11 static int g_is_by_qname = 0;
13 static inline int strnum_cmp(const char *a, const char *b)
16 pa = (char*)a; pb = (char*)b;
18 if (isdigit(*pa) && isdigit(*pb)) {
20 ai = strtol(pa, &pa, 10);
21 bi = strtol(pb, &pb, 10);
22 if (ai != bi) return ai<bi? -1 : ai>bi? 1 : 0;
24 if (*pa != *pb) break;
29 return (pa-a) < (pb-b)? -1 : (pa-a) > (pb-b)? 1 : 0;
30 return *pa<*pb? -1 : *pa>*pb? 1 : 0;
33 #define HEAP_EMPTY 0xffffffffffffffffull
41 #define __pos_cmp(a, b) ((a).pos > (b).pos || ((a).pos == (b).pos && ((a).i > (b).i || ((a).i == (b).i && (a).idx > (b).idx))))
43 static inline int heap_lt(const heap1_t a, const heap1_t b)
47 if (a.b == 0 || b.b == 0) return a.b == 0? 1 : 0;
48 t = strnum_cmp(bam1_qname(a.b), bam1_qname(b.b));
49 return (t > 0 || (t == 0 && __pos_cmp(a, b)));
50 } else return __pos_cmp(a, b);
53 KSORT_INIT(heap, heap1_t, heap_lt)
55 static void swap_header_targets(bam_header_t *h1, bam_header_t *h2)
58 t.n_targets = h1->n_targets, h1->n_targets = h2->n_targets, h2->n_targets = t.n_targets;
59 t.target_name = h1->target_name, h1->target_name = h2->target_name, h2->target_name = t.target_name;
60 t.target_len = h1->target_len, h1->target_len = h2->target_len, h2->target_len = t.target_len;
63 static void swap_header_text(bam_header_t *h1, bam_header_t *h2)
67 tempi = h1->l_text, h1->l_text = h2->l_text, h2->l_text = tempi;
68 temps = h1->text, h1->text = h2->text, h2->text = temps;
72 #define MERGE_UNCOMP 2
75 @abstract Merge multiple sorted BAM.
76 @param is_by_qname whether to sort by query name
77 @param out output BAM file name
78 @param headers name of SAM file from which to copy '@' header lines,
79 or NULL to copy them from the first file to be merged
80 @param n number of files to be merged
81 @param fn names of files to be merged
83 @discussion Padding information may NOT correctly maintained. This
84 function is NOT thread safe.
86 int bam_merge_core(int by_qname, const char *out, const char *headers, int n, char * const *fn,
87 int flag, const char *reg)
91 bam_header_t *hout = 0;
92 bam_header_t *hheaders = NULL;
93 int i, j, *RG_len = 0;
99 tamFile fpheaders = sam_open(headers);
100 if (fpheaders == 0) {
101 const char *message = strerror(errno);
102 fprintf(stderr, "[bam_merge_core] cannot open '%s': %s\n", headers, message);
105 hheaders = sam_header_read(fpheaders);
106 sam_close(fpheaders);
109 g_is_by_qname = by_qname;
110 fp = (bamFile*)calloc(n, sizeof(bamFile));
111 heap = (heap1_t*)calloc(n, sizeof(heap1_t));
112 iter = (bam_iter_t*)calloc(n, sizeof(bam_iter_t));
114 if (flag & MERGE_RG) {
115 RG = (char**)calloc(n, sizeof(void*));
116 RG_len = (int*)calloc(n, sizeof(int));
117 for (i = 0; i != n; ++i) {
118 int l = strlen(fn[i]);
119 const char *s = fn[i];
120 if (l > 4 && strcmp(s + l - 4, ".bam") == 0) l -= 4;
121 for (j = l - 1; j >= 0; --j) if (s[j] == '/') break;
123 RG[i] = calloc(l + 1, 1);
125 strncpy(RG[i], s + j, l);
129 for (i = 0; i != n; ++i) {
131 fp[i] = bam_open(fn[i], "r");
134 fprintf(stderr, "[bam_merge_core] fail to open file %s\n", fn[i]);
135 for (j = 0; j < i; ++j) bam_close(fp[j]);
136 free(fp); free(heap);
137 // FIXME: possible memory leak
140 hin = bam_header_read(fp[i]);
141 if (i == 0) { // the first BAM
143 } else { // validate multiple baf
144 int min_n_targets = hout->n_targets;
145 if (hin->n_targets < min_n_targets) min_n_targets = hin->n_targets;
147 for (j = 0; j < min_n_targets; ++j)
148 if (strcmp(hout->target_name[j], hin->target_name[j]) != 0) {
149 fprintf(stderr, "[bam_merge_core] different target sequence name: '%s' != '%s' in file '%s'\n",
150 hout->target_name[j], hin->target_name[j], fn[i]);
154 // If this input file has additional target reference sequences,
155 // add them to the headers to be output
156 if (hin->n_targets > hout->n_targets) {
157 swap_header_targets(hout, hin);
158 // FIXME Possibly we should also create @SQ text headers
159 // for the newly added reference sequences
162 bam_header_destroy(hin);
167 // If the text headers to be swapped in include any @SQ headers,
168 // check that they are consistent with the existing binary list
169 // of reference information.
