Currently I'm working on a set of benchmarks where I want to save the output of (at least) one run from each. --output should arguably be allowed N times, where N is the number of commands being benchmarked.

However, I would also argue that keeping each runs output is extremely useful, specifically for debugging/figuring out why statistical outliers exist.

As an arbitrary example:

$ hyperfine --output ./example.log "ls" "pwd"

will only show output from a single pwd. Passing multiple outputs does not work, as hyperfine only expects one.

As a proposal for how this could change there are two things that would be nice to see:

1. Allow multiple outputs, the same way --prepare is allowed per command

   $ hyperfine --output ./ls.log "ls" --output ./pwd.log "pwd"
   

2. Allow some parameter, like N, for the run number

   $ hyperfine --output "./ls-{N}.log" "ls"
   

For short running commands like ls this is a problem as it may run somewhere in the neighborhood of 5k times, so it may be a problem to create a new logfile for each run. Longer running commands obviously do not have this problem.