Browse code

embed samtools sources directly, since Rsamtools has moved to htslib

Michael Lawrence authored on 11/02/2019 20:33:07
Showing1 changed files
1 1
new file mode 100644
... ...
@@ -0,0 +1,474 @@
1
+#include <stdio.h>
2
+#include <ctype.h>
3
+#include <errno.h>
4
+#include <assert.h>
5
+#include "bam.h"
6
+#include "bam_endian.h"
7
+#include "kstring.h"
8
+#include "sam_header.h"
9
+
10
+int bam_is_be = 0, bam_verbose = 2, bam_no_B = 0;
11
+char *bam_flag2char_table = "pPuUrR12sfd\0\0\0\0\0";
12
+
13
+/**************************
14
+ * CIGAR related routines *
15
+ **************************/
16
+
17
+uint32_t bam_calend(const bam1_core_t *c, const uint32_t *cigar)
18
+{
19
+	int k, end = c->pos;
20
+	for (k = 0; k < c->n_cigar; ++k) {
21
+		int op  = bam_cigar_op(cigar[k]);
22
+		int len = bam_cigar_oplen(cigar[k]);
23
+		if (op == BAM_CBACK) { // move backward
24
+			int l, u, v;
25
+			if (k == c->n_cigar - 1) break; // skip trailing 'B'
26
+			for (l = k - 1, u = v = 0; l >= 0; --l) {
27
+				int op1  = bam_cigar_op(cigar[l]);
28
+				int len1 = bam_cigar_oplen(cigar[l]);
29
+				if (bam_cigar_type(op1)&1) { // consume query
30
+					if (u + len1 >= len) { // stop
31
+						if (bam_cigar_type(op1)&2) v += len - u;
32
+						break;
33
+					} else u += len1;
34
+				}
35
+				if (bam_cigar_type(op1)&2) v += len1;
36
+			}
37
+			end = l < 0? c->pos : end - v;
38
+		} else if (bam_cigar_type(op)&2) end += bam_cigar_oplen(cigar[k]);
39
+	}
40
+	return end;
41
+}
42
+
43
+int32_t bam_cigar2qlen(const bam1_core_t *c, const uint32_t *cigar)
44
+{
45
+	uint32_t k;
46
+	int32_t l = 0;
47
+	for (k = 0; k < c->n_cigar; ++k)
48
+		if (bam_cigar_type(bam_cigar_op(cigar[k]))&1)
49
+			l += bam_cigar_oplen(cigar[k]);
50
+	return l;
51
+}
52
+
53
+/********************
54
+ * BAM I/O routines *
55
+ ********************/
56
+
57
+bam_header_t *bam_header_init()
58
+{
59
+	bam_is_be = bam_is_big_endian();
60
+	return (bam_header_t*)calloc(1, sizeof(bam_header_t));
61
+}
62
+
63
+void bam_header_destroy(bam_header_t *header)
64
+{
65
+	int32_t i;
66
+	extern void bam_destroy_header_hash(bam_header_t *header);
67
+	if (header == 0) return;
68
+	if (header->target_name) {
69
+		for (i = 0; i < header->n_targets; ++i)
70
+			free(header->target_name[i]);
71
+		free(header->target_name);
72
+		free(header->target_len);
73
+	}
74
+	free(header->text);
75
+	if (header->dict) sam_header_free(header->dict);
76
+	if (header->rg2lib) sam_tbl_destroy(header->rg2lib);
77
+	bam_destroy_header_hash(header);
78
+	free(header);
79
+}
80
+
81
+bam_header_t *bam_header_read(bamFile fp)
82
+{
83
+	bam_header_t *header;
84
+	char buf[4];
85
+	int magic_len;
86
+	int32_t i = 1, name_len;
87
+	// check EOF
88
+	i = bgzf_check_EOF(fp);
89
+	if (i < 0) {
90
+		// If the file is a pipe, checking the EOF marker will *always* fail
91
+		// with ESPIPE.  Suppress the error message in this case.
