r815: fixed a memory leak

Makefile

@@ -1,7 +1,7 @@
 CFLAGS=		-g -Wall -O2 -Wc++-compat #-Wextra
 CPPFLAGS=	-DHAVE_KALLOC
 INCLUDES=
-OBJS=		kthread.o kalloc.o misc.o bseq.o sketch.o sdust.o options.o index.o chain.o align.o hit.o map.o format.o pe.o esterr.o ksw2_ll_sse.o
+OBJS=		kthread.o kalloc.o misc.o bseq.o sketch.o sdust.o options.o index.o chain.o align.o hit.o map.o format.o pe.o esterr.o splitidx.o ksw2_ll_sse.o
 PROG=		minimap2
 PROG_EXTRA=	sdust minimap2-lite
 LIBS=		-lm -lz -lpthread

align.c

@@ -199,12 +199,12 @@ static void mm_append_cigar(mm_reg1_t *r, uint32_t n_cigar, uint32_t *cigar) //
 	mm_extra_t *p;
 	if (n_cigar == 0) return;
 	if (r->p == 0) {
-		uint32_t capacity = n_cigar + sizeof(mm_extra_t); // TODO: should this be "n_cigar + sizeof(mm_extra_t)/4" instead?
+		uint32_t capacity = n_cigar + sizeof(mm_extra_t)/4;
 		kroundup32(capacity);
 		r->p = (mm_extra_t*)calloc(capacity, 4);
 		r->p->capacity = capacity;
-	} else if (r->p->n_cigar + n_cigar + sizeof(mm_extra_t) > r->p->capacity) {
-		r->p->capacity = r->p->n_cigar + n_cigar + sizeof(mm_extra_t);
+	} else if (r->p->n_cigar + n_cigar + sizeof(mm_extra_t)/4 > r->p->capacity) {
+		r->p->capacity = r->p->n_cigar + n_cigar + sizeof(mm_extra_t)/4;
 		kroundup32(r->p->capacity);
 		r->p = (mm_extra_t*)realloc(r->p, r->p->capacity * 4);
 	}
@@ -878,6 +878,6 @@ mm_reg1_t *mm_align_skeleton(void *km, const mm_mapopt_t *opt, const mm_idx_t *m
 	kfree(km, qseq0[0]);
 	kfree(km, ez.cigar);
 	mm_filter_regs(opt, qlen, n_regs_, regs);
-	mm_hit_sort_by_dp(km, n_regs_, regs);
+	mm_hit_sort(km, n_regs_, regs);
 	return regs;
 }

format.c

@@ -259,6 +259,10 @@ static inline void write_tags(kstring_t *s, const mm_reg1_t *r)
 void mm_write_paf(kstring_t *s, const mm_idx_t *mi, const mm_bseq1_t *t, const mm_reg1_t *r, void *km, int opt_flag)
 {
 	s->l = 0;
+	if (r == 0) {
+		mm_sprintf_lite(s, "%s\t%d", t->name, t->l_seq);
+		return;
+	}
 	mm_sprintf_lite(s, "%s\t%d\t%d\t%d\t%c\t", t->name, t->l_seq, r->qs, r->qe, "+-"[r->rev]);
 	if (mi->seq[r->rid].name) mm_sprintf_lite(s, "%s", mi->seq[r->rid].name);
 	else mm_sprintf_lite(s, "%d", r->rid);

hit.c

@@ -164,9 +164,9 @@ set_parent_test:
 	kfree(km, w);
 }
 
-void mm_hit_sort_by_dp(void *km, int *n_regs, mm_reg1_t *r)
+void mm_hit_sort(void *km, int *n_regs, mm_reg1_t *r)
 {
-	int32_t i, n_aux, n = *n_regs;
+	int32_t i, n_aux, n = *n_regs, has_cigar = 0, no_cigar = 0;
 	mm128_t *aux;
 	mm_reg1_t *t;
 
@@ -175,14 +175,20 @@ void mm_hit_sort_by_dp(void *km, int *n_regs, mm_reg1_t *r)
 	t = (mm_reg1_t*)kmalloc(km, n * sizeof(mm_reg1_t));
 	for (i = n_aux = 0; i < n; ++i) {
 		if (r[i].inv || r[i].cnt > 0) { // squeeze out elements with cnt==0 (soft deleted)
-			assert(r[i].p);
-			aux[n_aux].x = (uint64_t)r[i].p->dp_max << 32 | r[i].hash;
+			if (r[i].p) {
+				aux[n_aux].x = (uint64_t)r[i].p->dp_max << 32 | r[i].hash;
+				has_cigar = 1;
+			} else {
+				aux[n_aux].x = (uint64_t)r[i].score << 32 | r[i].hash;
+				no_cigar = 1;
+			}
 			aux[n_aux++].y = i;
 		} else if (r[i].p) {
 			free(r[i].p);
 			r[i].p = 0;
 		}
 	}
+	assert(has_cigar + no_cigar == 1);
 	radix_sort_128x(aux, aux + n_aux);
 	for (i = n_aux - 1; i >= 0; --i)
 		t[n_aux - 1 - i] = r[aux[i].y];

index.c

@@ -48,10 +48,12 @@ void mm_idx_destroy(mm_idx_t *mi)
 	uint32_t i;
 	if (mi == 0) return;
 	if (mi->h) kh_destroy(str, (khash_t(str)*)mi->h);
-	for (i = 0; i < 1U<<mi->b; ++i) {
-		free(mi->B[i].p);
-		free(mi->B[i].a.a);
-		kh_destroy(idx, (idxhash_t*)mi->B[i].h);
+	if (mi->B) {
+		for (i = 0; i < 1U<<mi->b; ++i) {
+			free(mi->B[i].p);
+			free(mi->B[i].a.a);
+			kh_destroy(idx, (idxhash_t*)mi->B[i].h);
+		}
 	}
 	if (!mi->km) {
 		for (i = 0; i < mi->n_seq; ++i)
@@ -563,7 +565,7 @@ mm_idx_t *mm_idx_reader_read(mm_idx_reader_t *r, int n_threads)
 		mi = mm_idx_gen(r->fp.seq, r->opt.w, r->opt.k, r->opt.bucket_bits, r->opt.flag, r->opt.mini_batch_size, n_threads, r->opt.batch_size);
 	if (mi) {
 		if (r->fp_out) mm_idx_dump(r->fp_out, mi);
-		++r->n_parts;
+		mi->index = r->n_parts++;
 	}
 	return mi;
 }

main.c

@@ -10,7 +10,7 @@
 #include "getopt.h"
 #endif
 
-#define MM_VERSION "2.11-r797"
+#define MM_VERSION "2.11-r815-dirty"
 