170 if (hheaders->n_targets > 0) {
171 if (hout->n_targets != hheaders->n_targets) {
172 fprintf(stderr, "[bam_merge_core] number of @SQ headers in '%s' differs from number of target sequences\n", headers);
175 for (j = 0; j < hout->n_targets; ++j)
176 if (strcmp(hout->target_name[j], hheaders->target_name[j]) != 0) {
177 fprintf(stderr, "[bam_merge_core] @SQ header '%s' in '%s' differs from target sequence\n", hheaders->target_name[j], headers);
182 swap_header_text(hout, hheaders);
183 bam_header_destroy(hheaders);
188 if (bam_parse_region(hout, reg, &tid, &beg, &end) < 0) {
189 fprintf(stderr, "[%s] Malformated region string or undefined reference name\n", __func__);
192 for (i = 0; i < n; ++i) {
194 idx = bam_index_load(fn[i]);
195 iter[i] = bam_iter_query(idx, tid, beg, end);
196 bam_index_destroy(idx);
200 for (i = 0; i < n; ++i) {
201 heap1_t *h = heap + i;
203 h->b = (bam1_t*)calloc(1, sizeof(bam1_t));
204 if (bam_iter_read(fp[i], iter[i], h->b) >= 0) {
205 h->pos = ((uint64_t)h->b->core.tid<<32) | (uint32_t)h->b->core.pos<<1 | bam1_strand(h->b);
208 else h->pos = HEAP_EMPTY;
210 if (flag & MERGE_UNCOMP) {
211 fpout = strcmp(out, "-")? bam_open(out, "wu") : bam_dopen(fileno(stdout), "wu");
213 fpout = strcmp(out, "-")? bam_open(out, "w") : bam_dopen(fileno(stdout), "w");
216 fprintf(stderr, "[%s] fail to create the output file.\n", __func__);
219 bam_header_write(fpout, hout);
220 bam_header_destroy(hout);
222 ks_heapmake(heap, n, heap);
223 while (heap->pos != HEAP_EMPTY) {
225 if ((flag & MERGE_RG) && bam_aux_get(b, "RG") == 0)
226 bam_aux_append(b, "RG", 'Z', RG_len[heap->i] + 1, (uint8_t*)RG[heap->i]);
227 bam_write1_core(fpout, &b->core, b->data_len, b->data);
228 if ((j = bam_iter_read(fp[heap->i], iter[heap->i], b)) >= 0) {
229 heap->pos = ((uint64_t)b->core.tid<<32) | (uint32_t)b->core.pos<<1 | bam1_strand(b);
231 } else if (j == -1) {
232 heap->pos = HEAP_EMPTY;
233 free(heap->b->data); free(heap->b);
235 } else fprintf(stderr, "[bam_merge_core] '%s' is truncated. Continue anyway.\n", fn[heap->i]);
236 ks_heapadjust(heap, 0, n, heap);
239 if (flag & MERGE_RG) {
240 for (i = 0; i != n; ++i) free(RG[i]);
241 free(RG); free(RG_len);
243 for (i = 0; i != n; ++i) {
244 bam_iter_destroy(iter[i]);
248 free(fp); free(heap); free(iter);
252 int bam_merge(int argc, char *argv[])
254 int c, is_by_qname = 0, flag = 0, ret = 0;
255 char *fn_headers = NULL, *reg = 0;
257 while ((c = getopt(argc, argv, "h:nruR:")) >= 0) {
259 case 'r': flag |= MERGE_RG; break;
260 case 'h': fn_headers = strdup(optarg); break;
261 case 'n': is_by_qname = 1; break;
262 case 'u': flag |= MERGE_UNCOMP; break;
263 case 'R': reg = strdup(optarg); break;
266 if (optind + 2 >= argc) {
267 fprintf(stderr, "\n");
268 fprintf(stderr, "Usage: samtools merge [-nr] [-h inh.sam] <out.bam> <in1.bam> <in2.bam> [...]\n\n");
269 fprintf(stderr, "Options: -n sort by read names\n");
270 fprintf(stderr, " -r attach RG tag (inferred from file names)\n");
271 fprintf(stderr, " -u uncompressed BAM output\n");
272 fprintf(stderr, " -R STR merge file in the specified region STR [all]\n");
273 fprintf(stderr, " -h FILE copy the header in FILE to <out.bam> [in1.bam]\n\n");
274 fprintf(stderr, "Note: Samtools' merge does not reconstruct the @RG dictionary in the header. Users\n");
275 fprintf(stderr, " must provide the correct header with -h, or uses Picard which properly maintains\n");
276 fprintf(stderr, " the header dictionary in merging.\n\n");
279 if (bam_merge_core(is_by_qname, argv[optind], fn_headers, argc - optind - 1, argv + optind + 1, flag, reg) < 0) ret = 1;