92
+		if (errno != ESPIPE) perror("[bam_header_read] bgzf_check_EOF");
93
+	}
94
+	else if (i == 0) fprintf(stderr, "[bam_header_read] EOF marker is absent. The input is probably truncated.\n");
95
+	// read "BAM1"
96
+	magic_len = bam_read(fp, buf, 4);
97
+	if (magic_len != 4 || strncmp(buf, "BAM\001", 4) != 0) {
98
+		fprintf(stderr, "[bam_header_read] invalid BAM binary header (this is not a BAM file).\n");
99
+		return 0;
100
+	}
101
+	header = bam_header_init();
102
+	// read plain text and the number of reference sequences
103
+	bam_read(fp, &header->l_text, 4);
104
+	if (bam_is_be) bam_swap_endian_4p(&header->l_text);
105
+	header->text = (char*)calloc(header->l_text + 1, 1);
106
+	bam_read(fp, header->text, header->l_text);
107
+	bam_read(fp, &header->n_targets, 4);
108
+	if (bam_is_be) bam_swap_endian_4p(&header->n_targets);
109
+	// read reference sequence names and lengths
110
+	header->target_name = (char**)calloc(header->n_targets, sizeof(char*));
111
+	header->target_len = (uint32_t*)calloc(header->n_targets, 4);
112
+	for (i = 0; i != header->n_targets; ++i) {
113
+		bam_read(fp, &name_len, 4);
114
+		if (bam_is_be) bam_swap_endian_4p(&name_len);
115
+		header->target_name[i] = (char*)calloc(name_len, 1);
116
+		bam_read(fp, header->target_name[i], name_len);
117
+		bam_read(fp, &header->target_len[i], 4);
118
+		if (bam_is_be) bam_swap_endian_4p(&header->target_len[i]);
119
+	}
120
+	return header;
121
+}
122
+
123
+int bam_header_write(bamFile fp, const bam_header_t *header)
124
+{
125
+	char buf[4];
126
+	int32_t i, name_len, x;
127
+	// write "BAM1"
128
+	strncpy(buf, "BAM\001", 4);
129
+	bam_write(fp, buf, 4);
130
+	// write plain text and the number of reference sequences
131
+	if (bam_is_be) {
132
+		x = bam_swap_endian_4(header->l_text);
133
+		bam_write(fp, &x, 4);
134
+		if (header->l_text) bam_write(fp, header->text, header->l_text);
135
+		x = bam_swap_endian_4(header->n_targets);
136
+		bam_write(fp, &x, 4);
137
+	} else {
138
+		bam_write(fp, &header->l_text, 4);
139
+		if (header->l_text) bam_write(fp, header->text, header->l_text);
140
+		bam_write(fp, &header->n_targets, 4);
141
+	}
142
+	// write sequence names and lengths
143
+	for (i = 0; i != header->n_targets; ++i) {
144
+		char *p = header->target_name[i];
145
+		name_len = strlen(p) + 1;
146
+		if (bam_is_be) {
147
+			x = bam_swap_endian_4(name_len);
148
+			bam_write(fp, &x, 4);
149
+		} else bam_write(fp, &name_len, 4);
150
+		bam_write(fp, p, name_len);
151
+		if (bam_is_be) {
152
+			x = bam_swap_endian_4(header->target_len[i]);
153
+			bam_write(fp, &x, 4);
154
+		} else bam_write(fp, &header->target_len[i], 4);
155
+	}
156
+	bgzf_flush(fp);
157
+	return 0;
158
+}
159
+
160
+static void swap_endian_data(const bam1_core_t *c, int data_len, uint8_t *data)
161
+{
162
+	uint8_t *s;
163
+	uint32_t i, *cigar = (uint32_t*)(data + c->l_qname);
164
+	s = data + c->n_cigar*4 + c->l_qname + c->l_qseq + (c->l_qseq + 1)/2;
165
+	for (i = 0; i < c->n_cigar; ++i) bam_swap_endian_4p(&cigar[i]);
166
+	while (s < data + data_len) {
167
+		uint8_t type;
168
+		s += 2; // skip key
