_ARRAY(type_t,set_missing,args->tmp_arr1,args->ntmp_arr1,1,nvals_ori,nvals); \
            vals = (type_t*) args->tmp_arr1; \
            for (i=1; i<nlines; i++) \
            { \
                int ntmp2 = args->ntmp_arr2 / sizeof(type_t); \
                int nvals2 = bcf_get_info_##type(args->hdr,lines[i],tag,&args->tmp_arr2,&ntmp2); \
                if (nvals2<0) continue; /* info tag does not exist in this record, skip */ \
                args->ntmp_arr2 = ntmp2 * sizeof(type_t); \
                if (nvals2!=lines[i]->n_allele) \
                    error("vcfnorm: could not merge INFO tag %s at position %s:%"PRId64"\n", tag, bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1); \
                vals2 = (type_t*) args->tmp_arr2; \
                for (k=0; k<nvals2; k++) \
                { \
                    if ( is_vector_end ) break; \
                    vals[ args->maps[i].map[k] ] = vals2[k]; \
                } \
            } \
            bcf_update_info_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals); \
        } \
        else if ( len==BCF_VL_G ) \
        { \
            /* expecting diploid gt in INFO */ \
            if (nvals_ori!=lines[0]->n_allele*(lines[0]->n_allele+1)/2) { \
                fprintf(stderr, "todo: merge Number=G INFO fields for haploid sites\n"); \
                error("vcfnorm: number of fields in first record at position %s:%"PRId64" for INFO tag %s not as expected [found: %d vs expected:%d]\n", bcf_seqname(args->hdr,lines[0]),(int64_t) lines[0]->pos+1, tag, nvals_ori, lines[0]->n_allele*(lines[0]->n_allele+1)/2); \
            } \
            int nvals = dst->n_allele*(dst->n_allele+1)/2; \
            ENLARGE_ARRAY(type_t,set_missing,args->tmp_arr1,args->ntmp_arr1,1,nvals_ori,nvals); \
            vals = (type_t*) args->tmp_arr1; \
            for (i=1; i<nlines; i++) \
            { \
                int ntmp2 = args->ntmp_arr2 / sizeof(type_t); \
                int nvals2 = bcf_get_info_##type(args->hdr,lines[i],tag,&args->tmp_arr2,&ntmp2); \
                if (nvals2<0) continue; /* info tag does not exist in this record, skip */ \
                args->ntmp_arr2 = ntmp2 * sizeof(type_t); \
                if (nvals2!=lines[i]->n_allele*(lines[i]->n_allele+1)/2) \
                    error("vcfnorm: could not merge INFO tag %s at position %s:%"PRId64"\n", tag, bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1); \
                vals2 = (type_t*) args->tmp_arr2; \
                int ia,ib; \
                k = 0; \
                for (ia=0; ia<lines[i]->n_allele; ia++) \
                { \
                    for (ib=0; ib<=ia; ib++) \
                    { \
                        if ( is_vector_end ) break; \
                        int l = bcf_alleles2gt(args->maps[i].map[ia],args->maps[i].map[ib]); \
                        vals[l] = vals2[k];  \
                        k++; \
                    } \
                } \
            } \
            bcf_update_info_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals); \
        } \
        else \
            bcf_update_info_##type(args->out_hdr,dst,tag,vals,nvals_ori); \
    }
    switch (bcf_hdr_id2type(args->hdr,BCF_HL_INFO,info->key))
    {
        case BCF_HT_INT:  BRANCH_NUMERIC(int32, int32_t, dst_ptr[k]=bcf_int32_missing, vals2[k]==bcf_int32_vector_end); break;
        case BCF_HT_REAL: BRANCH_NUMERIC(float, float, bcf_float_set_missing(dst_ptr[k]), bcf_float_is_vector_end(vals2[k])); break;
    }
    #undef BRANCH_NUMERIC
}
static void merge_info_flag(args_t *args, bcf1_t **lines, int nlines, bcf_info_t *info, bcf1_t *dst)
{
    const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,info->key);
    int ret = bcf_get_info_flag(args->hdr,lines[0],tag,&args->tmp_arr1,&args->ntmp_arr1);
    bcf_update_info_flag(args->out_hdr,dst,tag,NULL,ret);
}
int copy_string_field(char *src, int isrc, int src_len, kstring_t *dst, int idst); // see vcfmerge.c
static void merge_info_string(args_t *args, bcf1_t **lines, int nlines, bcf_info_t *info, bcf1_t *dst)
{
    const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,info->key);

    kstring_t str;
    str.m = args->ntmp_arr1;
    str.l = 0;
    str.s = (char*) args->tmp_arr1;

    int i, j, len = bcf_hdr_id2length(args->hdr,BCF_HL_INFO,info->key);
    if ( len==BCF_VL_A || len==BCF_VL_R )
    {
        int jfrom = len==BCF_VL_A ? 1 : 0;
        kputc('.',&str);
        for (i=jfrom+1; i<dst->n_allele; i++) kputs(",.",&str);
        for (i=0; i<nlines; i++)
        {
            bcf_info_t *src = bcf_get_info(args->hdr,lines[i],tag);
            if (!src) continue;
            for (j=jfrom; j<lines[i]->n_allele; j++)
                copy_string_field((char*)src->vptr, j-jfrom, src->len, &str, args->maps[i].map[j]-jfrom);
        }
        str.s[str.l] = 0;
        args->tmp_arr1  = (uint8_t*) str.s;
        args->ntmp_arr1 = str.m;
        bcf_update_info_string(args->out_hdr,dst,tag,str.s);
    }
    else if ( len==BCF_VL_G )
    {
        int ngts = dst->n_allele*(dst->n_allele+1)/2;
        kputc('.',&str);
        for (i=1; i<ngts; i++) kputs(",.",&str);
        for (i=0; i<nlines; i++)
        {
            bcf_info_t *src = bcf_get_info(args->hdr,lines[i],tag);
            if (!src) continue;
            int iori, jori, kori = 0;
            for (iori=0; iori<lines[i]->n_allele; iori++)
            {
                int inew = args->maps[i].map[iori];
                for (jori=0; jori<=iori; jori++)
                {
                    int jnew = args->maps[i].map[jori];
                    int knew = bcf_alleles2gt(inew,jnew);
                    copy_string_field((char*)src->vptr,kori,src->len,&str,knew);
                    kori++;
                }
            }
        }
        str.s[str.l] = 0;
        args->tmp_arr1  = (uint8_t*) str.s;
        args->ntmp_arr1 = str.m;
        bcf_update_info_string(args->out_hdr,dst,tag,str.s);
    }
    else
    {
        bcf_get_info_string(args->hdr,lines[0],tag,&args->tmp_arr1,&args->ntmp_arr1);
        bcf_update_info_string(args->out_hdr,dst,tag,args->tmp_arr1);
    }
}
static void merge_format_genotype(args_t *args, bcf1_t **lines, int nlines, bcf_fmt_t *fmt, bcf1_t *dst)
{
    // reusing int8_t arrays as int32_t arrays
    int ntmp = args->ntmp_arr1 / 4;
    int ngts = bcf_get_genotypes(args->hdr,lines[0],&args->tmp_arr1,&ntmp);
    args->ntmp_arr1 = ntmp * 4;
    assert( ngts >0 );

    int nsmpl = bcf_hdr_nsamples(args->hdr);
    ngts /= nsmpl;

    int i, j, k,k2;
    for (i=1; i<nlines; i++)
    {
        int ntmp2 = args->ntmp_arr2 / 4;
        int ngts2 = bcf_get_genotypes(args->hdr,lines[i],&args->tmp_arr2,&ntmp2);
        args->ntmp_arr2 = ntmp2 * 4;
        ngts2 /= nsmpl;
        if ( ngts!=ngts2 ) error("Error at %s:%"PRId64": cannot combine diploid with haploid genotype\n", bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1);

        int32_t *gt  = (int32_t*) args->tmp_arr1;       // the first, destination line
        int32_t *gt2 = (int32_t*) args->tmp_arr2;       // one of the subsequent lines, i.e. the source line
        for (j=0; j<nsmpl; j++)
        {
            // Take each source allele and apply to the first line. We try to preserve the order and phasing and we
            // never overwrite with ref allele
            for (k2=0; k2<ngts2; k2++)
            {
                if ( gt2[k2]==bcf_int32_vector_end ) break;
                if ( bcf_gt_is_missing(gt2[k2]) ) continue;
                int ial2 = bcf_gt_allele(gt2[k2]);
                if ( ial2==0 ) continue;    // never overwrite with ref
                if ( ial2>=args->maps[i].nals ) error("Error at %s:%"PRId64": incorrect allele index %d\n",bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1,ial2);

