Help understanding index creation using bcltools view

[Smeds]

Using the option --write-index together with bcltools view makes it possible to automatically create an index-file. Currenlty the default ouput (at least for a vcf.gz file) is a csi-file and I would like to have the possibility of having a tbi files as output.

I have been looking into adding the possibility of specifying output type for a vcf.gz index (csi or tbi) but got stuck trying to understand the code.

bcftools/version.c

Lines 116 to 141 in 95c1fb6

	int init_index(htsFile *fh, bcf_hdr_t *hdr, const char *fname, char **idx_fname)
	{
	int min_shift = 14; // CSI
	if ( !fname || !*fname || !strcmp(fname, "-") ) return -1;
	char *delim = strstr(fname, HTS_IDX_DELIM);
	if (delim)
	{
	delim += strlen(HTS_IDX_DELIM);
	*idx_fname = strdup(delim);
	if ( !*idx_fname ) return -1;
	size_t l = strlen(*idx_fname);
	if ( l >= 4 && strcmp(*idx_fname + l - 4, ".tbi")==0 ) min_shift = 0;
	}
	else
	{
	if ( !(*idx_fname = malloc(strlen(fname)+6)) ) return -1;
	sprintf(*idx_fname, "%s.csi", fname);
	}
	if ( bcf_idx_init(fh, hdr, min_shift, *idx_fname) < 0 ) return -1;
	return 0;
	}

If I understand the code correctly the function will look for occurrences of HTS_IDX_DELIM (##idx##) in the name of the file that we are trying to index, which I never will have in my filename.

Feels like I have misunderstood the code. Is the idea that the function should get the filename of the indexed-file, ex nam.vcf.gz.tbi as input? And then using some kind of regex (which isn't ##idx##) find the files that should get a tbi filetype?

[Smeds]

#2008