Detect a peak RSS regression

A change made the program hold much more memory at its peak, and you want the pull request to fail.

# yaml-language-server: $schema=../../schema/himorime.schema.json
#
# Recipe: detect a peak RSS regression.
# https://nao1215.github.io/himorime/cookbook/#detect-a-peak-rss-regression
#
#   himorime compare --against main examples/memory-regression
version: "1"

name: memory regression
description: Peak resident set size of the word counter, compared between two revisions.

build:
  command: [go, build, -o, "${artifact}", ../tools/wordcount]

defaults:
  warmup: 1
  runs: 12

benchmarks:
  - name: count 200k lines
    setup:
      - command: ["${artifact}", -gen, "200000", -o, "${workdir}/input.txt"]
    metrics:
      memory: true
    commands:
      wordcount:
        command: ["${artifact}", "${workdir}/input.txt"]
        budget:
          peak_rss:
            max: "<= 512MiB"
    regression:
      latency:
        max_percent: 50
      peak_rss:
        max_percent: 20
        min_difference: 4MiB

$ himorime compare --against main examples/memory-regression

A peak rss comparison table in MiB. When the head holds clearly more memory the row says REGRESSION and himorime exits 1. The end-to-end suite proves it by switching the word counter’s default implementation to one that reads the whole input into memory.

Peak RSS is the largest resident set size of one process in the tree, not a runtime heap or a sum of concurrent processes.
min_difference filters small changes; budget statistic meanings are in Configuration.
On Unix a peak RSS cannot be measured below a floor of a few MiB, the memory of the process that starts the command. A streaming base at the floor still shows a regression when the head clearly exceeds the floor; see The floor.

Example: examples/memory-regression