bzip2 often beats xz for highly regular data, like a big csv or jsonl file. So I usually try both approaches on those sorts of files. For everything else my go

bzip2 often beats xz for highly regular data, like a big csv or jsonl file. So I usually try both approaches on those sorts of files. For everything else my go to is zstd or xz. Xz usually wins by a little, but is a lot slower. Zstd is great in practice. Just yesterday I had a file where bzip2 won. I would love to see a modern version with the first few stages of bwt feeding into xz or zstd.