                // The destination allele
                int ial = args->maps[i].map[ial2];
                if ( gt[k2]==bcf_int32_vector_end || bcf_gt_is_missing(gt[k2]) || !bcf_gt_allele(gt[k2]) )
                    gt[k2] = bcf_gt_is_phased(gt[k2]) ? bcf_gt_phased(ial) : bcf_gt_unphased(ial);
                else
                {
                    // conflict, the first line has non-zero allele, use the old way, possibly disrupt the phasing
                    for (k=0; k<ngts; k++)
                        if ( gt[k]==bcf_int32_vector_end || bcf_gt_is_missing(gt[k]) || !bcf_gt_allele(gt[k]) ) break;
                    if ( k<ngts )
                        gt[k] = bcf_gt_unphased(ial);
                }
            }
            gt  += ngts;
            gt2 += ngts;
        }
    }
    bcf_update_genotypes(args->out_hdr,dst,args->tmp_arr1,ngts*nsmpl);
}
static int diploid_to_haploid(int size, int nsmpl, int nals, uint8_t *vals)
{
    int i, dsrc = size*nals*(nals+1)/2, ddst = size*nals;
    uint8_t *src_ptr = vals + dsrc, *dst_ptr = vals + ddst;
    for (i=1; i<nsmpl; i++)
    {
        memmove(dst_ptr,src_ptr,ddst);
        dst_ptr += ddst;
        src_ptr += dsrc;
    }
    return nals;
}
static void merge_format_numeric(args_t *args, bcf1_t **lines, int nlines, bcf_fmt_t *fmt, bcf1_t *dst)
{
    #define BRANCH_NUMERIC(type,type_t,set_missing,is_vector_end,set_vector_end) \
    { \
        const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,fmt->id); \
        int ntmp = args->ntmp_arr1 / sizeof(type_t); \
        int nvals_ori = bcf_get_format_##type(args->hdr,lines[0],tag,&args->tmp_arr1,&ntmp); \
        args->ntmp_arr1 = ntmp * sizeof(type_t); \
        assert( nvals_ori>0 ); \
        type_t *vals2, *vals = (type_t *) args->tmp_arr1; \
        int len = bcf_hdr_id2length(args->hdr,BCF_HL_FMT,fmt->id); \
        int i, j, k, nsmpl = bcf_hdr_nsamples(args->hdr); \
        nvals_ori /= nsmpl; \
        if ( len==BCF_VL_A ) \
        { \
            int nvals = dst->n_allele - 1; \
            ENLARGE_ARRAY(type_t,set_missing,args->tmp_arr1,args->ntmp_arr1,nsmpl,nvals_ori,nvals); \
            for (i=1; i<nlines; i++) \
            { \
                int ntmp2 = args->ntmp_arr2 / sizeof(type_t); \
                int nvals2 = bcf_get_format_##type(args->hdr,lines[i],tag,&args->tmp_arr2,&ntmp2); \
                if (nvals2<0) continue; /* format tag does not exist in this record, skip */ \
                args->ntmp_arr2 = ntmp2 * sizeof(type_t); \
                nvals2 /= nsmpl; \
                if (nvals2!=lines[i]->n_allele-1) \
                    error("vcfnorm: could not merge FORMAT tag %s at position %s:%"PRId64"\n", tag, bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1); \
                vals  = (type_t*) args->tmp_arr1; \
                vals2 = (type_t*) args->tmp_arr2; \
                for (j=0; j<nsmpl; j++) \
                { \
                    for (k=0; k<nvals2; k++) \
                    { \
                        if ( is_vector_end ) break; \
                        vals[ args->maps[i].map[k+1] - 1 ] = vals2[k]; \
                    } \
                    vals  += nvals; \
                    vals2 += nvals2; \
                } \
            } \
            bcf_update_format_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals*nsmpl); \
        } \
        else if ( len==BCF_VL_R ) \
        { \
            int nvals = dst->n_allele; \
            ENLARGE_ARRAY(type_t,set_missing,args->tmp_arr1,args->ntmp_arr1,nsmpl,nvals_ori,nvals); \
            for (i=1; i<nlines; i++) \
            { \
                int ntmp2 = args->ntmp_arr2 / sizeof(type_t); \
                int nvals2 = bcf_get_format_##type(args->hdr,lines[i],tag,&args->tmp_arr2,&ntmp2); \
                if (nvals2<0) continue; /* format tag does not exist in this record, skip */ \
                args->ntmp_arr2 = ntmp2 * sizeof(type_t); \
                nvals2 /= nsmpl; \
                if (nvals2!=lines[i]->n_allele) \
                    error("vcfnorm: could not merge FORMAT tag %s at position %s:%"PRId64"\n", tag, bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1); \
                vals  = (type_t*) args->tmp_arr1; \
                vals2 = (type_t*) args->tmp_arr2; \
                for (j=0; j<nsmpl; j++) \
                { \
                    for (k=0; k<nvals2; k++) \
                    { \
                        if ( is_vector_end ) break; \
                        vals[ args->maps[i].map[k] ] = vals2[k]; \
                    } \
                    vals  += nvals; \
                    vals2 += nvals2; \
                } \
            } \
            bcf_update_format_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals*nsmpl); \
        } \
        else if ( len==BCF_VL_G ) \
        { \
            /* which samples are diploid */ \
            memset(args->diploid,0,nsmpl); \
            int all_haploid = 1; \
            if ( nvals_ori > lines[0]->n_allele ) /* line possibly diploid */ \
            { \
                vals2 = (type_t*) args->tmp_arr1; \
                int ndiploid = lines[0]->n_allele*(lines[0]->n_allele+1)/2; \
                for (i=0; i<nsmpl; i++) \
                { \
                    if ( !args->diploid[i] ) \
                    { \
                        for (k=0; k<nvals_ori; k++) if ( is_vector_end ) break; \
                        if ( k==ndiploid ) { args->diploid[i] = 1; all_haploid = 0; }\
                    } \
                    vals2 += nvals_ori; \
                } \
            } \
            int nvals = dst->n_allele*(dst->n_allele+1)/2; \
            ENLARGE_ARRAY(type_t,set_missing,args->tmp_arr1,args->ntmp_arr1,nsmpl,nvals_ori,nvals); \
            for (i=1; i<nlines; i++) \
            { \
                int ntmp2 = args->ntmp_arr2 / sizeof(type_t); \
                int nvals2 = bcf_get_format_##type(args->hdr,lines[i],tag,&args->tmp_arr2,&ntmp2); \
                if (nvals2<0) continue; /* format tag does not exist in this record, skip */ \
                args->ntmp_arr2 = ntmp2 * sizeof(type_t); \
                nvals2 /= nsmpl; \
                int ndiploid = lines[i]->n_allele*(lines[i]->n_allele+1)/2; \
                int line_diploid = nvals2==ndiploid ? 1 : 0; \
                if (!(nvals2==1 || nvals2==lines[i]->n_allele || nvals2==lines[i]->n_allele*(lines[i]->n_allele+1)/2)) \
                    error("vcfnorm: could not merge FORMAT tag %s at position %s:%"PRId64"\n", tag, bcf_seqname(args->hdr,lines[i]),(int64_t) lines[i]->pos+1); \
                vals  = (type_t*) args->tmp_arr1; \
                vals2 = (type_t*) args->tmp_arr2; \
                for (j=0; j<nsmpl; j++) \
                { \
                    int smpl_diploid = line_diploid; \
                    if ( smpl_diploid ) \
                    { \
                        for (k=0; k<nvals2; k++) if ( is_vector_end ) break; \
                        if ( k!=ndiploid ) smpl_diploid = 0; \
                    } \
                    if ( smpl_diploid && !args->diploid[j] ) { args->diploid[j] = 1; all_haploid = 0; } \
                    if ( !smpl_diploid ) \
                    { \
                        for (k=0; k<lines[i]->n_allele; k++) vals[args->maps[i].map[k]] = vals2[k]; \
                    } \
                    else \
                    { \
                        k = 0; \
                        int ia,ib; \
                        for (ia=0; ia<lines[i]->n_allele; ia++) \
                        { \
                            for (ib=0; ib<=ia; ib++) \
                            { \
                                int l = bcf_alleles2gt(args->maps[i].map[ia],args->maps[i].map[ib]); \
                                vals[l] = vals2[k]; \
                                k++; \
                            } \
                        } \
                    } \
                    vals  += nvals; \
                    vals2 += nvals2; \
                } \
            } \
            if ( all_haploid ) \
                nvals = diploid_to_haploid(sizeof(type_t),nsmpl,dst->n_allele,args->tmp_arr1); \
            else \
            {\
                k = dst->n_allele;\
                vals2 = (type_t*) args->tmp_arr1;\
                for (i=0; i<nsmpl; i++)\
                {\
                    if ( !args->diploid[i] ) set_vector_end;\
                    vals2 += nvals;\
                }\
            }\
            bcf_update_format_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals*nsmpl); \
        } \
        else \
            bcf_update_format_##type(args->out_hdr,dst,tag,args->tmp_arr1,nvals_ori*nsmpl); \
    }
    switch (bcf_hdr_id2type(args->hdr,BCF_HL_FMT,fmt->id))
    {
        case BCF_HT_INT:  BRANCH_NUMERIC(int32, int32_t, dst_ptr[k]=bcf_int32_missing, vals2[k]==bcf_int32_vector_end, vals2[k]=bcf_int32_vector_end); break;
        case BCF_HT_REAL: BRANCH_NUMERIC(float, float, bcf_float_set_missing(dst_ptr[k]), bcf_float_is_vector_end(vals2[k]), bcf_float_set_vector_end(vals2[k])); break;
    }
    #undef BRANCH_NUMERIC
}
static void merge_format_string(args_t *args, bcf1_t **lines, int nlines, bcf_fmt_t *fmt, bcf1_t *dst)
{
    const char *tag = bcf_hdr_int2id(args->hdr,BCF_DT_ID,fmt->id);