169
+		type = toupper(*s); ++s; // skip type
170
+		if (type == 'C' || type == 'A') ++s;
171
+		else if (type == 'S') { bam_swap_endian_2p(s); s += 2; }
172
+		else if (type == 'I' || type == 'F') { bam_swap_endian_4p(s); s += 4; }
173
+		else if (type == 'D') { bam_swap_endian_8p(s); s += 8; }
174
+		else if (type == 'Z' || type == 'H') { while (*s) ++s; ++s; }
175
+		else if (type == 'B') {
176
+			int32_t n, Bsize = bam_aux_type2size(*s);
177
+			memcpy(&n, s + 1, 4);
178
+			if (1 == Bsize) {
179
+			} else if (2 == Bsize) {
180
+				for (i = 0; i < n; i += 2)
181
+					bam_swap_endian_2p(s + 5 + i);
182
+			} else if (4 == Bsize) {
183
+				for (i = 0; i < n; i += 4)
184
+					bam_swap_endian_4p(s + 5 + i);
185
+			}
186
+			bam_swap_endian_4p(s+1); 
187
+		}
188
+	}
189
+}
190
+
191
+int bam_read1(bamFile fp, bam1_t *b)
192
+{
193
+	bam1_core_t *c = &b->core;
194
+	int32_t block_len, ret, i;
195
+	uint32_t x[8];
196
+
197
+	assert(BAM_CORE_SIZE == 32);
198
+	if ((ret = bam_read(fp, &block_len, 4)) != 4) {
199
+		if (ret == 0) return -1; // normal end-of-file
200
+		else return -2; // truncated
201
+	}
202
+	if (bam_read(fp, x, BAM_CORE_SIZE) != BAM_CORE_SIZE) return -3;
203
+	if (bam_is_be) {
204
+		bam_swap_endian_4p(&block_len);
205
+		for (i = 0; i < 8; ++i) bam_swap_endian_4p(x + i);
206
+	}
207
+	c->tid = x[0]; c->pos = x[1];
208
+	c->bin = x[2]>>16; c->qual = x[2]>>8&0xff; c->l_qname = x[2]&0xff;
209
+	c->flag = x[3]>>16; c->n_cigar = x[3]&0xffff;
210
+	c->l_qseq = x[4];
211
+	c->mtid = x[5]; c->mpos = x[6]; c->isize = x[7];
212
+	b->data_len = block_len - BAM_CORE_SIZE;
213
+	if (b->m_data < b->data_len) {
214
+		b->m_data = b->data_len;
215
+		kroundup32(b->m_data);
216
+		b->data = (uint8_t*)realloc(b->data, b->m_data);
217
+	}
218
+	if (bam_read(fp, b->data, b->data_len) != b->data_len) return -4;
219
+	b->l_aux = b->data_len - c->n_cigar * 4 - c->l_qname - c->l_qseq - (c->l_qseq+1)/2;
220
+	if (bam_is_be) swap_endian_data(c, b->data_len, b->data);
221
+	if (bam_no_B) bam_remove_B(b);
222
+	return 4 + block_len;
223
+}
224
+
225
+inline int bam_write1_core(bamFile fp, const bam1_core_t *c, int data_len, uint8_t *data)
226
+{
227
+	uint32_t x[8], block_len = data_len + BAM_CORE_SIZE, y;
228
+	int i;
229
+	assert(BAM_CORE_SIZE == 32);
230
+	x[0] = c->tid;
231
+	x[1] = c->pos;
232
+	x[2] = (uint32_t)c->bin<<16 | c->qual<<8 | c->l_qname;
233
+	x[3] = (uint32_t)c->flag<<16 | c->n_cigar;
234
+	x[4] = c->l_qseq;
235
+	x[5] = c->mtid;
236
+	x[6] = c->mpos;
237
+	x[7] = c->isize;
238
+	bgzf_flush_try(fp, 4 + block_len);
239
+	if (bam_is_be) {
240
+		for (i = 0; i < 8; ++i) bam_swap_endian_4p(x + i);
241
+		y = block_len;
242
+		bam_write(fp, bam_swap_endian_4p(&y), 4);
243
+		swap_endian_data(c, data_len, data);
244
+	} else bam_write(fp, &block_len, 4);
245
+	bam_write(fp, x, BAM_CORE_SIZE);
246
+	bam_write(fp, data, data_len);
247
+	if (bam_is_be) swap_endian_data(c, data_len, data);
248
+	return 4 + block_len;
249
+}
250
+
251
+int bam_write1(bamFile fp, const bam1_t *b)
252
+{
253
+	return bam_write1_core(fp, &b->core, b->data_len, b->data);