 #ifdef __linux__
 #include <sys/resource.h>
@@ -60,6 +60,8 @@ static struct option long_options[] = {
 	{ "lj-min-ratio",   required_argument, 0, 0 },   // 30
 	{ "score-N",        required_argument, 0, 0 },   // 31
 	{ "eqx",            no_argument,       0, 0 },   // 32
+	{ "paf-no-hit",     no_argument,       0, 0 },   // 33
+	{ "split-prefix",   required_argument, 0, 0 },   // 34
 	{ "help",           no_argument,       0, 'h' },
 	{ "max-intron-len", required_argument, 0, 'G' },
 	{ "version",        no_argument,       0, 'V' },
@@ -100,7 +102,7 @@ int main(int argc, char *argv[])
 	const char *opt_str = "2aSDw:k:K:t:r:f:Vv:g:G:I:d:XT:s:x:Hcp:M:n:z:A:B:O:E:m:N:Qu:R:hF:LC:yY";
 	mm_mapopt_t opt;
 	mm_idxopt_t ipt;
-	int i, c, n_threads = 3, long_idx;
+	int i, c, n_threads = 3, n_parts, long_idx;
 	char *fnw = 0, *rg = 0, *s;
 	FILE *fp_help = stderr;
 	mm_idx_reader_t *idx_rdr;
@@ -179,6 +181,8 @@ int main(int argc, char *argv[])
 		else if (c == 0 && long_idx ==30) opt.min_join_flank_ratio = atof(optarg); // --lj-min-ratio
 		else if (c == 0 && long_idx ==31) opt.sc_ambi = atoi(optarg); // --score-N
 		else if (c == 0 && long_idx ==32) opt.flag |= MM_F_EQX; // --eqx
+		else if (c == 0 && long_idx ==33) opt.flag |= MM_F_PAF_NO_HIT; // --paf-no-hit
+		else if (c == 0 && long_idx ==34) opt.split_prefix = optarg; // --split-prefix
 		else if (c == 0 && long_idx == 14) { // --frag
 			yes_or_no(&opt, MM_F_FRAG_MODE, long_idx, optarg, 1);
 		} else if (c == 0 && long_idx == 15) { // --secondary
@@ -325,8 +329,8 @@ int main(int argc, char *argv[])
 				mm_write_sam_hdr(mi, rg, MM_VERSION, argc, argv);
 			} else {
 				mm_write_sam_hdr(0, rg, MM_VERSION, argc, argv);
-				if (mm_verbose >= 2)
-					fprintf(stderr, "[WARNING]\033[1;31m For a multi-part index, no @SQ lines will be outputted.\033[0m\n");
+				if (opt.split_prefix == 0 && mm_verbose >= 2)
+					fprintf(stderr, "[WARNING]\033[1;31m For a multi-part index, no @SQ lines will be outputted. Please use --split-prefix.\033[0m\n");
 			}
 		}
 		if (mm_verbose >= 3)
@@ -342,8 +346,12 @@ int main(int argc, char *argv[])
 		}
 		mm_idx_destroy(mi);
 	}
+	n_parts = idx_rdr->n_parts;
 	mm_idx_reader_close(idx_rdr);
 
+	if (opt.split_prefix)
+		mm_split_merge(argc - (optind + 1), (const char**)&argv[optind + 1], &opt, n_parts);
+
 	if (fflush(stdout) == EOF) {
 		fprintf(stderr, "[ERROR] failed to write the results\n");
 		exit(EXIT_FAILURE);

map.c

@@ -11,6 +11,7 @@
 
 struct mm_tbuf_s {
 	void *km;
+	int rep_len, frag_gap;
 };
 
 mm_tbuf_t *mm_tbuf_init(void)
@@ -330,6 +331,8 @@ void mm_map_frag(const mm_idx_t *mi, int n_segs, const int *qlens, const char **
 			a = mm_chain_dp(max_chain_gap_ref, max_chain_gap_qry, opt->bw, opt->max_chain_skip, opt->min_cnt, opt->min_chain_score, is_splice, n_segs, n_a, a, &n_regs0, &u, b->km);
 		}
 	}
+	b->frag_gap = max_chain_gap_ref;
+	b->rep_len = rep_len;
 
 	regs0 = mm_gen_regs(b->km, hash, qlen_sum, n_regs0, u, a);
 
@@ -394,13 +397,17 @@ typedef struct {
 	mm_bseq_file_t **fp;
 	const mm_idx_t *mi;
 	kstring_t str;
+
+	int n_parts;
+	uint32_t *rid_shift;
+	FILE *fp_split, **fp_parts;
 } pipeline_t;
 
 typedef struct {
 	const pipeline_t *p;
     int n_seq, n_frag;
 	mm_bseq1_t *seq;
-	int *n_reg, *seg_off, *n_seg;
+	int *n_reg, *seg_off, *n_seg, *rep_len, *frag_gap;
 	mm_reg1_t **reg;
 	mm_tbuf_t **buf;
 } step_t;
@@ -421,10 +428,17 @@ static void worker_for(void *_data, long i, int tid) // kt_for() callback
 		qseqs[j] = s->seq[off + j].seq;
 	}
 	if (s->p->opt->flag & MM_F_INDEPEND_SEG) {
-		for (j = 0; j < s->n_seg[i]; ++j)
+		for (j = 0; j < s->n_seg[i]; ++j) {
 			mm_map_frag(s->p->mi, 1, &qlens[j], &qseqs[j], &s->n_reg[off+j], &s->reg[off+j], b, s->p->opt, s->seq[off+j].name);
+			s->rep_len[off + j] = b->rep_len;
+			s->frag_gap[off + j] = b->frag_gap;
+		}
 	} else {
 		mm_map_frag(s->p->mi, s->n_seg[i], qlens, qseqs, &s->n_reg[off], &s->reg[off], b, s->p->opt, s->seq[off].name);
+		for (j = 0; j < s->n_seg[i]; ++j) {
+			s->rep_len[off + j] = b->rep_len;
+			s->frag_gap[off + j] = b->frag_gap;
+		}
 	}
 	for (j = 0; j < s->n_seg[i]; ++j) // flip the query strand and coordinate to the original read strand
 		if (s->n_seg[i] == 2 && ((j == 0 && (pe_ori>>1&1)) || (j == 1 && (pe_ori&1)))) {
@@ -440,6 +454,63 @@ static void worker_for(void *_data, long i, int tid) // kt_for() callback
 		}
 }
 
