I was hoping to use samtools split -d HP to split a phased bam file into its various haplotypes, but currently only string tags are supported. I tried to find a nicer solution, but this manual branch worked with existing htslib functions (pointers are not my friend) changing the line here. Given there is already a --max-split option, was there a reason why integers are generally unsuitable compared to strings? I would imagine splitting by phase is not an obscure thing to do (and easier than multiple samtools view -d HP:<phase>). The other option would be to also encode phase/haplotype as a string, but would be redundant information.

char *val[2];
if (strcmp(tag, "HP")) sprintf(val,"%ld", bam_aux2i(tag));
else *val = tag ? bam_aux2Z(tag) : NULL;

I also noticed that if you split by tag, the format string is forcibly overwritten from whatever you provide.

The format string documentation definitely seems to indicate you can also change the -f format for TAGs.

Format string expansions:
  %%     %
  %*     basename
  %#     index (of @RG in the header, or count of TAG values seen so far)
  %!     @RG ID or TAG value
  %.     filename extension for output format

Best,
Alex