Hunt down bazel build reproducibility issues

hunt.sh

#!/usr/bin/env bash

# usage: ./hunt //my-target [n_runs]

set -euo pipefail

filter='map
    (reduce . as $target
        ([]; . +
            $target.actualOutputs
        )
    ) | flatten'


target="$1"
n_runs="${2:-2}"


run_build() {
    local run_n="$1"

    echo "$(date) Running build $run_n/$n_runs"

    output_base="$PWD/,output-base-$run_n"

    for i in $(seq 1 "$run_n"); do
        output_base="$output_base/nest-$i"
    done


    echo "using output base '$output_base'"
    rm -rf "$output_base"

    # add extra startup_options if necessary
    startup_options=( --output_base="$output_base" )
    
    mkdir -p target # arbitrary dir where execlogs should be placed
    bazel \
        "${startup_options[@]}"
        build \
        "--execution_log_json_file=target/exec$run_n.json" \
        "$target"
}

compare_runs() {

    local run_a="$1"
    local run_b="$2"

    res=$(jq --slurp '(.[0] - .[1])' \
        <(jq -Mr <"target/exec$run_a.json" --slurp "$filter") \
        <(jq -Mr <"target/exec$run_b.json" --slurp "$filter")
    )


    if [ "$res" != "[]" ]; then
        echo "$(date) mismatch"
        echo "$res"
        exit 1
    fi

    echo "$(date) builds $run_a - $run_b: no diff 👌"
}

for run_n in $(seq 1 "$n_runs"); do
    run_build "$run_n"

    if [[ $run_n != "1" ]]; then
        compare_runs "1" "$run_n"
    fi
done