+static void merge_hits(step_t *s)
+{
+	int f, i, k0, k, max_seg = 0, *n_reg_part, *rep_len_part, *frag_gap_part, *qlens;
+	void *km;
+	FILE **fp = s->p->fp_parts;
+	const mm_mapopt_t *opt = s->p->opt;
+
+	km = km_init();
+	for (f = 0; f < s->n_frag; ++f)
+		max_seg = max_seg > s->n_seg[f]? max_seg : s->n_seg[f];
+	qlens = CALLOC(int, max_seg + s->p->n_parts * 3);
+	n_reg_part = qlens + max_seg;
+	rep_len_part = n_reg_part + s->p->n_parts;
+	frag_gap_part = rep_len_part + s->p->n_parts;
+	for (f = 0, k = k0 = 0; f < s->n_frag; ++f) {
+		k0 = k;
+		for (i = 0; i < s->n_seg[f]; ++i, ++k) {
+			int j, l, t, rep_len = 0;
+			qlens[i] = s->seq[k].l_seq;
+			for (j = 0, s->n_reg[k] = 0; j < s->p->n_parts; ++j) {
+				mm_err_fread(&n_reg_part[j],    sizeof(int), 1, fp[j]);
+				mm_err_fread(&rep_len_part[j],  sizeof(int), 1, fp[j]);
+				mm_err_fread(&frag_gap_part[j], sizeof(int), 1, fp[j]);
+				s->n_reg[k] += n_reg_part[j];
+				if (rep_len < rep_len_part[j])
+					rep_len = rep_len_part[j];
+			}
+			s->reg[k] = CALLOC(mm_reg1_t, s->n_reg[k]);
+			for (j = 0, l = 0; j < s->p->n_parts; ++j) {
+				for (t = 0; t < n_reg_part[j]; ++t, ++l) {
+					mm_reg1_t *r = &s->reg[k][l];
+					uint32_t capacity;
+					mm_err_fread(r, sizeof(mm_reg1_t), 1, fp[j]);
+					r->rid += s->p->rid_shift[j];
+					if (opt->flag & MM_F_CIGAR) {
+						mm_err_fread(&capacity, 4, 1, fp[j]);
+						r->p = (mm_extra_t*)calloc(capacity, 4);
+						r->p->capacity = capacity;
+						mm_err_fread(r->p, r->p->capacity, 4, fp[j]);
+					}
+				}
+			}
+			mm_hit_sort(km, &s->n_reg[k], s->reg[k]);
+			mm_set_parent(km, opt->mask_level, s->n_reg[k], s->reg[k], opt->a * 2 + opt->b);
+			if (!(opt->flag & MM_F_ALL_CHAINS)) {
+				mm_select_sub(km, opt->pri_ratio, s->p->mi->k*2, opt->best_n, &s->n_reg[k], s->reg[k]);
+				mm_set_sam_pri(s->n_reg[k], s->reg[k]);
+			}
+			mm_set_mapq(km, s->n_reg[k], s->reg[k], opt->min_chain_score, opt->a, rep_len, !!(opt->flag & MM_F_SR));
+		}
+		if (s->n_seg[f] == 2 && opt->pe_ori >= 0 && (opt->flag&MM_F_CIGAR))
+			mm_pair(km, frag_gap_part[0], opt->pe_bonus, opt->a * 2 + opt->b, opt->a, qlens, &s->n_reg[k0], &s->reg[k0]);
+	}
+	free(qlens);
+	km_destroy(km);
+}
+
 static void *worker_pipeline(void *shared, int step, void *in)
 {
 	int i, j, k;
@@ -459,9 +530,11 @@ static void *worker_pipeline(void *shared, int step, void *in)
 			s->buf = (mm_tbuf_t**)calloc(p->n_threads, sizeof(mm_tbuf_t*));
 			for (i = 0; i < p->n_threads; ++i)
 				s->buf[i] = mm_tbuf_init();
-			s->n_reg = (int*)calloc(3 * s->n_seq, sizeof(int));
-			s->seg_off = s->n_reg + s->n_seq; // seg_off and n_seg are allocated together with n_reg
+			s->n_reg = (int*)calloc(5 * s->n_seq, sizeof(int));
+			s->seg_off = s->n_reg + s->n_seq; // seg_off, n_seg, rep_len and frag_gap are allocated together with n_reg
 			s->n_seg = s->seg_off + s->n_seq;
+			s->rep_len = s->n_seg + s->n_seq;
+			s->frag_gap = s->rep_len + s->n_seq;
 			s->reg = (mm_reg1_t**)calloc(s->n_seq, sizeof(mm_reg1_t*));
 			for (i = 1, j = 0; i <= s->n_seq; ++i)
 				if (i == s->n_seq || !frag_mode || !mm_qname_same(s->seq[i-1].name, s->seq[i].name)) {
@@ -472,7 +545,8 @@ static void *worker_pipeline(void *shared, int step, void *in)
 			return s;
 		} else free(s);
     } else if (step == 1) { // step 1: map
-		kt_for(p->n_threads, worker_for, in, ((step_t*)in)->n_frag);
+		if (p->n_parts > 0) merge_hits((step_t*)in);
+		else kt_for(p->n_threads, worker_for, in, ((step_t*)in)->n_frag);
 		return in;
     } else if (step == 2) { // step 2: output
 		void *km = 0;
@@ -485,19 +559,35 @@ static void *worker_pipeline(void *shared, int step, void *in)
 			int seg_st = s->seg_off[k], seg_en = s->seg_off[k] + s->n_seg[k];
 			for (i = seg_st; i < seg_en; ++i) {
 				mm_bseq1_t *t = &s->seq[i];
-				for (j = 0; j < s->n_reg[i]; ++j) {
-					mm_reg1_t *r = &s->reg[i][j];
-					assert(!r->sam_pri || r->id == r->parent);
-					if ((p->opt->flag & MM_F_NO_PRINT_2ND) && r->id != r->parent)
-						continue;
+				if (p->opt->split_prefix && p->n_parts == 0) { // then write to temporary files
+					mm_err_fwrite(&s->n_reg[i],    sizeof(int), 1, p->fp_split);
+					mm_err_fwrite(&s->rep_len[i],  sizeof(int), 1, p->fp_split);
+					mm_err_fwrite(&s->frag_gap[i], sizeof(int), 1, p->fp_split);
+					for (j = 0; j < s->n_reg[i]; ++j) {
+						mm_reg1_t *r = &s->reg[i][j];
+						mm_err_fwrite(r, sizeof(mm_reg1_t), 1, p->fp_split);
+						if (p->opt->flag & MM_F_CIGAR) {
+							mm_err_fwrite(&r->p->capacity, 4, 1, p->fp_split);
+							mm_err_fwrite(r->p, r->p->capacity, 4, p->fp_split);
+						}
+					}
+				} else if (s->n_reg[i] > 0) { // the query has at least one hit
+					for (j = 0; j < s->n_reg[i]; ++j) {
+						mm_reg1_t *r = &s->reg[i][j];
+						assert(!r->sam_pri || r->id == r->parent);
+						if ((p->opt->flag & MM_F_NO_PRINT_2ND) && r->id != r->parent)
+							continue;
+						if (p->opt->flag & MM_F_OUT_SAM)
+							mm_write_sam2(&p->str, mi, t, i - seg_st, j, s->n_seg[k], &s->n_reg[seg_st], (const mm_reg1_t*const*)&s->reg[seg_st], km, p->opt->flag);
+						else
+							mm_write_paf(&p->str, mi, t, r, km, p->opt->flag);
+						mm_err_puts(p->str.s);
+					}
+				} else if (p->opt->flag & (MM_F_OUT_SAM|MM_F_PAF_NO_HIT)) { // output an empty hit, if requested
 					if (p->opt->flag & MM_F_OUT_SAM)
-						mm_write_sam2(&p->str, mi, t, i - seg_st, j, s->n_seg[k], &s->n_reg[seg_st], (const mm_reg1_t*const*)&s->reg[seg_st], km, p->opt->flag);
+						mm_write_sam2(&p->str, mi, t, i - seg_st, -1, s->n_seg[k], &s->n_reg[seg_st], (const mm_reg1_t*const*)&s->reg[seg_st], km, p->opt->flag);
 					else
-						mm_write_paf(&p->str, mi, t, r, km, p->opt->flag);
-					mm_err_puts(p->str.s);
-				}
-				if (s->n_reg[i] == 0 && (p->opt->flag & MM_F_OUT_SAM)) { // write an unmapped record
-					mm_write_sam2(&p->str, mi, t, i - seg_st, -1, s->n_seg[k], &s->n_reg[seg_st], (const mm_reg1_t*const*)&s->reg[seg_st], km, p->opt->flag);
+						mm_write_paf(&p->str, mi, t, 0, 0, p->opt->flag);
 					mm_err_puts(p->str.s);
 				}
 			}
@@ -508,7 +598,7 @@ static void *worker_pipeline(void *shared, int step, void *in)
 				if (s->seq[i].qual) free(s->seq[i].qual);
 			}
 		}
-		free(s->reg); free(s->n_reg); free(s->seq); // seg_off and n_seg were allocated with reg; no memory leak here
+		free(s->reg); free(s->n_reg); free(s->seq); // seg_off, n_seg, rep_len and frag_gap were allocated with reg; no memory leak here
 		km_destroy(km);
 		if (mm_verbose >= 3)
 			fprintf(stderr, "[M::%s::%.3f*%.2f] mapped %d sequences\n", __func__, realtime() - mm_realtime0, cputime() / (realtime() - mm_realtime0), s->n_seq);
@@ -517,32 +607,44 @@ static void *worker_pipeline(void *shared, int step, void *in)
     return 0;
 }
 
