How long does it take for a large file?

I'll test. Meanwhile, I messed up the GH commits with another PR, so will fix and force push

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.17%. Comparing base (a080e8f) to head (3421907).
Report is 8 commits behind head on main.

@@           Coverage Diff           @@
##             main     #960   +/-   ##
=======================================
  Coverage   93.16%   93.17%           
=======================================
  Files          18       18           
  Lines        6337     6340    +3     
  Branches     1133     1135    +2     
=======================================
+ Hits         5904     5907    +3     
  Misses        294      294           
  Partials      139      139           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

LGTM - I don't think the loop through the alleles will be that expensive for most cases - unless lots of sites have hundreds of alleles.

This has made me realise the vcfzarr spec should probably be more explicit that zero-length alleles are not allowed. It is currently there as implicit as it states "" is a fill value.

This has made me realise the vcfzarr spec should probably be more explicit that zero-length alleles are not allowed. It is currently there as implicit as it states "" is a fill value.

Zero-length alleles are not allowed in VCF files, so I guess that's another way in which it is implicit. However, we might want to allow representation of indels internally using a "", otherwise there are forms of genetic variation that can be represented in a VCF that can't be represented in the tsinfer input (see #893 (comment))

Rolled into #963 as part of review, and tested with a large file.