Step 1: Gather bench data

With JSON set to true in src/main.rs run ./run.sh | jq --slurp "sort_by(.threads)" > comparison_data.json to benchmark all pairs of SHOULD_DROP and TOTAL_THREAD_COUNT, you may need to adjust the TOTAL_THREAD_COUNT_VALUES variable, if you don't have 12 threads or want to benchmark a different configuraiton.

The jq option --slurp turns the JSON Lines coming from our program into a normal JSON array, which is then sorted.

NOTE: It may also be necessary to remove the no option from SHOULD_DROP_VALUES, since it uses quite a bit of memory with the default amount of samples.

Step 2: Transform to CSV for further processing

Libreoffice calc does not accept JSON files, so we use the python script to_csv.py to transform the data to CSV. The command is python3 to_csv.py < comparison_data.json > comparison_data.csv. This also deduplicates the date. Mainly, because of the repeated testing of Rc<str>.

Be ware that the script requires pandas.

Step 3: Use the data

Libreoffice Calc or Excel in my case