+static mm_bseq_file_t **open_bseqs(int n, const char **fn)
+{
+	mm_bseq_file_t **fp;
+	int i, j;
+	fp = (mm_bseq_file_t**)calloc(n, sizeof(mm_bseq_file_t*));
+	for (i = 0; i < n; ++i) {
+		if ((fp[i] = mm_bseq_open(fn[i])) == 0) {
+			if (mm_verbose >= 1)
+				fprintf(stderr, "ERROR: failed to open file '%s'\n", fn[i]);
+			for (j = 0; j < i; ++j)
+				mm_bseq_close(fp[j]);
+			free(fp);
+			return 0;
+		}
+	}
+	return fp;
+}
+
 int mm_map_file_frag(const mm_idx_t *idx, int n_segs, const char **fn, const mm_mapopt_t *opt, int n_threads)
 {
-	int i, j, pl_threads;
+	int i, pl_threads;
 	pipeline_t pl;
 	if (n_segs < 1) return -1;
 	memset(&pl, 0, sizeof(pipeline_t));
 	pl.n_fp = n_segs;
-	pl.fp = (mm_bseq_file_t**)calloc(n_segs, sizeof(mm_bseq_file_t*));
-	for (i = 0; i < n_segs; ++i) {
-		pl.fp[i] = mm_bseq_open(fn[i]);
-		if (pl.fp[i] == 0) {
-			if (mm_verbose >= 1)
-				fprintf(stderr, "ERROR: failed to open file '%s'\n", fn[i]);
-			for (j = 0; j < i; ++j)
-				mm_bseq_close(pl.fp[j]);
-			free(pl.fp);
-			return -1;
-		}
-	}
+	pl.fp = open_bseqs(pl.n_fp, fn);
+	if (pl.fp == 0) return -1;
 	pl.opt = opt, pl.mi = idx;
 	pl.n_threads = n_threads > 1? n_threads : 1;
 	pl.mini_batch_size = opt->mini_batch_size;
+	if (opt->split_prefix)
+		pl.fp_split = mm_split_init(opt->split_prefix, idx);
 	pl_threads = n_threads == 1? 1 : (opt->flag&MM_F_2_IO_THREADS)? 3 : 2;
 	kt_pipeline(pl_threads, worker_pipeline, &pl, 3);
+
 	free(pl.str.s);
-	for (i = 0; i < n_segs; ++i)
+	if (pl.fp_split) fclose(pl.fp_split);
+	for (i = 0; i < pl.n_fp; ++i)
 		mm_bseq_close(pl.fp[i]);
 	free(pl.fp);
 	return 0;
@@ -552,3 +654,48 @@ int mm_map_file(const mm_idx_t *idx, const char *fn, const mm_mapopt_t *opt, int
 {
 	return mm_map_file_frag(idx, 1, &fn, opt, n_threads);
 }
+
+int mm_split_merge(int n_segs, const char **fn, const mm_mapopt_t *opt, int n_split_idx)
+{
+	int i;
+	pipeline_t pl;
+	mm_idx_t *mi;
+	if (n_segs < 1 || n_split_idx < 1) return -1;
+	memset(&pl, 0, sizeof(pipeline_t));
+	pl.n_fp = n_segs;
+	pl.fp = open_bseqs(pl.n_fp, fn);
+	if (pl.fp == 0) return -1;
+	pl.opt = opt;
+	pl.mini_batch_size = opt->mini_batch_size;
+
+	pl.n_parts = n_split_idx;
+	pl.fp_parts  = CALLOC(FILE*, pl.n_parts);
+	pl.rid_shift = CALLOC(uint32_t, pl.n_parts);
+	pl.mi = mi = mm_split_merge_prep(opt->split_prefix, n_split_idx, pl.fp_parts, pl.rid_shift);
+	if (pl.mi == 0) {
+		free(pl.fp_parts);
+		free(pl.rid_shift);
+		return -1;
+	}
+	for (i = n_split_idx - 1; i > 0; --i)
+		pl.rid_shift[i] = pl.rid_shift[i - 1];
+	for (pl.rid_shift[0] = 0, i = 1; i < n_split_idx; ++i)
+		pl.rid_shift[i] += pl.rid_shift[i - 1];
+	if (opt->flag & MM_F_OUT_SAM)
+		for (i = 0; i < pl.mi->n_seq; ++i)
+			printf("@SQ\tSN:%s\tLN:%d\n", pl.mi->seq[i].name, pl.mi->seq[i].len);
+
+	kt_pipeline(2, worker_pipeline, &pl, 3);
+
+	free(pl.str.s);
+	mm_idx_destroy(mi);
+	free(pl.rid_shift);
+	for (i = 0; i < n_split_idx; ++i)
+		fclose(pl.fp_parts[i]);
+	free(pl.fp_parts);
+	for (i = 0; i < pl.n_fp; ++i)
+		mm_bseq_close(pl.fp[i]);
+	free(pl.fp);
+	mm_split_rm_tmp(opt->split_prefix, n_split_idx);
+	return 0;
+}