254
+}
255
+
256
+char *bam_format1_core(const bam_header_t *header, const bam1_t *b, int of)
257
+{
258
+	uint8_t *s = bam1_seq(b), *t = bam1_qual(b);
259
+	int i;
260
+	const bam1_core_t *c = &b->core;
261
+	kstring_t str;
262
+	str.l = str.m = 0; str.s = 0;
263
+
264
+	kputsn(bam1_qname(b), c->l_qname-1, &str); kputc('\t', &str);
265
+	if (of == BAM_OFDEC) { kputw(c->flag, &str); kputc('\t', &str); }
266
+	else if (of == BAM_OFHEX) ksprintf(&str, "0x%x\t", c->flag);
267
+	else { // BAM_OFSTR
268
+		for (i = 0; i < 16; ++i)
269
+			if ((c->flag & 1<<i) && bam_flag2char_table[i])
270
+				kputc(bam_flag2char_table[i], &str);
271
+		kputc('\t', &str);
272
+	}
273
+	if (c->tid < 0) kputsn("*\t", 2, &str);
274
+	else {
275
+		if (header) kputs(header->target_name[c->tid] , &str);
276
+		else kputw(c->tid, &str);
277
+		kputc('\t', &str);
278
+	}
279
+	kputw(c->pos + 1, &str); kputc('\t', &str); kputw(c->qual, &str); kputc('\t', &str);
280
+	if (c->n_cigar == 0) kputc('*', &str);
281
+	else {
282
+		uint32_t *cigar = bam1_cigar(b);
283
+		for (i = 0; i < c->n_cigar; ++i) {
284
+			kputw(bam1_cigar(b)[i]>>BAM_CIGAR_SHIFT, &str);
285
+			kputc(bam_cigar_opchr(cigar[i]), &str);
286
+		}
287
+	}
288
+	kputc('\t', &str);
289
+	if (c->mtid < 0) kputsn("*\t", 2, &str);
290
+	else if (c->mtid == c->tid) kputsn("=\t", 2, &str);
291
+	else {
292
+		if (header) kputs(header->target_name[c->mtid], &str);
293
+		else kputw(c->mtid, &str);
294
+		kputc('\t', &str);
295
+	}
296
+	kputw(c->mpos + 1, &str); kputc('\t', &str); kputw(c->isize, &str); kputc('\t', &str);
297
+	if (c->l_qseq) {
298
+		for (i = 0; i < c->l_qseq; ++i) kputc(bam_nt16_rev_table[bam1_seqi(s, i)], &str);
299
+		kputc('\t', &str);
300
+		if (t[0] == 0xff) kputc('*', &str);
301
+		else for (i = 0; i < c->l_qseq; ++i) kputc(t[i] + 33, &str);
302
+	} else kputsn("*\t*", 3, &str);
303
+	s = bam1_aux(b);
304
+	while (s < b->data + b->data_len) {
305
+		uint8_t type, key[2];
306
+		key[0] = s[0]; key[1] = s[1];
307
+		s += 2; type = *s; ++s;
308
+		kputc('\t', &str); kputsn((char*)key, 2, &str); kputc(':', &str);
309
+		if (type == 'A') { kputsn("A:", 2, &str); kputc(*s, &str); ++s; }
310
+		else if (type == 'C') { kputsn("i:", 2, &str); kputw(*s, &str); ++s; }
311
+		else if (type == 'c') { kputsn("i:", 2, &str); kputw(*(int8_t*)s, &str); ++s; }
312
+		else if (type == 'S') { kputsn("i:", 2, &str); kputw(*(uint16_t*)s, &str); s += 2; }
313
+		else if (type == 's') { kputsn("i:", 2, &str); kputw(*(int16_t*)s, &str); s += 2; }
314
+		else if (type == 'I') { kputsn("i:", 2, &str); kputuw(*(uint32_t*)s, &str); s += 4; }
315
+		else if (type == 'i') { kputsn("i:", 2, &str); kputw(*(int32_t*)s, &str); s += 4; }
316
+		else if (type == 'f') { ksprintf(&str, "f:%g", *(float*)s); s += 4; }
317
+		else if (type == 'd') { ksprintf(&str, "d:%lg", *(double*)s); s += 8; }
318
+		else if (type == 'Z' || type == 'H') { kputc(type, &str); kputc(':', &str); while (*s) kputc(*s++, &str); ++s; }
319
+		else if (type == 'B') {
320
+			uint8_t sub_type = *(s++);
321
+			int32_t n;
322
+			memcpy(&n, s, 4);
323
