clintval / vartovcf / 39886296310

98%

Build Type

push

github

Committed by web-flow

Commit Message

refactor!: replace rust-htslib with noodles (#161)

* refactor!: write VCF headers and records without rust-htslib

Format the header and each record line in a new vcf module instead of
building htslib records and having htslib print them. The module writes
what htslib wrote: ##fileformat, then the header lines in the order they
were pushed with repeats dropped, and record fields with a missing value
as `.`, an empty one left out, and floats as htslib's kputd prints them.
The float formatting was compared with kputd over every 32-bit float.

The header is no longer formatted through a temporary file, so the
--tmp-dir help names only the records sorting spills.

BREAKING CHANGE: tumor_only_header and tumor_normal_header return a
vcf::Header, contigs_to_vcf_header, fasta_contigs_to_vcf_header and
fasta_path_to_vcf_header take one, gt_value returns vcf::GenotypeAllele
values, and the missing-value trait is vcf::Numeric.

* refactor: read the reference FASTA with noodles-bgzf instead of htslib

Read a window of bases from the byte the FASTA index places its first
base at, keeping each printable character other than a space, as
htslib's faidx reads them, so a line with trailing whitespace reads the
same. A bgzipped FASTA is read through its .gzi index with noodles-bgzf.
A contig the index lists twice is read from its first entry, as htslib
read it.

* refactor!: write and index bgzipped output with noodles, not rust-htslib

Write a .gz output with noodles-bgzf and index it as `tabix -p vcf`
does, with noodles-csi and noodles-tabix: a record spans its POS to the
end of its REF allele, or to a later INFO END, and the records of a
contig must be together, sorted and within what the index addresses.
Refusing them now names the contig and the positions. A CSI index takes
its depth from the longest ##contig line as tabix does, and sets each
bin's offset to the linear index's at the bin's first window, as htslib
does, so a reader does not skip a longer record tha... (continued)

Coverage Stats

1000 of 1026 new or added lines in 7 files covered. (97.47%)

1 existing line in 1 file now uncovered.

4280 of 4363 relevant lines covered (98.1%)

165232.6 hits per line