minimap.h

@@ -32,6 +32,7 @@
 #define MM_F_OUT_MD        0x1000000
 #define MM_F_COPY_COMMENT  0x2000000
 #define MM_F_EQX           0x4000000 // use =/X instead of M
+#define MM_F_PAF_NO_HIT    0x8000000 // output unmapped reads to PAF
 
 #define MM_I_HPC          0x1
 #define MM_I_NO_SEQ       0x2
@@ -59,6 +60,7 @@ typedef struct {
 typedef struct {
 	int32_t b, w, k, flag;
 	uint32_t n_seq;            // number of reference sequences
+	int32_t index;
 	mm_idx_seq_t *seq;         // sequence name, length and offset
 	uint32_t *S;               // 4-bit packed sequence
 	struct mm_idx_bucket_s *B; // index (hidden)
@@ -135,6 +137,8 @@ typedef struct {
 	int32_t mid_occ;     // ignore seeds with occurrences above this threshold
 	int32_t max_occ;
 	int mini_batch_size; // size of a batch of query bases to process in parallel
+
+	const char *split_prefix;
 } mm_mapopt_t;
 
 // index reader

minimap2.1

@@ -239,7 +239,7 @@ Disable the long gap patching heuristic. When this option is applied, the
 maximum alignment gap is mostly controlled by
 .BR -r .
 .TP
-.B --lj-min-ratio \ FLOAT
+.BI --lj-min-ratio \ FLOAT
 Fraction of query sequence length required to bridge a long gap [0.5]. A
 smaller value helps to recover longer gaps, at the cost of more false gaps.
 .TP
@@ -252,6 +252,9 @@ applies a second round of chaining with a higher minimizer occurrence threshold
 if no good chain is found. In addition, minimap2 attempts to patch gaps between
 seeds with ungapped alignment.
 .TP
+.BI --split-prefix \ STR
+Prefix to create temporary files. Typically used for a multi-part index.
+.TP
 .BR --frag = no | yes
 Whether to enable the fragment mode [no]
 .TP

misc.c

@@ -131,6 +131,26 @@ void mm_err_puts(const char *str)
 	}
 }
 
+void mm_err_fwrite(const void *p, size_t size, size_t nitems, FILE *fp)
+{
+	int ret;
+	ret = fwrite(p, size, nitems, fp);
+	if (ret == EOF) {
+		fprintf(stderr, "[ERROR] failed to write data\n");
+		exit(EXIT_FAILURE);
+	}
+}
+
+void mm_err_fread(void *p, size_t size, size_t nitems, FILE *fp)
+{
+	int ret;
+	ret = fread(p, size, nitems, fp);
+	if (ret == EOF) {
+		fprintf(stderr, "[ERROR] failed to read data\n");
+		exit(EXIT_FAILURE);
+	}
+}
+
 #include "ksort.h"
 
 #define sort_key_128x(a) ((a).x)

misc/paftools.js

@@ -340,12 +340,13 @@ function paf_liftover(args)
 function paf_call(args)
 {
 	var re_cs = /([:=*+-])(\d+|[A-Za-z]+)/g, re_tag = /\t(\S\S:[AZif]):(\S+)/g;
-	var c, min_cov_len = 10000, min_var_len = 50000, gap_thres = 50, min_mapq = 5;
+	var c, min_cov_len = 10000, min_var_len = 50000, gap_thres = 50, gap_thres_long = 1000, min_mapq = 5;
 	var fa_tmp = null, fa, fa_lens, is_vcf = false;
 	while ((c = getopt(args, "l:L:g:q:B:f:")) != null) {
 		if (c == 'l') min_cov_len = parseInt(getopt.arg);
 		else if (c == 'L') min_var_len = parseInt(getopt.arg);
 		else if (c == 'g') gap_thres = parseInt(getopt.arg);
+		else if (c == 'G') gap_thres_long = parseInt(getopt.arg);
 		else if (c == 'q') min_mapq = parseInt(getopt.arg);
 		else if (c == 'f') fa_tmp = fasta_read(getopt.arg, fa_lens);
 	}
@@ -364,7 +365,7 @@ function paf_call(args)
 
