`seqkit shuffle` is deterministic by default

[apcamargo]

This is not strictly a bug, but it may lead to unexpected behavior for users. In many tools that perform shuffling (for example, the shuf command), the default behavior is non-deterministic, typically using a time-based seed. If deterministic output is required, users are expected to set a seed explicitly.

In contrast, seqkit shuffle uses a fixed seed by default via the -s parameter. As a result, users need to pass something like -s $(date +%s) to obtain non-deterministic behavior.

This may be an intentional design choice. However, because users might reasonably assume that shuffling is non-deterministic by default, this behavior could lead to unexpected results.

[shenwei356]

How about adding a new flag to make it non-deterministic >

[apcamargo]

That would work. Maybe some short warning in the help message would help making users aware.

[shenwei356]

added

Attention:
  2. By default, the output is deterministic; that is, given the same input and random seed,
     seqkit shuf will always generate identical results across different runs.
     For 'true randomness', please add '-r/--non-deterministic', which uses a time-based seed.

Flags:
  -r, --non-deterministic   use a time-based seed to generate non-deterministic (truly random) results

[apcamargo]

Thank you, @shenwei356 !

[apcamargo]

I just noticed that seqkit sample has the same behaviour. Would you consider adding the parameter and the warning there as well?

[shenwei356]

added

seqkit_linux_amd64.tar.gz
seqkit_darwin_arm64.tar.gz