    int i, j, k, len = bcf_hdr_id2length(args->hdr,BCF_HL_FMT,fmt->id);
    if ( len!=BCF_VL_A && len!=BCF_VL_R && len!=BCF_VL_G )
    {
        int nret = bcf_get_format_char(args->hdr,lines[0],tag,&args->tmp_arr1,&args->ntmp_arr1);
        bcf_update_format_char(args->out_hdr,dst,tag,args->tmp_arr1,nret);
        return;
    }

    int nsmpl = bcf_hdr_nsamples(args->hdr);
    for (i=0; i<nsmpl; i++) args->tmp_str[i].l = 0;

    if ( len==BCF_VL_A || len==BCF_VL_R )
    {
        int jfrom = len==BCF_VL_A ? 1 : 0;
        for (i=0; i<nsmpl; i++)
        {
            kstring_t *tmp = &args->tmp_str[i];
            kputc('.',tmp);
            for (k=jfrom+1; k<dst->n_allele; k++) kputs(",.",tmp);
        }
        for (i=0; i<nlines; i++)
        {
            int nret = bcf_get_format_char(args->hdr,lines[i],tag,&args->tmp_arr1,&args->ntmp_arr1);
            if (nret<0) continue; /* format tag does not exist in this record, skip */
            nret /= nsmpl;
            for (k=0; k<nsmpl; k++)
            {
                kstring_t *tmp = &args->tmp_str[k];
                char *src = (char*)args->tmp_arr1 + k*nret;
                for (j=jfrom; j<lines[i]->n_allele; j++)
                    copy_string_field(src, j-jfrom, nret, tmp, args->maps[i].map[j]-jfrom);
            }
        }
    }
    else if ( len==BCF_VL_G )
    {
        hts_expand(uint8_t,nsmpl,args->ntmp_arr2,args->tmp_arr2);
        uint8_t *haploid = args->tmp_arr2;
        int nret = bcf_get_format_char(args->hdr,lines[0],tag,&args->tmp_arr1,&args->ntmp_arr1);
        nret /= nsmpl;
        for (i=0; i<nsmpl; i++)
        {
            char *ss = (char*)args->tmp_arr1 + i*nret, *se = ss+nret;
            int nfields = 1;
            while ( *ss && ss<se )
            {
                if ( *ss==',' ) nfields++;
                ss++;
            }
            if ( nfields==lines[0]->n_allele )
            {
                haploid[i] = 1;
                nfields = dst->n_allele;
            }
            else if ( nfields==lines[0]->n_allele*(lines[0]->n_allele+1)/2 )
            {
                haploid[i] = 0;
                nfields = dst->n_allele*(dst->n_allele+1)/2;
            }
            else error("The field %s at %s:%"PRId64" neither diploid nor haploid?\n", tag,bcf_seqname(args->hdr,dst),(int64_t) dst->pos+1);

            kstring_t *tmp = &args->tmp_str[i];
            kputc('.',tmp);
            for (j=1; j<nfields; j++) kputs(",.",tmp);
        }
        for (i=0; i<nlines; i++)
        {
            if ( i ) // we already have a copy
            {
                nret = bcf_get_format_char(args->hdr,lines[i],tag,&args->tmp_arr1,&args->ntmp_arr1);
                if (nret<0) continue; /* format tag does not exist in this record, skip */
                nret /= nsmpl;
            }
            for (k=0; k<nsmpl; k++)
            {
                kstring_t *tmp = &args->tmp_str[k];
                char *src = (char*)args->tmp_arr1 + k*nret;
                if ( haploid[k] )
                {
                    for (j=0; j<lines[i]->n_allele; j++)
                        copy_string_field(src,j,nret, tmp, args->maps[i].map[j]);
                }
                else
                {
                    int iori, jori, kori = 0;
                    for (iori=0; iori<lines[i]->n_allele; iori++)
                    {
                        int inew = args->maps[i].map[iori];
                        for (jori=0; jori<=iori; jori++)
                        {
                            int jnew = args->maps[i].map[jori];
                            int knew = bcf_alleles2gt(inew,jnew);
                            copy_string_field(src,kori,nret,tmp,knew);
                            kori++;
                        }
                    }
                }
            }
        }
    }
    kstring_t str;
    str.m = args->ntmp_arr2;
    str.l = 0;
    str.s = (char*) args->tmp_arr2;