 	var file = args[getopt.ind] == '-'? new File() : new File(args[getopt.ind]);
 	var buf = new Bytes();
-	var tot_len = 0, n_sub = [0, 0, 0], n_ins = [0, 0, 0, 0], n_del = [0, 0, 0, 0];
+	var tot_len = 0, n_sub = [0, 0, 0], n_ins = [0, 0, 0, 0, 0], n_del = [0, 0, 0, 0, 0];
 
 	function print_vcf(o, fa)
 	{
@@ -396,13 +397,15 @@ function paf_call(args)
 			if (l == 1) ++n_ins[0];
 			else if (l == 2) ++n_ins[1];
 			else if (l < gap_thres) ++n_ins[2];
-			else ++n_ins[3];
+			else if (l < gap_thres_long) ++n_ins[3];
+			else ++n_ins[4];
 		} else if (o[6] == '-') { // deletion
 			var l = o[5].length;
 			if (l == 1) ++n_del[0];
 			else if (l == 2) ++n_del[1];
 			else if (l < gap_thres) ++n_del[2];
-			else ++n_del[3];
+			else if (l < gap_thres_long) ++n_del[3];
+			else ++n_del[4];
 		} else {
 			++n_sub[0];
 			var s = (o[5] + o[6]).toLowerCase();
@@ -547,14 +550,196 @@ function paf_call(args)
 	warn(n_ins[1] + " 2bp insertions");
 	warn(n_del[2] + " [3,"+gap_thres+") deletions");
 	warn(n_ins[2] + " [3,"+gap_thres+") insertions");
-	warn(n_del[3] + " >="+gap_thres+" deletions");
-	warn(n_ins[3] + " >="+gap_thres+" insertions");
+	warn(n_del[3] + " ["+gap_thres+","+gap_thres_long+") deletions");
+	warn(n_ins[3] + " ["+gap_thres+","+gap_thres_long+") insertions");
+	warn(n_del[4] + " >=" + gap_thres_long + " deletions");
+	warn(n_ins[4] + " >=" + gap_thres_long + " insertions");
 
