tree 2653538a7b913d6b294070deedeed7c00010a580
parent d8e30eaef3dd59a3c7efef8fe282eed1c62fbb9b
author M. J. Fromberger <michael.j.fromberger@gmail.com> 1559532961 -0700
committer M. J. Fromberger <michael.j.fromberger@gmail.com> 1559533187 -0700

Benchmark memory more realistically.

Using shell.Split to benchmark produces a lot of noise from the accumulation of
the results into the output slice. To get a more accurate memory profile, use
the scanner directly and discard the tokens as they are processed.

This allows us to (still) count the total memory allocated during the split,
but does not charge it against the wrong code.

This change makes the benchmark look a little bit faster, but not enough to
worry about.