    int max_len = 0;
    for (i=0; i<nsmpl; i++)
        if ( max_len < args->tmp_str[i].l ) max_len = args->tmp_str[i].l;
    for (i=0; i<nsmpl; i++)
    {
        kstring_t *tmp = &args->tmp_str[i];
        kputsn(tmp->s,tmp->l,&str);
        for (j=tmp->l; j<max_len; j++) kputc('\0',&str);
    }
    args->ntmp_arr2 = str.m;
    args->tmp_arr2  = (uint8_t*)str.s;
    bcf_update_format_char(args->out_hdr,dst,tag,str.s,str.l);
}

char **merge_alleles(char **a, int na, int *map, char **b, int *nb, int *mb);   // see vcfmerge.c
static void merge_biallelics_to_multiallelic(args_t *args, bcf1_t *dst, bcf1_t **lines, int nlines)
{
    int i;
    for (i=0; i<nlines; i++)
        bcf_unpack(lines[i], BCF_UN_ALL);

    dst->rid  = lines[0]->rid;
    dst->pos  = lines[0]->pos;

    // take max for QUAL
    bcf_float_set_missing(dst->qual);
    for (i=0; i<nlines; i++) {
        if (bcf_float_is_missing(lines[i]->qual)) continue;
        if (bcf_float_is_missing(dst->qual) || dst->qual<lines[i]->qual)
            dst->qual = lines[i]->qual;
    }

    bcf_update_id(args->out_hdr, dst, lines[0]->d.id);

    // Merge and set the alleles, create a mapping from source allele indexes to dst idxs
    hts_expand0(map_t,nlines,args->mmaps,args->maps);   // a mapping for each line
    args->nals = args->maps[0].nals = lines[0]->n_allele;
    hts_expand(int,args->maps[0].nals,args->maps[0].mals,args->maps[0].map);
    hts_expand(char*,args->nals,args->mals,args->als);
    for (i=0; i<args->maps[0].nals; i++)
    {
        args->maps[0].map[i] = i;
        args->als[i] = strdup(lines[0]->d.allele[i]);
    }
    for (i=1; i<nlines; i++)
    {
        if (lines[i]->d.id[0]!='.' || lines[i]->d.id[1]) bcf_add_id(args->out_hdr, dst, lines[i]->d.id);
        args->maps[i].nals = lines[i]->n_allele;
        hts_expand(int,args->maps[i].nals,args->maps[i].mals,args->maps[i].map);
        args->als = merge_alleles(lines[i]->d.allele, lines[i]->n_allele, args->maps[i].map, args->als, &args->nals, &args->mals);
        if ( !args->als ) error("Failed to merge alleles at %s:%"PRId64"\n", bcf_seqname(args->hdr,dst),(int64_t) dst->pos+1);
    }
    bcf_update_alleles(args->out_hdr, dst, (const char**)args->als, args->nals);
    for (i=0; i<args->nals; i++)
    {
        free(args->als[i]);
        args->als[i] = NULL;
    }

    if ( lines[0]->d.n_flt ) bcf_update_filter(args->out_hdr, dst, lines[0]->d.flt, lines[0]->d.n_flt);
    for (i=1; i<nlines; i++) {
        int j;
        for (j=0; j<lines[i]->d.n_flt; j++) {
            // if strict_filter, set FILTER to PASS if any site PASS
            // otherwise accumulate FILTERs
            if (lines[i]->d.flt[j] == bcf_hdr_id2int(args->hdr, BCF_DT_ID, "PASS")) {
                if (args->strict_filter) {
                    bcf_update_filter(args->out_hdr, dst, lines[i]->d.flt, lines[i]->d.n_flt);
                    break;
                }
                else
                    continue;
            }
            bcf_add_filter(args->out_hdr, dst, lines[i]->d.flt[j]);
        }
    }

    // merge info
    for (i=0; i<lines[0]->n_info; i++)
    {
        bcf_info_t *info = &lines[0]->d.info[i];
        int type = bcf_hdr_id2type(args->hdr,BCF_HL_INFO,info->key);
        if ( type==BCF_HT_INT || type==BCF_HT_REAL ) merge_info_numeric(args, lines, nlines, info, dst);
        else if ( type==BCF_HT_FLAG ) merge_info_flag(args, lines, nlines, info, dst);
        else merge_info_string(args, lines, nlines, info, dst);
    }

    // merge format
    int gt_id = bcf_hdr_id2int(args->hdr,BCF_DT_ID,"GT");
    dst->n_sample = lines[0]->n_sample;
    for (i=0; i<lines[0]->n_fmt; i++)
    {
        bcf_fmt_t *fmt = &lines[0]->d.fmt[i];
        int type = bcf_hdr_id2type(args->hdr,BCF_HL_FMT,fmt->id);
        if ( fmt->id==gt_id ) merge_format_genotype(args, lines, nlines, fmt, dst);
        else if ( type==BCF_HT_INT || type==BCF_HT_REAL ) merge_format_numeric(args, lines, nlines, fmt, dst);
        else merge_format_string(args, lines, nlines, fmt, dst);
    }
    args->njoined++;
}

#define SWAP(type_t, a, b) { type_t t = a; a = b; b = t; }
static void mrows_schedule(args_t *args, bcf1_t **line)
{
    int i,m;
    if ( args->mrows_collapse==COLLAPSE_ANY         // merge all record types together
        || bcf_get_variant_types(*line)&VCF_SNP     // SNP, put into alines
        || bcf_get_variant_types(*line)==VCF_REF )  // ref
    {
        args->nalines++;
        m = args->malines;
        hts_expand(bcf1_t*,args->nalines,args->malines,args->alines);
        for (i=m; i<args->malines; i++) args->alines[i] = bcf_init1();
        SWAP(bcf1_t*, args->alines[args->nalines-1], *line);
    }
    else
    {
        args->nblines++;
        m = args->mblines;
        hts_expand(bcf1_t*,args->nblines,args->mblines,args->blines);
        for (i=m; i<args->mblines; i++) args->blines[i] = bcf_init1();
        SWAP(bcf1_t*, args->blines[args->nblines-1], *line);
    }
}
static int mrows_ready_to_flush(args_t *args, bcf1_t *line)
{
    if ( args->nalines && (args->alines[0]->rid!=line->rid || args->alines[0]->pos!=line->pos) ) return 1;
    if ( args->nblines && (args->blines[0]->rid!=line->rid || args->blines[0]->pos!=line->pos) ) return 1;
    return 0;
}
static bcf1_t *mrows_flush(args_t *args)
{
    if ( args->nblines && args->nalines==1 && bcf_get_variant_types(args->alines[0])==VCF_REF )
    {
        // By default, REF lines are merged with SNPs if SNPs and indels are to be kept separately.
        // However, if there are indels only and a single REF line, merge it with indels.
        args->nblines++;
        int i,m = args->mblines;
        hts_expand(bcf1_t*,args->nblines,args->mblines,args->blines);
        for (i=m; i<args->mblines; i++) args->blines[i] = bcf_init1();
        SWAP(bcf1_t*, args->blines[args->nblines-1], args->alines[0]);
        args->nalines--;
    }
    if ( args->nalines )
    {
        if ( args->nalines==1 )
        {
            args->nalines = 0;
            return args->alines[0];
        }
        bcf_clear(args->mrow_out);
        merge_biallelics_to_multiallelic(args, args->mrow_out, args->alines, args->nalines);
        args->nalines = 0;
        return args->mrow_out;
    }
    else if ( args->nblines )
    {
        if ( args->nblines==1 )
        {
            args->nblines = 0;
            return args->blines[0];
        }
        bcf_clear(args->mrow_out);
        merge_biallelics_to_multiallelic(args, args->mrow_out, args->blines, args->nblines);
        args->nblines = 0;
        return args->mrow_out;
    }
    return NULL;
}
static void cmpals_add(cmpals_t *ca, bcf1_t *rec)
{
    ca->ncmpals++;
    hts_expand0(cmpals1_t, ca->ncmpals, ca->mcmpals, ca->cmpals);
    cmpals1_t *cmpals = ca->cmpals + ca->ncmpals - 1;
    free(cmpals->ref);
    cmpals->ref = strdup(rec->d.allele[0]);
    cmpals->n   = rec->n_allele;
    if ( rec->n_allele==2 )
    {
        free(cmpals->alt);
        cmpals->alt = strdup(rec->d.allele[1]);
    }
    else
    {
        if ( cmpals->hash ) khash_str2int_destroy_free(cmpals->hash);
        cmpals->hash = khash_str2int_init();
        int i;
        for (i=1; i<rec->n_allele; i++)
            khash_str2int_inc(cmpals->hash, strdup(rec->d.allele[i]));
    }
}
static int cmpals_match(cmpals_t *ca, bcf1_t *rec)
{
    int i, j;
    for (i=0; i<ca->ncmpals; i++)
    {
        cmpals1_t *cmpals = ca->cmpals + i;
        if ( rec->n_allele != cmpals->n ) continue;