+			s += 4; // no point to the start of the array
324
+			kputc(type, &str); kputc(':', &str); kputc(sub_type, &str); // write the typing
325
+			for (i = 0; i < n; ++i) {
326
+				kputc(',', &str);
327
+				if ('c' == sub_type || 'c' == sub_type) { kputw(*(int8_t*)s, &str); ++s; }
328
+				else if ('C' == sub_type) { kputw(*(uint8_t*)s, &str); ++s; }
329
+				else if ('s' == sub_type) { kputw(*(int16_t*)s, &str); s += 2; }
330
+				else if ('S' == sub_type) { kputw(*(uint16_t*)s, &str); s += 2; }
331
+				else if ('i' == sub_type) { kputw(*(int32_t*)s, &str); s += 4; }
332
+				else if ('I' == sub_type) { kputuw(*(uint32_t*)s, &str); s += 4; }
333
+				else if ('f' == sub_type) { ksprintf(&str, "%g", *(float*)s); s += 4; }
334
+			}
335
+		}
336
+	}
337
+	return str.s;
338
+}
339
+
340
+char *bam_format1(const bam_header_t *header, const bam1_t *b)
341
+{
342
+	return bam_format1_core(header, b, BAM_OFDEC);
343
+}
344
+
345
+void bam_view1(const bam_header_t *header, const bam1_t *b)
346
+{
347
+	char *s = bam_format1(header, b);
348
+	puts(s);
349
+	free(s);
350
+}
351
+
352
+int bam_validate1(const bam_header_t *header, const bam1_t *b)
353
+{
354
+	char *s;
355
+
356
+	if (b->core.tid < -1 || b->core.mtid < -1) return 0;
357
+	if (header && (b->core.tid >= header->n_targets || b->core.mtid >= header->n_targets)) return 0;
358
+
359
+	if (b->data_len < b->core.l_qname) return 0;
360
+	s = memchr(bam1_qname(b), '\0', b->core.l_qname);
361
+	if (s != &bam1_qname(b)[b->core.l_qname-1]) return 0;
362
+
363
+	// FIXME: Other fields could also be checked, especially the auxiliary data
364
+
365
+	return 1;
366
+}
367
+
368
+// FIXME: we should also check the LB tag associated with each alignment
369
+const char *bam_get_library(bam_header_t *h, const bam1_t *b)
370
+{
371
+	const uint8_t *rg;
372
+	if (h->dict == 0) h->dict = sam_header_parse2(h->text);
373
+	if (h->rg2lib == 0) h->rg2lib = sam_header2tbl(h->dict, "RG", "ID", "LB");
374
+	rg = bam_aux_get(b, "RG");
375
+	return (rg == 0)? 0 : sam_tbl_get(h->rg2lib, (const char*)(rg + 1));
376
+}
377
+
378
+/************
379
+ * Remove B *
380
+ ************/
381
+
382
+int bam_remove_B(bam1_t *b)
383
+{
384
+	int i, j, end_j, k, l, no_qual;
385
+	uint32_t *cigar, *new_cigar;
386
+	uint8_t *seq, *qual, *p;
387
+	// test if removal is necessary
388
+	if (b->core.flag & BAM_FUNMAP) return 0; // unmapped; do nothing
389
+	cigar = bam1_cigar(b);
390
+	for (k = 0; k < b->core.n_cigar; ++k)
391
+		if (bam_cigar_op(cigar[k]) == BAM_CBACK) break;
392
+	if (k == b->core.n_cigar) return 0; // no 'B'
393
+	if (bam_cigar_op(cigar[0]) == BAM_CBACK) goto rmB_err; // cannot be removed
394
+	// allocate memory for the new CIGAR
395
+	if (b->data_len + (b->core.n_cigar + 1) * 4 > b->m_data) { // not enough memory
396
+		b->m_data = b->data_len + b->core.n_cigar * 4;
397
+		kroundup32(b->m_data);
398
+		b->data = (uint8_t*)realloc(b->data, b->m_data);
399
+		cigar = bam1_cigar(b); // after realloc, cigar may be changed
400
+	}
401
+	new_cigar = (uint32_t*)(b->data + (b->m_data - b->core.n_cigar * 4)); // from the end of b->data