 	buf.destroy();
 	file.close();
 	if (fa != null) fasta_free(fa);
 }
 
+function paf_asmstat(args)
+{
+	var c, min_seg_len = 10000, max_diff = 0.01;
+	while ((c = getopt(args, "l:d:")) != null) {
+		if (c == 'l') min_seg_len = parseInt(getopt.arg);
+		else if (c == 'd') max_diff = parseFloat(getopt.arg);
+	}
+	if (getopt.ind == args.length) {
+		print("Usage: paftools.js asmstat [options] <ref.fa.fai> <asm1.paf> [...]");
+		print("Options:");
+		print("  -l INT     min alignment block length [" + min_seg_len + "]");
+		print("  -d FLOAT   max gap-compressed sequence divergence [" + max_diff + "]");
+		exit(1);
+	}
+
+	var file, buf = new Bytes();
+
+	var ref_len = 0;
+	file = new File(args[getopt.ind]);
+	while (file.readline(buf) >= 0) {
+		var t = buf.toString().split("\t");
+		ref_len += parseInt(t[1]);
+	}
+	file.close();
+
+	function process_query(qblocks, qblock_len, bp) {
+		qblocks.sort(function(a,b) { return a[0]-b[0]; });
+		var last_k = null, last_blen = null, st = -1, en = -1, qcov = 0;
+		for (var k = 0; k < qblocks.length; ++k) {
+			var blen = qblocks[k][1] - qblocks[k][0];
+			if (k > 0 && qblocks[k][0] < qblocks[k-1][1]) {
+				if (qblocks[k][1] < qblocks[k-1][1]) continue;
+				blen = qblocks[k][1] - qblocks[k-1][1];
+			}
+			qblock_len.push(blen);
+			if (qblocks[k][0] > en) {
+				qcov += en - st;
+				st = qblocks[k][0];
+				en = qblocks[k][1];
+			} else en = en > qblocks[k][1]? en : qblocks[k][1];
+			if (last_k != null) {
+				var gap = 1000000000;
+				if (qblocks[k][2] == qblocks[last_k][2] && qblocks[k][3] == qblocks[last_k][3]) { // same chr and strand
+					var g1 = qblocks[k][0] - qblocks[last_k][1];
+					var g2 = qblocks[k][2] == '+'? qblocks[k][4] - qblocks[last_k][5] : qblocks[last_k][4] - qblocks[k][5];
+					gap = g1 > g2? g1 - g2 : g2 - g1;
+				}
+				var min = blen < last_blen? blen : last_blen;
+				var flank = k == 0? min : blen;
+				bp.push([flank, gap]);
+			}
+			last_k = k, last_blen = blen;
+		}
+		qcov += en - st;
+		return qcov;
+	}
+
+	function N50(lens, tot, quantile) {
+		lens.sort(function(a,b) { return b - a; });
+		if (tot == null) {
+			tot = 0;
+			for (var k = 0; k < lens.length; ++k)
+				tot += lens[k];
+		}
+		var sum = 0;
+		for (var k = 0; k < lens.length; ++k) {
+			if (sum <= quantile * tot && sum + lens[k] > quantile * tot)
+				return lens[k];
+			sum += lens[k];
+		}
+	}
+
+	function count_bp(bp, min_blen, min_gap) {
+		var n_bp = 0;
+		for (var k = 0; k < bp.length; ++k)
+			if (bp[k][0] >= min_blen && bp[k][1] >= min_gap)
+				++n_bp;
+		return n_bp;
+	}
+
+	function compute_diff(cigar, NM) {
+		var m, re = /(\d+)([MID])/g;
+		var n_M = 0, n_gapo = 0, n_gaps = 0;
+		while ((m = re.exec(cigar)) != null) {
+			var len = parseInt(m[1]);
+			if (m[2] == 'M') n_M += len;
+			else ++n_gapo, n_gaps += len;
+		}
+		if (NM < n_gaps) throw Error('NM is smaller the number of gaps');
+		return (NM - n_gaps + n_gapo) / (n_M + n_gapo);
+	}
+
+	var labels = ['Length', 'NG50', 'Coverage', 'Qcov', 'NGA50', '#breaks', 'bp(' + min_seg_len + ',0)', 'bp(' + min_seg_len + ',10k)'];
+	var rst = [];
+	for (var i = 0; i < labels.length; ++i)
+		rst[i] = [];
+
+	var n_asm = args.length - (getopt.ind + 1);
+	var header = ["Metric"];
+	for (var i = 0; i < n_asm; ++i) {
+		var n_breaks = 0, qcov = 0;
+		var fn = args[getopt.ind + 1 + i];
+		header.push(fn.replace(/.paf(.gz)?$/, ""));
+		var ref_blocks = [], qblock_len = [], qblocks = [], bp = [];
+		var query = {};
+		var last_qname = null;
+		file = new File(fn);
+		while (file.readline(buf) >= 0) {
+			var m, line = buf.toString();
+			var t = line.split("\t");
+			t[1] = parseInt(t[1]);
+			if (t.length >= 2) query[t[0]] = t[1];
+			if (t.length < 9) continue;
+			if (!/\ttp:A:[PI]/.test(line)) continue;
+			if ((m = /\tcg:Z:(\S+)/.exec(line)) == null) continue;
+			var cigar = m[1];
+			if ((m = /\tNM:i:(\d+)/.exec(line)) == null) continue;
+			var NM = parseInt(m[1]);
+			var diff = compute_diff(cigar, NM);
+			t[2] = parseInt(t[2]);
+			t[3] = parseInt(t[3]);
+			t[7] = parseInt(t[7]);
+			t[8] = parseInt(t[8]);
+			if (t[0] == last_qname) ++n_breaks;
+			if (diff > max_diff) continue;
+			if (t[3] - t[2] < min_seg_len) continue;
+			if (t[0] != last_qname) {
+				if (last_qname != null)
+					qcov += process_query(qblocks, qblock_len, bp);
+				qblocks = [];
+				last_qname = t[0];
+			}
+			ref_blocks.push([t[5], t[7], t[8]]);
+			qblocks.push([t[2], t[3], t[4], t[5], t[7], t[8]]);
+		}
+		if (last_qname != null)
+			qcov += process_query(qblocks, qblock_len, bp);
+		file.close();
+
+		// compute NG50
+		var asm_len = 0, asm_lens = []
+		for (var ctg in query) {
+			asm_len += query[ctg];
+			asm_lens.push(query[ctg]);
+		}
+		rst[0][i] = asm_len;
+		rst[1][i] = N50(asm_lens, ref_len, 0.5);
+
+		// compute coverage
+		var l_cov = 0;
+		ref_blocks.sort(function(a, b) { return a[0] > b[0]? 1 : a[0] < b[0]? -1 : a[1] - b[1]; });
+		var last_ref = null, st = -1, en = -1;
+		for (var j = 0; j < ref_blocks.length; ++j) {
+			if (ref_blocks[j][0] != last_ref || ref_blocks[j][1] > en) {
+				l_cov += en - st;
+				last_ref = ref_blocks[j][0];
+				st = ref_blocks[j][1];
+				en = ref_blocks[j][2];
+			} else en = en > ref_blocks[j][2]? en : ref_blocks[j][2];
+		}
+		l_cov += en - st;
+		rst[2][i] = (100.0 * (l_cov / ref_len)).toFixed(2) + '%';
+		rst[3][i] = (100.0 * (qcov  / asm_len)).toFixed(2) + '%';
+
+		// compute NGA50
+		rst[4][i] = N50(qblock_len, ref_len, 0.5);
+
+		// compute break points
+		rst[5][i] = n_breaks;
+		rst[6][i] = count_bp(bp, 500, 0);
+		rst[7][i] = count_bp(bp, 500, 10000);
+	}
+
+	print(header.join("\t"));
+	for (var i = 0; i < labels.length; ++i)
+		print(labels[i], rst[i].join("\t"));
+
+	buf.destroy();
+}
+
 function paf_stat(args)
 {
 	var c, gap_out_len = null;
@@ -2001,6 +2186,7 @@ function main(args)
 		print("  gff2bed    convert GTF/GFF3 to BED12");
 		print("");
 		print("  stat       collect basic mapping information in PAF/SAM");
+		print("  asmstat    collect basic assembly information");
 		print("  liftover   simplistic liftOver");
 		print("  call       call variants from asm-to-ref alignment with the cs tag");
 		print("  bedcov     compute the number of bases covered");
@@ -2021,6 +2207,7 @@ function main(args)
 	else if (cmd == 'splice2bed') paf_splice2bed(args);
 	else if (cmd == 'gff2bed') paf_gff2bed(args);
 	else if (cmd == 'stat') paf_stat(args);
+	else if (cmd == 'asmstat') paf_asmstat(args);
 	else if (cmd == 'liftover' || cmd == 'liftOver') paf_liftover(args);
 	else if (cmd == 'call') paf_call(args);
 	else if (cmd == 'mapeval') paf_mapeval(args);

mmpriv.h

@@ -28,6 +28,9 @@
 #define mm_seq4_set(s, i, c) ((s)[(i)>>3] |= (uint32_t)(c) << (((i)&7)<<2))
 #define mm_seq4_get(s, i)    ((s)[(i)>>3] >> (((i)&7)<<2) & 0xf)
 
+#define MALLOC(type, len) ((type*)malloc((len) * sizeof(type)))
+#define CALLOC(type, len) ((type*)calloc((len), sizeof(type)))
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -77,7 +80,7 @@ void mm_select_sub(void *km, float pri_ratio, int min_diff, int best_n, int *n_,
 void mm_select_sub_multi(void *km, float pri_ratio, float pri1, float pri2, int max_gap_ref, int min_diff, int best_n, int n_segs, const int *qlens, int *n_, mm_reg1_t *r);
 void mm_filter_regs(const mm_mapopt_t *opt, int qlen, int *n_regs, mm_reg1_t *regs);
 void mm_join_long(void *km, const mm_mapopt_t *opt, int qlen, int *n_regs, mm_reg1_t *regs, mm128_t *a);
-void mm_hit_sort_by_dp(void *km, int *n_regs, mm_reg1_t *r);
+void mm_hit_sort(void *km, int *n_regs, mm_reg1_t *r);
 void mm_set_mapq(void *km, int n_regs, mm_reg1_t *regs, int min_chain_sc, int match_sc, int rep_len, int is_sr);
 
 void mm_est_err(const mm_idx_t *mi, int qlen, int n_regs, mm_reg1_t *regs, const mm128_t *a, int32_t n, const uint64_t *mini_pos);
@@ -86,7 +89,14 @@ mm_seg_t *mm_seg_gen(void *km, uint32_t hash, int n_segs, const int *qlens, int
 void mm_seg_free(void *km, int n_segs, mm_seg_t *segs);
 void mm_pair(void *km, int max_gap_ref, int dp_bonus, int sub_diff, int match_sc, const int *qlens, int *n_regs, mm_reg1_t **regs);
 