        // NB. assuming both are normalized
        if ( strcasecmp(rec->d.allele[0], cmpals->ref) ) continue;

        // the most frequent case
        if ( rec->n_allele==2 )
        {
            if ( strcasecmp(rec->d.allele[1], cmpals->alt) ) continue;
            return 1;
        }

        khash_t(str2int) *hash = (khash_t(str2int)*) cmpals->hash;
        for (j=1; j<rec->n_allele; j++)
            if ( !khash_str2int_has_key(hash, rec->d.allele[j]) ) break;
        if ( j<rec->n_allele ) continue;
        return 1;
    }
    return 0;
}
static void cmpals_reset(cmpals_t *ca) { ca->ncmpals = 0; }
static void cmpals_destroy(cmpals_t *ca)
{
    int i;
    for (i=0; i<ca->mcmpals; i++)
    {
        cmpals1_t *cmpals = ca->cmpals + i;
        free(cmpals->ref);
        free(cmpals->alt);
        if ( cmpals->hash ) khash_str2int_destroy_free(cmpals->hash);
    }
    free(ca->cmpals);
}

static void flush_buffer(args_t *args, htsFile *file, int n)
{
    bcf1_t *line;
    int i, k;
    int prev_rid = -1, prev_pos = -1, prev_type = 0;
    for (i=0; i<n; i++)
    {
        k = rbuf_shift(&args->rbuf);
        if ( args->mrows_op==MROWS_MERGE )
        {
            if ( mrows_ready_to_flush(args, args->lines[k]) )
            {
                while ( (line=mrows_flush(args)) )
                    if ( bcf_write1(file, args->out_hdr, line)!=0 ) error("[%s] Error: cannot write to %s\n", __func__,args->output_fname);
            }
            int merge = 1;
            if ( args->mrows_collapse!=COLLAPSE_BOTH && args->mrows_collapse!=COLLAPSE_ANY )
            {
                if ( !(bcf_get_variant_types(args->lines[k]) & args->mrows_collapse) ) merge = 0;
            }
            if ( merge )
            {
                mrows_schedule(args, &args->lines[k]);
                continue;
            }
        }
        else if ( args->rmdup )
        {
            int line_type = bcf_get_variant_types(args->lines[k]);
            if ( prev_rid>=0 && prev_rid==args->lines[k]->rid && prev_pos==args->lines[k]->pos )
            {
                if ( args->rmdup & BCF_SR_PAIR_ANY ) continue;    // rmdup by position only
                if ( args->rmdup & BCF_SR_PAIR_SNPS && line_type&(VCF_SNP|VCF_MNP) && prev_type&(VCF_SNP|VCF_MNP) ) continue;
                if ( args->rmdup & BCF_SR_PAIR_INDELS && line_type&(VCF_INDEL) && prev_type&(VCF_INDEL) ) continue;
                if ( args->rmdup & BCF_SR_PAIR_EXACT && cmpals_match(&args->cmpals_out, args->lines[k]) ) continue;
            }
            else
            {
                prev_rid  = args->lines[k]->rid;
                prev_pos  = args->lines[k]->pos;
                prev_type = 0;
                if ( args->rmdup & BCF_SR_PAIR_EXACT ) cmpals_reset(&args->cmpals_out);
            }
            prev_type |= line_type;
            if ( args->rmdup & BCF_SR_PAIR_EXACT ) cmpals_add(&args->cmpals_out, args->lines[k]);
        }
        if ( bcf_write1(file, args->out_hdr, args->lines[k])!=0 ) error("[%s] Error: cannot write to %s\n", __func__,args->output_fname);
    }
    if ( args->mrows_op==MROWS_MERGE && !args->rbuf.n )
    {
        while ( (line=mrows_flush(args)) )
            if ( bcf_write1(file, args->out_hdr, line)!=0 ) error("[%s] Error: cannot write to %s\n", __func__,args->output_fname);
    }
}

static void init_data(args_t *args)
{
    args->hdr = args->files->readers[0].header;
    if ( args->keep_sum_ad )
    {
        args->keep_sum_ad = bcf_hdr_id2int(args->hdr,BCF_DT_ID,"AD");
        if ( args->keep_sum_ad < 0 ) error("Error: --keep-sum-ad requested but the tag AD is not present\n");
    }
    else
        args->keep_sum_ad = -1;

    args->out_hdr = args->hdr;
    if ( args->old_rec_tag )
        bcf_hdr_printf(args->out_hdr,"##INFO=<ID=%s,Number=1,Type=String,Description=\"Original variant. Format: CHR|POS|REF|ALT|USED_ALT_IDX\">",args->old_rec_tag);

    rbuf_init(&args->rbuf, 100);
    args->lines = (bcf1_t**) calloc(args->rbuf.m, sizeof(bcf1_t*));
    if ( args->ref_fname )
    {
        args->fai = fai_load(args->ref_fname);
        if ( !args->fai ) error("Failed to load the fai index: %s\n", args->ref_fname);
    }
    if ( args->mrows_op==MROWS_MERGE )
    {
        args->mrow_out = bcf_init1();
        args->tmp_str = (kstring_t*) calloc(bcf_hdr_nsamples(args->hdr),sizeof(kstring_t));
        args->diploid = (uint8_t*) malloc(bcf_hdr_nsamples(args->hdr));
    }
    if ( args->atomize==SPLIT )
    {
        args->abuf = abuf_init(args->hdr, SPLIT);
        abuf_set_opt(args->abuf, bcf_hdr_t*, BCF_HDR, args->out_hdr);
        if ( args->old_rec_tag )
        {
            abuf_set_opt(args->abuf, const char*, INFO_TAG, args->old_rec_tag);
            if ( bcf_hdr_sync(args->out_hdr)!=0 ) error("bcf_hdr_sync failed\n");
        }
        abuf_set_opt(args->abuf, int, STAR_ALLELE, args->use_star_allele);
    }
    if ( args->gff_fname )
    {
        args->gff = gff_init(args->gff_fname);
        gff_set(args->gff,verbosity,1);
        gff_set(args->gff,strip_chr_names,1);
        gff_parse(args->gff);
        args->idx_tscript = gff_get(args->gff,idx_tscript);
        args->itr_tscript = regitr_init(NULL);
    }
    args->out_hdr = bcf_hdr_dup(args->out_hdr);
}