402
+	// the core loop
403
+	seq = bam1_seq(b); qual = bam1_qual(b);
404
+	no_qual = (qual[0] == 0xff); // test whether base quality is available
405
+	i = j = 0; end_j = -1;
406
+	for (k = l = 0; k < b->core.n_cigar; ++k) {
407
+		int op  = bam_cigar_op(cigar[k]);
408
+		int len = bam_cigar_oplen(cigar[k]);
409
+		if (op == BAM_CBACK) { // the backward operation
410
+			int t, u;
411
+			if (k == b->core.n_cigar - 1) break; // ignore 'B' at the end of CIGAR
412
+			if (len > j) goto rmB_err; // an excessively long backward
413
+			for (t = l - 1, u = 0; t >= 0; --t) { // look back
414
+				int op1  = bam_cigar_op(new_cigar[t]);
415
+				int len1 = bam_cigar_oplen(new_cigar[t]);
416
+				if (bam_cigar_type(op1)&1) { // consume the query
417
+					if (u + len1 >= len) { // stop
418
+						new_cigar[t] -= (len - u) << BAM_CIGAR_SHIFT;
419
+						break;
420
+					} else u += len1;
421
+				}
422
+			}
423
+			if (bam_cigar_oplen(new_cigar[t]) == 0) --t; // squeeze out the zero-length operation
424
+			l = t + 1;
425
+			end_j = j; j -= len;
426
+		} else { // other CIGAR operations
427
+			new_cigar[l++] = cigar[k];
428
+			if (bam_cigar_type(op)&1) { // consume the query
429
+				if (i != j) { // no need to copy if i == j
430
+					int u, c, c0;
431
+					for (u = 0; u < len; ++u) { // construct the consensus
432
+						c = bam1_seqi(seq, i+u);
433
+						if (j + u < end_j) { // in an overlap
434
+							c0 = bam1_seqi(seq, j+u);
435
+							if (c != c0) { // a mismatch; choose the better base
436
+								if (qual[j+u] < qual[i+u]) { // the base in the 2nd segment is better
437
+									bam1_seq_seti(seq, j+u, c);
438
+									qual[j+u] = qual[i+u] - qual[j+u];
439
+								} else qual[j+u] -= qual[i+u]; // the 1st is better; reduce base quality
440
+							} else qual[j+u] = qual[j+u] > qual[i+u]? qual[j+u] : qual[i+u];
441
+						} else { // not in an overlap; copy over
442
+							bam1_seq_seti(seq, j+u, c);
443
+							qual[j+u] = qual[i+u];
444
+						}
445
+					}
446
+				}
447
+				i += len, j += len;
448
+			}
449
+		}
450
+	}
451
+	if (no_qual) qual[0] = 0xff; // in very rare cases, this may be modified
452
+	// merge adjacent operations if possible
453
+	for (k = 1; k < l; ++k)
454
+		if (bam_cigar_op(new_cigar[k]) == bam_cigar_op(new_cigar[k-1]))
455
+			new_cigar[k] += new_cigar[k-1] >> BAM_CIGAR_SHIFT << BAM_CIGAR_SHIFT, new_cigar[k-1] &= 0xf;
456
+	// kill zero length operations
457
+	for (k = i = 0; k < l; ++k)
458
+		if (new_cigar[k] >> BAM_CIGAR_SHIFT)
459
+			new_cigar[i++] = new_cigar[k];
460
+	l = i;
461
+	// update b
462
+	memcpy(cigar, new_cigar, l * 4); // set CIGAR
463
+	p = b->data + b->core.l_qname + l * 4;
464
+	memmove(p, seq, (j+1)>>1); p += (j+1)>>1; // set SEQ
465
+	memmove(p, qual, j); p += j; // set QUAL
466
+	memmove(p, bam1_aux(b), b->l_aux); p += b->l_aux; // set optional fields
467
+	b->core.n_cigar = l, b->core.l_qseq = j; // update CIGAR length and query length
468
+	b->data_len = p - b->data; // update record length
469
+	return 0;
470
+
471
+rmB_err:
472
+	b->core.flag |= BAM_FUNMAP;
473
+	return -1;
474
+}