int cram_reheader_inplace(cram_fd *fd, const bam_hdr_t *h, const char *arg_list,
int add_PG)
{
switch (cram_major_vers(fd)) {
case 2: return cram_reheader_inplace2(fd, h, arg_list, add_PG);
case 3: return cram_reheader_inplace3(fd, h, arg_list, add_PG);
default:
fprintf(stderr, "[%s] unsupported CRAM version %d\n", __func__,
cram_major_vers(fd));
return -1;
}
}
/*
* Reads a version 2 CRAM file and replaces the header in-place,
* provided the header is small enough to fit without growing the
* entire file.
*
* Version 2 format has an uncompressed SAM header with multiple nul
* termination bytes to permit inline header editing.
*
* Returns 0 on success;
* -1 on general failure;
* -2 on failure due to insufficient size
*/
int cram_reheader_inplace2(cram_fd *fd, const bam_hdr_t *h, const char *arg_list,
int add_PG)
{
cram_container *c = NULL;
cram_block *b = NULL;
SAM_hdr *hdr = NULL;
off_t start;
int ret = -1;
if (cram_major_vers(fd) < 2 ||
cram_major_vers(fd) > 3) {
fprintf(stderr, "[%s] unsupported CRAM version %d\n", __func__,
cram_major_vers(fd));
goto err;
}
if (!(hdr = sam_hdr_parse_(h->text, h->l_text)))
goto err;
if (add_PG && sam_hdr_add_PG(hdr, "samtools", "VN", samtools_version(),
arg_list ? "CL": NULL,
arg_list ? arg_list : NULL,
NULL))
goto err;
int header_len = sam_hdr_length(hdr);
/* Fix M5 strings? Maybe out of scope for this tool */
// Load the existing header
if ((start = hseek(cram_fd_get_fp(fd), 26, SEEK_SET)) != 26)
goto err;
if (!(c = cram_read_container(fd)))
goto err;
// Version 2.1 has a single uncompressed block which is nul
// terminated with many nuls to permit growth.
//
// So load old block and keep all contents identical bar the
// header text itself
if (!(b = cram_read_block(fd)))
goto err;
if (cram_block_get_uncomp_size(b) < header_len+4) {
fprintf(stderr, "New header will not fit. Use non-inplace version (%d > %d)\n",
header_len+4, cram_block_get_uncomp_size(b));
ret = -2;
goto err;
}
cram_block_set_offset(b, 0); // rewind block
int32_put_blk(b, header_len);
cram_block_append(b, sam_hdr_str(hdr), header_len);
// Zero the remaining block
memset(cram_block_get_data(b)+cram_block_get_offset(b), 0,
cram_block_get_uncomp_size(b) - cram_block_get_offset(b));
// Make sure all sizes and byte-offsets are consistent after memset
cram_block_set_offset(b, cram_block_get_uncomp_size(b));
cram_block_set_comp_size(b, cram_block_get_uncomp_size(b));
if (hseek(cram_fd_get_fp(fd), start, SEEK_SET) != start)
goto err;
if (cram_write_container(fd, c) == -1)
goto err;
if (cram_write_block(fd, b) == -1)
goto err;
ret = 0;
err:
if (c) cram_free_container(c);
if (b) cram_free_block(b);
if (hdr) sam_hdr_free(hdr);
return ret;
}
/*
* Reads a version 3 CRAM file and replaces the header in-place,
* provided the header is small enough to fit without growing the
* entire file.
*
* Version 3 format has a SAM header held as an (optionally)
* compressed block within the header container. Additional
* uncompressed blocks or simply unallocated space (the difference
* between total block sizes and the container size) are used to
* provide room for growth or contraction of the compressed header.
*
* Returns 0 on success;
* -1 on general failure;
* -2 on failure due to insufficient size
*/
int cram_reheader_inplace3(cram_fd *fd, const bam_hdr_t *h, const char *arg_list,
int add_PG)
{
cram_container *c = NULL;
cram_block *b = NULL;
SAM_hdr *hdr = NULL;
off_t start, sz, end;
int container_sz, max_container_sz;
char *buf = NULL;
int ret = -1;
if (cram_major_vers(fd) < 2 ||
cram_major_vers(fd) > 3) {
fprintf(stderr, "[%s] unsupported CRAM version %d\n", __func__,
cram_major_vers(fd));
goto err;
}
if (!(hdr = sam_hdr_parse_(h->text, h->l_text)))
goto err;
if (add_PG && sam_hdr_add_PG(hdr, "samtools", "VN", samtools_version(),
arg_list ? "CL": NULL,
arg_list ? arg_list : NULL,
NULL))
goto err;
int header_len = sam_hdr_length(hdr);
/* Fix M5 strings? Maybe out of scope for this tool */
// Find current size of SAM header block
if ((start = hseek(cram_fd_get_fp(fd), 26, SEEK_SET)) != 26)
goto err;
if (!(c = cram_read_container(fd)))
goto err;
// +5 allows num_landmarks to increase from 0 to 1 (Cramtools)
max_container_sz = cram_container_size(c)+5;
sz = htell(cram_fd_get_fp(fd)) + cram_container_get_length(c) - start;
end = htell(cram_fd_get_fp(fd)) + cram_container_get_length(c);
// We force 1 block instead of (optionally) 2. C CRAM
// implementations for v3 were writing 1 compressed block followed
// by 1 uncompressed block. However this is tricky to deal with
// as changing block sizes can mean the block header also changes
// size due to itf8 and variable size integers.