static void destroy_data(args_t *args)
{
    if ( args->gff )
    {
        gff_destroy(args->gff);
        regitr_destroy(args->itr_tscript);
    }
    cmpals_destroy(&args->cmpals_in);
    cmpals_destroy(&args->cmpals_out);
    int i;
    for (i=0; i<args->rbuf.m; i++)
        if ( args->lines[i] ) bcf_destroy1(args->lines[i]);
    free(args->lines);
    for (i=0; i<args->mtmp_lines; i++)
        if ( args->tmp_lines[i] ) bcf_destroy1(args->tmp_lines[i]);
    free(args->tmp_lines);
    for (i=0; i<args->malines; i++)
        bcf_destroy1(args->alines[i]);
    free(args->alines);
    for (i=0; i<args->mblines; i++)
        bcf_destroy1(args->blines[i]);
    free(args->blines);
    for (i=0; i<args->mmaps; i++)
        free(args->maps[i].map);
    for (i=0; i<args->ntmp_als; i++)
        free(args->tmp_als[i].s);
    for (i=0; i<args->ntmp_del; i++)
        free(args->tmp_del[i].s);
    free(args->tmp_als);
    free(args->tmp_del);
    free(args->tmp_kstr.s);
    if ( args->tmp_str )
    {
        for (i=0; i<bcf_hdr_nsamples(args->hdr); i++) free(args->tmp_str[i].s);
        free(args->tmp_str);
    }
    free(args->maps);
    free(args->als);
    free(args->int32_arr);
    free(args->tmp_arr1);
    free(args->tmp_arr2);
    free(args->diploid);
    if ( args->abuf ) abuf_destroy(args->abuf);
    bcf_hdr_destroy(args->out_hdr);
    if ( args->mrow_out ) bcf_destroy1(args->mrow_out);
    if ( args->fai ) fai_destroy(args->fai);
    if ( args->mseq ) free(args->seq);
}


static void normalize_line(args_t *args, bcf1_t *line)
{
    if ( args->fai )
    {
        if ( args->check_ref & CHECK_REF_FIX ) fix_ref(args, line);
        if ( args->do_indels )
        {
            int ret = realign(args, line);

            // exclude broken VCF lines
            if ( ret==ERR_REF_MISMATCH && args->check_ref & CHECK_REF_SKIP )
            {
                args->nskipped++;
                return;
            }
            if ( ret==ERR_DUP_ALLELE )
            {
                if ( args->check_ref & CHECK_REF_FIX )
                    fix_dup_alt(args, line);
                else if ( args->check_ref==CHECK_REF_EXIT )
                    error("Duplicate alleles at %s:%"PRId64"; run with -cw to turn the error into warning or with -cs to fix.\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1);
                else if ( args->check_ref & CHECK_REF_WARN )
                    fprintf(stderr,"ALT_DUP\t%s\t%"PRId64"\n", bcf_seqname(args->hdr,line),(int64_t) line->pos+1);
            }
        }
    }

    // insert into sorted buffer
    rbuf_expand0(&args->rbuf,bcf1_t*,args->rbuf.n+1,args->lines);
    int i,j;
    i = j = rbuf_append(&args->rbuf);
    if ( args->lines[i] ) bcf_destroy(args->lines[i]);
    args->lines[i] = bcf_dup(line);
    while ( rbuf_prev(&args->rbuf,&i) )
    {
        if ( args->lines[i]->pos > args->lines[j]->pos ) SWAP(bcf1_t*, args->lines[i], args->lines[j]);
        j = i;
    }
}

static bcf1_t *next_atomized_line(args_t *args)
{
    bcf1_t *rec = NULL;
    if ( args->atomize==SPLIT )
    {
        rec = abuf_flush(args->abuf, 0);
        if ( rec ) return rec;
    }

    if ( !bcf_sr_next_line(args->files) ) return NULL;

    if ( args->atomize==SPLIT )
    {
        abuf_push(args->abuf,bcf_sr_get_line(args->files,0));
        return abuf_flush(args->abuf, 0);
    }
    return bcf_sr_get_line(args->files,0);
}
static void normalize_vcf(args_t *args)
{
    char wmode[8];
    set_wmode(wmode,args->output_type,args->output_fname,args->clevel);
    args->out = hts_open(args->output_fname ? args->output_fname : "-", wmode);
    if ( args->out == NULL ) error("Can't write to \"%s\": %s\n", args->output_fname, strerror(errno));
    if ( args->n_threads )
        hts_set_opt(args->out, HTS_OPT_THREAD_POOL, args->files->p);
    if (args->record_cmd_line) bcf_hdr_append_version(args->out_hdr, args->argc, args->argv, "bcftools_norm");
    if ( bcf_hdr_write(args->out, args->out_hdr)!=0 ) error("[%s] Error: cannot write to %s\n", __func__,args->output_fname);
    if ( args->write_index && init_index(args->out,args->out_hdr,args->output_fname,&args->index_fn)<0 ) error("Error: failed to initialise index for %s\n",args->output_fname);

    bcf1_t *line;
    int prev_rid = -1, prev_pos = -1, prev_type = 0;
    while ( (line = next_atomized_line(args)) )
    {
        args->ntotal++;
        if ( args->rmdup )
        {
            int line_type = bcf_get_variant_types(line);
            if ( prev_rid>=0 && prev_rid==line->rid && prev_pos==line->pos )
            {
                if ( args->rmdup & BCF_SR_PAIR_ANY ) continue;    // rmdup by position only
                if ( args->rmdup & BCF_SR_PAIR_SNPS && line_type&(VCF_SNP|VCF_MNP) && prev_type&(VCF_SNP|VCF_MNP) ) continue;
                if ( args->rmdup & BCF_SR_PAIR_INDELS && line_type&(VCF_INDEL) && prev_type&(VCF_INDEL) ) continue;
                if ( args->rmdup & BCF_SR_PAIR_EXACT && cmpals_match(&args->cmpals_in, line) ) continue;
            }
            else
            {
                prev_rid  = line->rid;
                prev_pos  = line->pos;
                prev_type = 0;
                if ( args->rmdup & BCF_SR_PAIR_EXACT ) cmpals_reset(&args->cmpals_in);
            }
            prev_type |= line_type;
            if ( args->rmdup & BCF_SR_PAIR_EXACT ) cmpals_add(&args->cmpals_in, line);
        }

        // still on the same chromosome?
        int i,j,ilast = rbuf_last(&args->rbuf);
        if ( ilast>=0 && line->rid != args->lines[ilast]->rid ) flush_buffer(args, args->out, args->rbuf.n); // new chromosome

        int split = 0;
        if ( args->mrows_op==MROWS_SPLIT )
        {
            split = 1;
            if ( args->mrows_collapse!=COLLAPSE_BOTH && args->mrows_collapse!=COLLAPSE_ANY )
            {
                if ( !(bcf_get_variant_types(line) & args->mrows_collapse) ) split = 0;
            }
            if ( split && line->n_allele>2 )
            {
                args->nsplit++;
                split_multiallelic_to_biallelics(args, line);
                for (j=0; j<args->ntmp_lines; j++)
                    normalize_line(args, args->tmp_lines[j]);
            }
            else
                split = 0;
        }
        if ( !split )
            normalize_line(args, line);