+FILE *mm_split_init(const char *prefix, const mm_idx_t *mi);
+mm_idx_t *mm_split_merge_prep(const char *prefix, int n_splits, FILE **fp, uint32_t *n_seq_part);
+int mm_split_merge(int n_segs, const char **fn, const mm_mapopt_t *opt, int n_split_idx);
+void mm_split_rm_tmp(const char *prefix, int n_splits);
+
 void mm_err_puts(const char *str);
+void mm_err_fwrite(const void *p, size_t size, size_t nitems, FILE *fp);
+void mm_err_fread(void *p, size_t size, size_t nitems, FILE *fp);
 
 #ifdef __cplusplus
 }

options.c

@@ -49,7 +49,7 @@ void mm_mapopt_init(mm_mapopt_t *opt)
 
 void mm_mapopt_update(mm_mapopt_t *opt, const mm_idx_t *mi)
 {
-	if ((opt->flag & MM_F_SPLICE_FOR) && (opt->flag & MM_F_SPLICE_REV))
+	if ((opt->flag & MM_F_SPLICE_FOR) || (opt->flag & MM_F_SPLICE_REV))
 		opt->flag |= MM_F_SPLICE;
 	if (opt->mid_occ <= 0)
 		opt->mid_occ = mm_idx_cal_max_occ(mi, opt->mid_occ_frac);

python/cmappy.h

@@ -73,13 +73,17 @@ static inline void mm_reset_timer(void)
 extern unsigned char seq_comp_table[256];
 static inline mm_reg1_t *mm_map_aux(const mm_idx_t *mi, const char *seq1, const char *seq2, int *n_regs, mm_tbuf_t *b, const mm_mapopt_t *opt)
 {
+	mm_reg1_t *r;
+
+	Py_BEGIN_ALLOW_THREADS
 	if (seq2 == 0) {
-		return mm_map(mi, strlen(seq1), seq1, n_regs, b, opt, NULL);
+		r = mm_map(mi, strlen(seq1), seq1, n_regs, b, opt, NULL);
 	} else {
 		int _n_regs[2];
 		mm_reg1_t *regs[2];
 		char *seq[2];
 		int i, len[2];
+
 		len[0] = strlen(seq1);
 		len[1] = strlen(seq2);
 		seq[0] = (char*)seq1;
@@ -97,8 +101,11 @@ static inline mm_reg1_t *mm_map_aux(const mm_idx_t *mi, const char *seq1, const
 		regs[0] = (mm_reg1_t*)realloc(regs[0], sizeof(mm_reg1_t) * (*n_regs));
 		memcpy(&regs[0][_n_regs[0]], regs[1], _n_regs[1] * sizeof(mm_reg1_t));
 		free(regs[1]);
-		return regs[0];
+		r = regs[0];
 	}
+	Py_END_ALLOW_THREADS
+
+	return r;
 }
 
 static inline char *mappy_revcomp(int len, const uint8_t *seq)

splitidx.c

@@ -0,0 +1,80 @@
+#include <string.h>
+#include <assert.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include "mmpriv.h"
+
+FILE *mm_split_init(const char *prefix, const mm_idx_t *mi)
+{
+	char *fn;
+	FILE *fp;
+	uint32_t i, k = mi->k;
+	fn = (char*)calloc(strlen(prefix) + 10, 1);
+	sprintf(fn, "%s.%.4d.tmp", prefix, mi->index);
+	fp = fopen(fn, "wb");
+	assert(fp);
+	mm_err_fwrite(&k, 4, 1, fp);
+	mm_err_fwrite(&mi->n_seq, 4, 1, fp);
+	for (i = 0; i < mi->n_seq; ++i) {
+		uint8_t l;
+		l = strlen(mi->seq[i].name);
+		mm_err_fwrite(&l, 1, 1, fp);
+		mm_err_fwrite(mi->seq[i].name, 1, l, fp);
+		mm_err_fwrite(&mi->seq[i].len, 4, 1, fp);
+	}
+	free(fn);
+	return fp;
+}
+
+mm_idx_t *mm_split_merge_prep(const char *prefix, int n_splits, FILE **fp, uint32_t *n_seq_part)
+{
+	mm_idx_t *mi = 0;
+	char *fn;
+	int i, j;
+
+	if (n_splits < 1) return 0;
+	fn = CALLOC(char, strlen(prefix) + 10);
+	for (i = 0; i < n_splits; ++i) {
+		sprintf(fn, "%s.%.4d.tmp", prefix, i);
+		if ((fp[i] = fopen(fn, "rb")) == 0) {
+			if (mm_verbose >= 1)
+				fprintf(stderr, "ERROR: failed to open temporary file '%s'\n", fn);
+			for (j = 0; j < i; ++j)
+				fclose(fp[j]);
+			free(fn);
+			return 0;
+		}
+	}
+	free(fn);
+
+	mi = CALLOC(mm_idx_t, 1);
+	for (i = 0; i < n_splits; ++i) {
+		mm_err_fread(&mi->k, 4, 1, fp[i]); // TODO: check if k is all the same
+		mm_err_fread(&n_seq_part[i], 4, 1, fp[i]);
+		mi->n_seq += n_seq_part[i];
+	}
+	mi->seq = CALLOC(mm_idx_seq_t, mi->n_seq);
+	for (i = j = 0; i < n_splits; ++i) {
+		uint32_t k;
+		for (k = 0; k < n_seq_part[i]; ++k, ++j) {
+			uint8_t l;
+			mm_err_fread(&l, 1, 1, fp[i]);
+			mi->seq[j].name = (char*)calloc(l + 1, 1);
+			mm_err_fread(mi->seq[j].name, 1, l, fp[i]);
+			mm_err_fread(&mi->seq[j].len, 4, 1, fp[i]);
+		}
+	}
+	return mi;
+}
+
+void mm_split_rm_tmp(const char *prefix, int n_splits)
+{
+	int i;
+	char *fn;
+	fn = CALLOC(char, strlen(prefix) + 10);
+	for (i = 0; i < n_splits; ++i) {
+		sprintf(fn, "%s.%.4d.tmp", prefix, i);
+		remove(fn);
+	}
+	free(fn);
+}