/*
* Renumbers RG numbers in a cram compression header.
*
* CRAM stores RG as the Nth number in the header, rather than a
* string holding the ID: tag. This is smaller in space, but means
* "samtools cat" to join files together that contain single but
* different RG lines needs a way of renumbering them.
*
* The file descriptor is expected to be immediately after the
* cram_container structure (ie before the cram compression header).
* Due to the nature of the CRAM format, this needs to read and write
* the blocks itself. Note that there may be multiple slices within
* the container, meaning multiple compression headers to manipulate.
* Changing RG may change the size of the compression header and
* therefore the length field in the container. Hence we rewrite all
* blocks just incase and also emit the adjusted container.
*
* The current implementation can only cope with renumbering a single
* RG (and only then if it is using HUFFMAN or BETA codecs). In
* theory it *may* be possible to renumber multiple RGs if they use
* HUFFMAN to the CORE block or use an external block unshared by any
* other data series. So we have an API that can be upgraded to
* support this, but do not implement it for now. An example
* implementation of RG as an EXTERNAL block would be to find that
* block and rewrite it, returning the number of blocks consumed.
*
* Returns 0 on success;
* -1 if unable to edit;
* -2 on other errors (eg I/O).
*/
int cram_transcode_rg(cram_fd *in, cram_fd *out,
cram_container *c,
int nrg, int *in_rg, int *out_rg) {
int new_rg = *out_rg, old_size, new_size;
cram_block *o_blk, *n_blk;
cram_block_compression_hdr *ch;
if (nrg != 1) {
hts_log_error("CRAM transcode supports only a single RG");
return -2;
}
// Produce a new block holding the updated compression header,
// with RG transcoded to a new value. (Single only supported.)
o_blk = cram_read_block(in);
old_size = cram_block_size(o_blk);
ch = cram_decode_compression_header(in, o_blk);
if (cram_block_compression_hdr_set_rg(ch, new_rg) != 0)
return -1;
cram_block_compression_hdr_decoder2encoder(in, ch);
n_blk = cram_encode_compression_header(in, c, ch);
cram_free_compression_header(ch);
/*
* Warning: this has internal knowledge of the cram compression
* header format.
*
* The decoder doesn't set c->tags_used, so the encoder puts a two
* byte blank segment. This means n_blk is too short. We skip
* through the decoded old block (o_blk) and copy from there.
*/
char *cp = cram_block_get_data(o_blk);
char *op = cp;
char *endp = cp + cram_block_get_uncomp_size(o_blk);
//fprintf(stderr, "sz = %d\n", (int)(endp-cp));
int32_t i32;
cp += safe_itf8_get(cp, endp, &i32);
cp += i32;
cp += safe_itf8_get(cp, endp, &i32);
cp += i32;
op = cp;
cp += safe_itf8_get(cp, endp, &i32);
i32 += (cp-op);
//fprintf(stderr, "remaining %d bytes\n", i32);
cram_block_set_size(n_blk, cram_block_get_size(n_blk)-2);
cram_block_append(n_blk, op, i32);
cram_block_update_size(n_blk);
new_size = cram_block_size(n_blk);
//fprintf(stderr, "size %d -> %d\n", old_size, new_size);
// Now we've constructedthe updated compression header,
// amend the container too (it may have changed size).
int32_t *landmarks, num_landmarks;
landmarks = cram_container_get_landmarks(c, &num_landmarks);
if (old_size != new_size) {
int diff = new_size - old_size, j;
for (j = 0; j < num_landmarks; j++)
landmarks[j] += diff;
//cram_container_set_landmarks(c, num_landmarks, landmarks);
cram_container_set_length(c, cram_container_get_length(c) + diff);
}
// Finally write it all out; container, compression header,
// and then all the remaining slice blocks.
if (cram_write_container(out, c) != 0)
return -2;
cram_write_block(out, n_blk);
cram_free_block(o_blk);
cram_free_block(n_blk);
// Container num_blocks can be invalid, due to a bug.
// Instead we iterate in slice context instead.
return cram_copy_slice(in, out, num_landmarks);
}
/*
* 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;
}