        // find out how many sites to flush
        ilast = rbuf_last(&args->rbuf);
        j = 0;
        for (i=-1; rbuf_next(&args->rbuf,&i); )
        {
            if ( args->lines[ilast]->pos - args->lines[i]->pos < args->buf_win ) break;
            j++;
        }
        if ( j>0 ) flush_buffer(args, args->out, j);
    }
    flush_buffer(args, args->out, args->rbuf.n);
    if ( args->write_index )
    {
        if ( bcf_idx_save(args->out)<0 )
        {
            if ( hts_close(args->out)!=0 ) error("Error: close failed .. %s\n", args->output_fname?args->output_fname:"stdout");
            error("Error: cannot write to index %s\n", args->index_fn);
        }
        free(args->index_fn);
    }
    if ( hts_close(args->out)!=0 ) error("[%s] Error: close failed .. %s\n", __func__,args->output_fname);

    fprintf(stderr,"Lines   total/split/joined/realigned/skipped:\t%d/%d/%d/%d/%d\n", args->ntotal,args->nsplit,args->njoined,args->nchanged,args->nskipped);
    if ( args->check_ref & CHECK_REF_FIX )
        fprintf(stderr,"REF/ALT total/modified/added:  \t%d/%d/%d\n", args->nref.tot,args->nref.swap,args->nref.set);
}

static void usage(void)
{
    fprintf(stderr, "\n");
    fprintf(stderr, "About:   Left-align and normalize indels; check if REF alleles match the reference;\n");
    fprintf(stderr, "         split multiallelic sites into multiple rows; recover multiallelics from\n");
    fprintf(stderr, "         multiple rows.\n");
    fprintf(stderr, "Usage:   bcftools norm [options] <in.vcf.gz>\n");
    fprintf(stderr, "\n");
    fprintf(stderr, "Options:\n");
    fprintf(stderr, "    -a, --atomize                   Decompose complex variants (e.g. MNVs become consecutive SNVs)\n");
    fprintf(stderr, "        --atom-overlaps '*'|.       Use the star allele (*) for overlapping alleles or set to missing (.) [*]\n");
    fprintf(stderr, "    -c, --check-ref e|w|x|s         Check REF alleles and exit (e), warn (w), exclude (x), or set (s) bad sites [e]\n");
    fprintf(stderr, "    -D, --remove-duplicates         Remove duplicate lines of the same type.\n");
    fprintf(stderr, "    -d, --rm-dup TYPE               Remove duplicate snps|indels|both|all|exact\n");
    fprintf(stderr, "    -f, --fasta-ref FILE            Reference sequence\n");
    fprintf(stderr, "        --force                     Try to proceed even if malformed tags are encountered. Experimental, use at your own risk\n");
    fprintf(stderr, "    -g, --gff-annot FILE            Follow HGVS 3'rule and right-align variants in transcripts on the forward strand\n");
    fprintf(stderr, "        --keep-sum TAG,..           Keep vector sum constant when splitting multiallelics (see github issue #360)\n");
    fprintf(stderr, "    -m, --multiallelics -|+TYPE     Split multiallelics (-) or join biallelics (+), type: snps|indels|both|any [both]\n");
    fprintf(stderr, "        --multi-overlaps 0|.        Fill in the reference (0) or missing (.) allele when splitting multiallelics [0]\n");
    fprintf(stderr, "        --no-version                Do not append version and command line to the header\n");
    fprintf(stderr, "    -N, --do-not-normalize          Do not normalize indels (with -m or -c s)\n");
    fprintf(stderr, "        --old-rec-tag STR           Annotate modified records with INFO/STR indicating the original variant\n");
    fprintf(stderr, "    -o, --output FILE               Write output to a file [standard output]\n");
    fprintf(stderr, "    -O, --output-type u|b|v|z[0-9]  u/b: un/compressed BCF, v/z: un/compressed VCF, 0-9: compression level [v]\n");
    fprintf(stderr, "    -r, --regions REGION            Restrict to comma-separated list of regions\n");
    fprintf(stderr, "    -R, --regions-file FILE         Restrict to regions listed in a file\n");
    fprintf(stderr, "        --regions-overlap 0|1|2     Include if POS in the region (0), record overlaps (1), variant overlaps (2) [1]\n");
    fprintf(stderr, "    -s, --strict-filter             When merging (-m+), merged site is PASS only if all sites being merged PASS\n");
    fprintf(stderr, "    -t, --targets REGION            Similar to -r but streams rather than index-jumps\n");
    fprintf(stderr, "    -T, --targets-file FILE         Similar to -R but streams rather than index-jumps\n");
    fprintf(stderr, "        --targets-overlap 0|1|2     Include if POS in the region (0), record overlaps (1), variant overlaps (2) [0]\n");
    fprintf(stderr, "        --threads INT               Use multithreading with INT worker threads [0]\n");
    fprintf(stderr, "    -w, --site-win INT              Buffer for sorting lines which changed position during realignment [1000]\n");
    fprintf(stderr, "        --write-index               Automatically index the output files [off]\n");
    fprintf(stderr, "\n");
    fprintf(stderr, "Examples:\n");
    fprintf(stderr, "   # normalize and left-align indels\n");
    fprintf(stderr, "   bcftools norm -f ref.fa in.vcf\n");
    fprintf(stderr, "\n");
    fprintf(stderr, "   # split multi-allelic sites\n");
    fprintf(stderr, "   bcftools norm -m- in.vcf\n");
    fprintf(stderr, "\n");
    exit(1);
}

int main_vcfnorm(int argc, char *argv[])
{
    int c;
    args_t *args  = (args_t*) calloc(1,sizeof(args_t));
    args->argc    = argc; args->argv = argv;
    args->files   = bcf_sr_init();
    args->output_fname = "-";
    args->output_type = FT_VCF;
    args->n_threads = 0;
    args->record_cmd_line = 1;
    args->aln_win = 100;
    args->buf_win = 1000;
    args->mrows_collapse = COLLAPSE_BOTH;
    args->do_indels = 1;
    args->ma_use_ref_allele = 1;
    args->clevel = -1;
    int region_is_file  = 0;
    int targets_is_file = 0;
    args->use_star_allele = 1;
    int regions_overlap = 1;
    int targets_overlap = 0;

