+ // fill the bases array
+ memset(r, 0, sizeof(bcf_callret1_t));
+ r->indelreg = 10000;
+ n_ins = n_del = 0;
+ for (i = n = 0; i < _n; ++i) {
+ const bam_pileup1_t *p = pl + i;
+ int q, b, mapQ, indelQ, is_diff, min_dist;
+ if (p->is_del || p->is_refskip || (p->b->core.flag&BAM_FUNMAP)) continue;
+ { // compute indel (base) quality
+ // this can be made more efficient, but realignment is the bottleneck anyway
+ int j, k, x, y, op, len = 0, max_left, max_rght, seqQ, indelreg;
+ bam1_core_t *c = &p->b->core;
+ uint32_t *cigar = bam1_cigar(p->b);
+ uint8_t *qual = bam1_qual(p->b);
+ for (k = y = 0, x = c->pos; k < c->n_cigar && y <= p->qpos; ++k) {
+ op = cigar[k]&0xf;
+ len = cigar[k]>>4;
+ if (op == BAM_CMATCH) {
+ if (pos > x && pos < x + len) break;
+ x += len; y += len;
+ } else if (op == BAM_CINS || op == BAM_CSOFT_CLIP) y += len;
+ else if (op == BAM_CDEL || op == BAM_CREF_SKIP) x += len;
+ }
+ if (k == c->n_cigar) continue; // this actually should not happen
+ max_left = max_rght = 0; indelreg = 0;
+ if (pos == x + len - 1 && k+2 < c->n_cigar && ((cigar[k+1]&0xf) == BAM_CINS || (cigar[k+1]&0xf) == BAM_CDEL)
+ && (cigar[k+2]&0xf) == BAM_CMATCH)
+ {
+ for (j = y; j < y + len; ++j)
+ if (max_left < qual[j]) max_left = qual[j];
+ if ((cigar[k+1]&0xf) == BAM_CINS) y += cigar[k+1]>>4;
+ else x += cigar[k+1]>>4;
+ op = cigar[k+2]&0xf; len = cigar[k+2]>>4;
+ for (j = y; j < y + len; ++j) {
+ if (max_rght < qual[j]) max_rght = qual[j];
+ if (qual[j] > BAM2BCF_INDELREG_THRES && indelreg == 0)
+ indelreg = j - y + 1;
+ }
+ if (r->indelreg > indelreg) r->indelreg = indelreg;
+ } else {
+ for (j = y; j <= p->qpos; ++j)
+ if (max_left < qual[j]) max_left = qual[j];
+ for (j = p->qpos + 1; j < y + len; ++j)
+ if (max_rght < qual[j]) max_rght = qual[j];
+
+ }
+ indelQ = max_left < max_rght? max_left : max_rght;
+ // estimate the sequencing error rate
+ seqQ = bca->openQ;
+ if (p->indel != 0) seqQ += bca->extQ * (abs(p->indel) - 1); // FIXME: better to model homopolymer
+ if (p->indel > 0) {
+ ++n_ins; r->ins_len += p->indel;
+ } else if (p->indel < 0) {
+ ++n_del; r->del_len += -p->indel;
+ }
+ if (p->indel != 0) { // a different model for tandem repeats
+ uint8_t *seq = bam1_seq(p->b);
+ int tandemQ, qb = bam1_seqi(seq, p->qpos), l;
+ for (j = p->qpos + 1; j < c->l_qseq; ++j)
+ if (qb != bam1_seqi(seq, j)) break;
+ l = j;
+ for (j = (int)p->qpos - 1; j >= 0; --j)
+ if (qb != bam1_seqi(seq, j)) break;
+ l = l - (j + 1);
+ tandemQ = (int)((double)(abs(p->indel)) / l * bca->tandemQ + .499);
+ if (seqQ > tandemQ) seqQ = tandemQ;
+ }
+// fprintf(stderr, "%s\t%d\t%d\t%d\t%d\t%d\t%d\n", bam1_qname(p->b), pos+1, p->indel, indelQ, seqQ, max_left, max_rght);
+ if (indelQ > seqQ) indelQ = seqQ;
+ q = indelQ;