Picard file loading benchmark script (PICARD-3281)

bench_load.sh

#!/bin/bash
# Benchmark file loading performance across Picard commits.
# Usage: ./scripts/bench_load.sh /path/to/music/directory ref1 [ref2 ...]
#
# Examples:
#   ./scripts/bench_load.sh ~/Music f14c46c upstream/master HEAD
#   ./scripts/bench_load.sh ~/Music f14c46c a25fffa~5 a25fffa
#
# Requirements: run from the picard source directory with a working venv.

set -e

LC_NUMERIC=C
export LC_NUMERIC

MUSIC_DIR="${1:?Usage: $0 /path/to/music/directory ref1 [ref2 ...]}"
MUSIC_DIR=$(realpath "$MUSIC_DIR")
shift

if [ $# -lt 1 ]; then
    echo "Error: provide at least one commit ref"
    exit 1
fi

if [ ! -d "$MUSIC_DIR" ]; then
    echo "Error: $MUSIC_DIR is not a directory"
    exit 1
fi

FILE_COUNT=$(find "$MUSIC_DIR" -type f \( -name "*.flac" -o -name "*.mp3" -o -name "*.ogg" -o -name "*.wav" -o -name "*.m4a" -o -name "*.wma" \) | wc -l)
echo "=== Picard Load Benchmark ==="
echo "Directory: $MUSIC_DIR"
echo "Audio files: $FILE_COUNT"
echo ""

CURRENT_BRANCH=$(git rev-parse --abbrev-ref HEAD)
LOGDIR=$(mktemp -d)

run_test() {
    local ref="$1"
    local short=$(git rev-parse --short "$ref" 2>/dev/null || echo "$ref")
    local label="$short"
    local log="$LOGDIR/${label}.log"
    local marker="MainPanel sort=True"

    printf "  %-40s " "$ref ($short)"

    # Checkout the version to test
    git checkout "$ref" -- picard/tagger.py picard/cluster.py 2>/dev/null
    git checkout "$ref" -- picard/debug_opts.py 2>/dev/null || true

    # Start Picard
    python -u tagger.py --no-restore --debug --debug-opts=timings -e "LOAD $MUSIC_DIR" > "$log" 2>&1 &
    local pid=$!

    # Wait for loading to complete
    local elapsed=0
    while ! grep -q "$marker" "$log" 2>/dev/null; do
        sleep 0.5
        elapsed=$((elapsed + 1))
        if [ $elapsed -gt 240 ]; then
            printf "TIMEOUT\n"
            kill $pid 2>/dev/null; wait $pid 2>/dev/null
            git checkout "$CURRENT_BRANCH" -- picard/tagger.py picard/cluster.py picard/debug_opts.py 2>/dev/null || true
            return
        fi
    done
    sleep 0.5

    # Quit via remote command
    python tagger.py -e "QUIT" 2>/dev/null || true
    wait $pid 2>/dev/null

    # Extract timings
    local start=$(grep "Adding files" "$log" | head -1 | grep -oP '\d+:\d+:\d+,\d+')
    local end=$(grep "$marker" "$log" | head -1 | grep -oP '\d+:\d+:\d+,\d+')

    if [ -n "$start" ] && [ -n "$end" ]; then
        local s_sec=$(echo "$start" | awk -F'[,:]' '{print $1*3600 + $2*60 + $3 + $4/1000}')
        local e_sec=$(echo "$end" | awk -F'[,:]' '{print $1*3600 + $2*60 + $3 + $4/1000}')
        local duration=$(echo "$e_sec - $s_sec" | bc)
        printf "%6.1fs\n" "$duration"
    else
        printf "FAILED\n"
    fi

    # Restore
    git checkout "$CURRENT_BRANCH" -- picard/tagger.py picard/cluster.py picard/debug_opts.py 2>/dev/null || true
}

echo "Running benchmarks..."
echo ""
for ref in "$@"; do
    run_test "$ref"
done

echo ""
echo "Logs: $LOGDIR"