//
// If we had 1 block, this doesn't change anything.
// If we had 2 blocks, the new container header will be smaller by
// 1+ bytes, requiring the cram_container_get_length(c) to be larger in value.
// However this is an int32 instead of itf8 so the container
// header structure stays the same size. This means we can always
// reduce the number of blocks without running into size problems.
cram_container_set_num_blocks(c, 1);
int32_t *landmark;
int32_t num_landmarks;
landmark = cram_container_get_landmarks(c, &num_landmarks);
if (num_landmarks && landmark) {
num_landmarks = 1;
landmark[0] = 0;
} else {
num_landmarks = 0;
}
cram_container_set_landmarks(c, num_landmarks, landmark);
buf = malloc(max_container_sz);
container_sz = max_container_sz;
if (cram_store_container(fd, c, buf, &container_sz) != 0)
goto err;
if (!buf)
goto err;
// Proposed new length, but changing cram_container_get_length(c) may change the
// container_sz and thus the remainder (cram_container_get_length(c) itself).
cram_container_set_length(c, sz - container_sz);
int old_container_sz = container_sz;
container_sz = max_container_sz;
if (cram_store_container(fd, c, buf, &container_sz) != 0)
goto err;
if (old_container_sz != container_sz) {
fprintf(stderr, "Quirk of fate makes this troublesome! "
"Please use non-inplace version.\n");
goto err;
}
// Version 3.0 supports compressed header
b = cram_new_block(FILE_HEADER, 0);
int32_put_blk(b, header_len);
cram_block_append(b, sam_hdr_str(hdr), header_len);
cram_block_update_size(b);
cram_compress_block(fd, b, NULL, -1, -1);
if (hseek(cram_fd_get_fp(fd), 26, SEEK_SET) != 26)
goto err;
if (cram_block_size(b) > cram_container_get_length(c)) {
fprintf(stderr, "New header will not fit. Use non-inplace version"
" (%d > %d)\n",
(int)cram_block_size(b), cram_container_get_length(c));
ret = -2;
goto err;
}
if (cram_write_container(fd, c) == -1)
goto err;
if (cram_write_block(fd, b) == -1)
goto err;
// Blank out the remainder
int rsz = end - htell(cram_fd_get_fp(fd));
assert(rsz >= 0);
if (rsz) {
char *rem = calloc(1, rsz);
ret = hwrite(cram_fd_get_fp(fd), rem, rsz) == rsz ? 0 : -1;
free(rem);
}
err:
if (c) cram_free_container(c);
if (buf) free(buf);
if (b) cram_free_block(b);
if (hdr) sam_hdr_free(hdr);
return ret;
}
/*
* Check the files are consistent and capable of being concatenated.
* Also fills out the rg2id read-group hash and the version numbers
* and produces a new bam_hdr_t structure with merged RG lines.
* Note it is only a simple merge, as we lack the niceties of a proper
* header API.
*
* Returns updated header on success;
* NULL on failure.
*/
static bam_hdr_t *cram_cat_check_hdr(int nfn, char * const *fn, const bam_hdr_t *h,
khash_s2i **rg2id, int *vers_maj_p, int *vers_min_p) {
int i, vers_maj = -1, vers_min = -1;
bam_hdr_t *new_h = NULL;
if (h) {
new_h = bam_hdr_dup(h);
*rg2id = hash_rg(new_h);
}
for (i = 0; i < nfn; ++i) {
samFile *in;
cram_fd *in_c;
khint_t ki;
int new_rg = -1;
in = sam_open(fn[i], "rc");
if (in == 0) {
fprintf(stderr, "[%s] ERROR: fail to open file '%s'.\n", __func__, fn[i]);
return NULL;
}
in_c = in->fp.cram;
int vmaj = cram_major_vers(in_c);
int vmin = cram_minor_vers(in_c);
if ((vers_maj != -1 && vers_maj != vmaj) ||
(vers_min != -1 && vers_min != vmin)) {
fprintf(stderr, "[%s] ERROR: input files have differing version numbers.\n",
__func__);
return NULL;
}
vers_maj = vmaj;
vers_min = vmin;
bam_hdr_t *old = sam_hdr_read(in);
khash_s2i *rg2id_in = hash_rg(old);
if (!new_h) {
new_h = bam_hdr_dup(old);
*rg2id = hash_rg(new_h);
}
// Add any existing @RG entries to our global @RG hash.
for (ki = 0; ki < rg2id_in->n_id; ki++) {
int added;
new_rg = hash_s2i_inc(*rg2id, rg2id_in->id[ki], rg2id_in->line[ki], &added);
//fprintf(stderr, "RG %s: #%d -> #%d\n",
// rg2id_in->id[ki], ki, new_rg);
if (added) {
// Also add to new_h
const char *line = rg2id_in->line[ki];
const char *line_end = line;
while (*line && *line_end++ != '\n')
;
new_h->l_text += line_end - line;
new_h->text = realloc(new_h->text, new_h->l_text+1);
strncat(&new_h->text[new_h->l_text - (line_end - line)],
line, line_end - line);
}
if (new_rg != ki && rg2id_in->n_id > 1) {
fprintf(stderr, "[%s] ERROR: Same size @RG lists but differing order / contents\n",
__func__);
return NULL;
}
}
hash_s2i_free(rg2id_in);
bam_hdr_destroy(old);
sam_close(in);
}
*vers_maj_p = vers_maj;
*vers_min_p = vers_min;
return new_h;
}