    static struct option loptions[] =
    {
        {"help",no_argument,NULL,'h'},
        {"force",no_argument,NULL,7},
        {"atomize",no_argument,NULL,'a'},
        {"atom-overlaps",required_argument,NULL,11},
        {"old-rec-tag",required_argument,NULL,12},
        {"keep-sum",required_argument,NULL,10},
        {"fasta-ref",required_argument,NULL,'f'},
        {"gff-annot",required_argument,NULL,'g'},
        {"right-align",no_argument,NULL,15},            // undocumented, only for debugging
        {"do-not-normalize",no_argument,NULL,'N'},
        {"multiallelics",required_argument,NULL,'m'},
        {"multi-overlaps",required_argument,NULL,13},
        {"regions",required_argument,NULL,'r'},
        {"regions-file",required_argument,NULL,'R'},
        {"regions-overlap",required_argument,NULL,1},
        {"targets",required_argument,NULL,'t'},
        {"targets-file",required_argument,NULL,'T'},
        {"targets-overlap",required_argument,NULL,2},
        {"site-win",required_argument,NULL,'w'},
        {"remove-duplicates",no_argument,NULL,'D'},
        {"rm-dup",required_argument,NULL,'d'},
        {"output",required_argument,NULL,'o'},
        {"output-type",required_argument,NULL,'O'},
        {"threads",required_argument,NULL,9},
        {"check-ref",required_argument,NULL,'c'},
        {"strict-filter",no_argument,NULL,'s'},
        {"no-version",no_argument,NULL,8},
        {"write-index",no_argument,NULL,14},
        {NULL,0,NULL,0}
    };
    char *tmp;
    while ((c = getopt_long(argc, argv, "hr:R:f:w:Dd:o:O:c:m:t:T:sNag:",loptions,NULL)) >= 0) {
        switch (c) {
            case  10:
                // possibly generalize this also to INFO/AD and other tags
                if ( strcasecmp("ad",optarg) )
                    error("Error: only --keep-sum AD is currently supported. See https://github.com/samtools/bcftools/issues/360 for more.\n");
                args->keep_sum_ad = 1;  // this will be set to the header id or -1 in init_data
                break;
            case 'g': args->gff_fname = optarg; break;
            case 'a': args->atomize = SPLIT; break;
            case 11 :
                if ( optarg[0]=='*' ) args->use_star_allele = 1;
                else if ( optarg[0]=='.' ) args->use_star_allele = 0;
                else error("Invalid argument to --atom-overlaps. Perhaps you wanted: \"--atom-overlaps '*'\"?\n");
                break;
            case 12 : args->old_rec_tag = optarg; break;
            case 13 :
                if ( optarg[0]=='0' ) args->ma_use_ref_allele = 1;
                else if ( optarg[0]=='.' ) args->ma_use_ref_allele = 0;
                else error("Invalid argument to --multi-overlaps\n");
                break;
            case 14 : args->write_index = 1; break;
            case 15 : args->right_align = 1; break;
            case 'N': args->do_indels = 0; break;
            case 'd':
                if ( !strcmp("snps",optarg) ) args->rmdup = BCF_SR_PAIR_SNPS;
                else if ( !strcmp("indels",optarg) ) args->rmdup = BCF_SR_PAIR_INDELS;
                else if ( !strcmp("both",optarg) ) args->rmdup = BCF_SR_PAIR_BOTH;
                else if ( !strcmp("all",optarg) ) args->rmdup = BCF_SR_PAIR_ANY;
                else if ( !strcmp("any",optarg) ) args->rmdup = BCF_SR_PAIR_ANY;
                else if ( !strcmp("none",optarg) ) args->rmdup = BCF_SR_PAIR_EXACT;
                else if ( !strcmp("exact",optarg) ) args->rmdup = BCF_SR_PAIR_EXACT;
                else error("The argument to -d not recognised: %s\n", optarg);
                break;
            case 'm':
                if ( optarg[0]=='-' ) args->mrows_op = MROWS_SPLIT;
                else if ( optarg[0]=='+' ) args->mrows_op = MROWS_MERGE;
                else error("Expected '+' or '-' with -m\n");
                if ( optarg[1]!=0 )
                {
                    if ( !strcmp("snps",optarg+1) ) args->mrows_collapse = COLLAPSE_SNPS;
                    else if ( !strcmp("indels",optarg+1) ) args->mrows_collapse = COLLAPSE_INDELS;
                    else if ( !strcmp("both",optarg+1) ) args->mrows_collapse = COLLAPSE_BOTH;
                    else if ( !strcmp("any",optarg+1) ) args->mrows_collapse = COLLAPSE_ANY;
                    else error("The argument to -m not recognised: %s\n", optarg);
                }
                break;
            case 'c':
                if ( strchr(optarg,'w') ) args->check_ref |= CHECK_REF_WARN;
                if ( strchr(optarg,'x') ) args->check_ref |= CHECK_REF_SKIP;
                if ( strchr(optarg,'s') ) args->check_ref |= CHECK_REF_FIX;
                if ( strchr(optarg,'e') ) args->check_ref = CHECK_REF_EXIT; // overrides the above
                break;
            case 'O':
                switch (optarg[0]) {
                    case 'b': args->output_type = FT_BCF_GZ; break;
                    case 'u': args->output_type = FT_BCF; break;
                    case 'z': args->output_type = FT_VCF_GZ; break;
                    case 'v': args->output_type = FT_VCF; break;
                    default:
                    {
                        args->clevel = strtol(optarg,&tmp,10);
                        if ( *tmp || args->clevel<0 || args->clevel>9 ) error("The output type \"%s\" not recognised\n", optarg);
                    }
                }
                if ( optarg[1] )
                {
                    args->clevel = strtol(optarg+1,&tmp,10);
                    if ( *tmp || args->clevel<0 || args->clevel>9 ) error("Could not parse argument: --compression-level %s\n", optarg+1);
                }
                break;
            case 'o': args->output_fname = optarg; break;
            case 'D':
                fprintf(stderr,"Warning: `-D` is functional but deprecated, replaced by and alias of `-d none`.\n");
                args->rmdup = BCF_SR_PAIR_EXACT;
                break;
            case 's': args->strict_filter = 1; break;
            case 'f': args->ref_fname = optarg; break;
            case 'r': args->region = optarg; break;
            case 'R': args->region = optarg; region_is_file = 1; break;
            case 't': args->targets = optarg; break;
            case 'T': args->targets = optarg; targets_is_file = 1; break;
            case 'w':
                args->buf_win = strtol(optarg,&tmp,10);
                if ( *tmp ) error("Could not parse argument: --site-win %s\n", optarg);
                break;
            case  9 : args->n_threads = strtol(optarg, 0, 0); break;
            case  8 : args->record_cmd_line = 0; break;
            case  7 : args->force = 1; break;
            case  1 :
                regions_overlap = parse_overlap_option(optarg);
                if ( regions_overlap < 0 ) error("Could not parse: --regions-overlap %s\n",optarg);
                break;
            case  2 :
                targets_overlap = parse_overlap_option(optarg);
                if ( targets_overlap < 0 ) error("Could not parse: --targets-overlap %s\n",optarg);
                break;
            case 'h':
            case '?': usage(); break;
            default: error("Unknown argument: %s\n", optarg);
        }
    }

    char *fname = NULL;
    if ( optind>=argc )
    {
        if ( !isatty(fileno((FILE *)stdin)) ) fname = "-";  // reading from stdin
        else usage();
    }
    else fname = argv[optind];

    if ( !args->ref_fname && !args->mrows_op && !args->rmdup && args->atomize==NONE ) error("Expected -a, -f, -m, -D or -d option\n");
    if ( !args->check_ref && args->ref_fname ) args->check_ref = CHECK_REF_EXIT;
    if ( args->check_ref && !args->ref_fname ) error("Expected --fasta-ref with --check-ref\n");

    if ( args->region )
    {
        bcf_sr_set_opt(args->files,BCF_SR_REGIONS_OVERLAP,regions_overlap);
        if ( bcf_sr_set_regions(args->files, args->region,region_is_file)<0 )
            error("Failed to read the regions: %s\n", args->region);
    }
    if ( args->targets )
    {
        bcf_sr_set_opt(args->files,BCF_SR_TARGETS_OVERLAP,targets_overlap);
        if ( bcf_sr_set_targets(args->files, args->targets,targets_is_file, 0)<0 )
            error("Failed to read the targets: %s\n", args->targets);
    }

    if ( bcf_sr_set_threads(args->files, args->n_threads)<0 ) error("Failed to create threads\n");
    if ( !bcf_sr_add_reader(args->files, fname) ) error("Failed to read from %s: %s\n", !strcmp("-",fname)?"standard input":fname,bcf_sr_strerror(args->files->errnum));
    if ( args->mrows_op&MROWS_SPLIT && args->rmdup ) error("Cannot combine -D and -m-\n");
    init_data(args);
    normalize_vcf(args);
    destroy_data(args);
    bcf_sr_destroy(args->files);
    free(args);
    return 0;
}