fixref round up float numbers, convert small numbers to 0.

[logust79]

Hello developers!

I noticed that the fixref plugin (v1.22.1) will round up float numbers, and flatten small numbers to 0. Is this expected?

Example input vcf:

##fileformat=VCFv4.2
##INFO=<ID=AF,Number=A,Type=Float>
##INFO=<ID=SmallNumber,Number=A,Type=Float>
##contig=<ID=chr1,length=1000000000000>
#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO
chr1	55038469	.	C	A	.	.	AF=0.123894357;SmallNumber=2.03e-78

Example output vcf:

##fileformat=VCFv4.2
##FILTER=<ID=PASS,Description="All filters passed">
##INFO=<ID=AF,Number=A,Type=Float>
##INFO=<ID=SmallNumber,Number=A,Type=Float>
##contig=<ID=chr1,length=1000000000000>
##INFO=<ID=FIXREF,Number=.,Type=String,Description="The change made by bcftools/fixref">
##bcftools_pluginVersion=6b463daf-dirty+htslib-1.22.1
##bcftools_pluginCommand=plugin fixref -Ov -o out.vcf -- test.vcf -f /data/human_ref/Homo_sapiens_assembly38.fasta.gz -m swap; Date=Thu Nov 27 11:11:24 2025
#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO
chr1	55038469	.	C	A	.	.	AF=0.123894;SmallNumber=0;FIXREF=none

Notice that AF is rounded up to the sixth digit after decimal point and SmallNumber got flattened to 0.
command:

bcftools +fixref test.vcf -Ov -o out.vcf -- -f <ref> -m swap

[logust79]

I did some searches in the issues and this behaviour seems to be expected. I agree for variables such as AF this makes sense. However some small numbers, say, coming from p value calculations in a GWAS study, can easily go below 1e-45, which would be converted to 0 by norm and fixref. What is the recommended way to deal with this sort of cases? By coding SmallNumber as String?