Petr Danecek

The tag PL was encountered in the VCF body but was not defined in the header. When such tag is encountered, the reader can decide to abort (but that would...

No. The first is printed while reading and can be recovered from, hence a warning. It does not know that the record will be later used for writing. Only then...

That's a neat idea of generating the header on the fly and adding it later. There is a problem with the annotation type: in order to proceed we declare it...

Just to confirm (as it was me who added the abort) - a proper error propagation is preferred indeed. However, if the abort is going to be removed, it would...

@daviesrob `bcf_update_format()` uses your `serialize_float_array()` which calls `float_to_le()`, so no special handling of `vector_end` values should be required.

The specification requires that all values after `vector_end` are filled with `vector_end` values (if the sample has fewer values than the rest of the samples). The `bcf_update_format()` function assumes that...

Out of curiosity, have you done any benchmarks? When you say slow CRC, I am wondering how slow it really is.

I agree that'd be a reasonable behavior, happy to accept a pull request. Thank you for raising the issue.

Thanks for the analysis! My thoughts about this are: - I agree replacing with mpileup2 is a better approach and also gives more flexibility - a secondary "just gone by"...

This was never explicitly defined. I am open to modifying the existing specification and htslib as mixed ploidy is not a widely used feature.