285 typedef bam1_t *bam1_p;
287 static inline int bam1_lt(const bam1_p a, const bam1_p b)
290 int t = strnum_cmp(bam1_qname(a), bam1_qname(b));
291 return (t < 0 || (t == 0 && (((uint64_t)a->core.tid<<32|a->core.pos) < ((uint64_t)b->core.tid<<32|b->core.pos))));
292 } else return (((uint64_t)a->core.tid<<32|a->core.pos) < ((uint64_t)b->core.tid<<32|b->core.pos));
294 KSORT_INIT(sort, bam1_p, bam1_lt)
296 static void sort_blocks(int n, int k, bam1_p *buf, const char *prefix, const bam_header_t *h, int is_stdout)
301 ks_mergesort(sort, k, buf, 0);
302 name = (char*)calloc(strlen(prefix) + 20, 1);
303 if (n >= 0) sprintf(name, "%s.%.4d.bam", prefix, n);
304 else sprintf(name, "%s.bam", prefix);
305 fp = is_stdout? bam_dopen(fileno(stdout), "w") : bam_open(name, "w");
307 fprintf(stderr, "[sort_blocks] fail to create file %s.\n", name);
309 // FIXME: possible memory leak
313 bam_header_write(fp, h);
314 for (i = 0; i < k; ++i)
315 bam_write1_core(fp, &buf[i]->core, buf[i]->data_len, buf[i]->data);
320 @abstract Sort an unsorted BAM file based on the chromosome order
321 and the leftmost position of an alignment
323 @param is_by_qname whether to sort by query name
324 @param fn name of the file to be sorted
325 @param prefix prefix of the output and the temporary files; upon
326 sucessess, prefix.bam will be written.
327 @param max_mem approxiate maximum memory (very inaccurate)
329 @discussion It may create multiple temporary subalignment files
330 and then merge them by calling bam_merge_core(). This function is
333 void bam_sort_core_ext(int is_by_qname, const char *fn, const char *prefix, size_t max_mem, int is_stdout)
337 bam_header_t *header;
341 g_is_by_qname = is_by_qname;
343 fp = strcmp(fn, "-")? bam_open(fn, "r") : bam_dopen(fileno(stdin), "r");
345 fprintf(stderr, "[bam_sort_core] fail to open file %s\n", fn);
348 header = bam_header_read(fp);
349 buf = (bam1_t**)calloc(max_mem / BAM_CORE_SIZE, sizeof(bam1_t*));
352 if (buf[k] == 0) buf[k] = (bam1_t*)calloc(1, sizeof(bam1_t));
354 if ((ret = bam_read1(fp, b)) < 0) break;
357 if (mem >= max_mem) {
358 sort_blocks(n++, k, buf, prefix, header, 0);
363 fprintf(stderr, "[bam_sort_core] truncated file. Continue anyway.\n");
364 if (n == 0) sort_blocks(-1, k, buf, prefix, header, is_stdout);
367 fprintf(stderr, "[bam_sort_core] merging from %d files...\n", n+1);
368 sort_blocks(n++, k, buf, prefix, header, 0);
369 fnout = (char*)calloc(strlen(prefix) + 20, 1);
370 if (is_stdout) sprintf(fnout, "-");
371 else sprintf(fnout, "%s.bam", prefix);
372 fns = (char**)calloc(n, sizeof(char*));
373 for (i = 0; i < n; ++i) {
374 fns[i] = (char*)calloc(strlen(prefix) + 20, 1);
375 sprintf(fns[i], "%s.%.4d.bam", prefix, i);
377 bam_merge_core(is_by_qname, fnout, 0, n, fns, 0, 0);
379 for (i = 0; i < n; ++i) {
385 for (k = 0; k < max_mem / BAM_CORE_SIZE; ++k) {
392 bam_header_destroy(header);
396 void bam_sort_core(int is_by_qname, const char *fn, const char *prefix, size_t max_mem)
398 bam_sort_core_ext(is_by_qname, fn, prefix, max_mem, 0);
401 int bam_sort(int argc, char *argv[])
403 size_t max_mem = 500000000;
404 int c, is_by_qname = 0, is_stdout = 0;
405 while ((c = getopt(argc, argv, "nom:")) >= 0) {
407 case 'o': is_stdout = 1; break;
408 case 'n': is_by_qname = 1; break;
409 case 'm': max_mem = atol(optarg); break;
412 if (optind + 2 > argc) {
413 fprintf(stderr, "Usage: samtools sort [-on] [-m <maxMem>] <in.bam> <out.prefix>\n");
416 bam_sort_core_ext(is_by_qname, argv[optind], argv[optind+1], max_mem, is_stdout);