#!/bin/bash # min-stack.sh - Smallest main-thread stack that runs a SQL script # # Usage: # scripts/stack/min-stack.sh [-b BINARY]... [-S] [-e SQL]... [SQL_FILE]... # # -b BINARY tursodb binary to measure; repeat to compare binaries # (default: target/release/tursodb) # -S also measure the sqlite3 on PATH, as a reference # -e SQL SQL text to run; may be repeated # SQL_FILE SQL script to run; may be repeated # # Prints one row per script and one column per binary, in KiB. # # Examples: # scripts/stack/min-stack.sh -S upsert.sql # scripts/stack/min-stack.sh -b /tmp/tursodb-main -b target/release/tursodb -e "SELECT 1;" *.sql # # Each binary runs the script on an in-memory database with the stack limited # by `ulimit -s`. The CLI runs statements on the main thread, so the limit # applies to them. A run passes if its exit code and output equal those of a # run with the largest stack the shell allows; this way a SQL error is not # mistaken for a stack overflow. The result is a binary search to 4 KiB. # "crash" means the script crashes even with the largest stack. # # The number includes everything the process needs before the first # statement (for example, SELECT 1 alone needs some stack). Compare against # that baseline, or against sqlite3 with -S. Use a release build: debug frames # are much larger. set -uo pipefail usage() { sed -n '2,/^$/p' "$0" | sed 's/^# \{0,1\}//' exit "${1:-0}" } binaries=() with_sqlite=0 texts=() while getopts "b:Se:h" opt; do case "$opt" in b) binaries+=("$OPTARG") ;; S) with_sqlite=1 ;; e) texts+=("$OPTARG") ;; h) usage 0 ;; *) usage 1 ;; esac done shift $((OPTIND - 1)) files=("$@") if [ ${#binaries[@]} -eq 0 ]; then binaries=(target/release/tursodb) fi for bin in "${binaries[@]}"; do if [ ! -x "$bin" ]; then echo "error: $bin is not an executable" >&2 exit 1 fi done kinds=() for _ in "${binaries[@]}"; do kinds+=(tursodb); done if [ "$with_sqlite" -eq 1 ]; then if ! command -v sqlite3 > /dev/null; then echo "error: -S needs sqlite3 on PATH" >&2 exit 1 fi binaries+=("$(command -v sqlite3)") kinds+=(sqlite3) fi if [ ${#files[@]} -eq 0 ] && [ ${#texts[@]} -eq 0 ]; then usage 1 fi workdir=$(mktemp -d) trap 'rm -rf "$workdir"' EXIT scripts=() labels=() for f in "${files[@]+"${files[@]}"}"; do if [ ! -f "$f" ]; then echo "error: $f does not exist" >&2 exit 1 fi scripts+=("$f") labels+=("$(basename "$f")") done i=0 for text in "${texts[@]+"${texts[@]}"}"; do i=$((i + 1)) printf '%s\n' "$text" > "$workdir/e$i.sql" scripts+=("$workdir/e$i.sql") label="$text" if [ ${#label} -gt 30 ]; then label="${label:0:27}..."; fi labels+=("$label") done max_kib=$(ulimit -H -s) if [ "$max_kib" = "unlimited" ] || [ "$max_kib" -gt 1048576 ]; then max_kib=1048576 fi # Prints the exit code and output of one run with the given stack size. run() { local kind="$1" bin="$2" script="$3" kib="$4" out code if [ "$kind" = sqlite3 ]; then out=$( (ulimit -s "$kib" && "$bin" :memory: < "$script") 2>&1) else out=$( (ulimit -s "$kib" && "$bin" -q -m list :memory: < "$script") 2>&1) fi code=$? printf '%s\n%s' "$code" "$out" } measure() { local kind="$1" bin="$2" script="$3" expected lo hi mid expected=$(run "$kind" "$bin" "$script" "$max_kib") if [ "$(head -n 1 <<< "$expected")" -gt 128 ]; then echo crash return fi lo=0 hi=$max_kib while [ $((hi - lo)) -gt 4 ]; do mid=$(((lo + hi) / 2)) if [ "$(run "$kind" "$bin" "$script" "$mid")" = "$expected" ]; then hi=$mid; else lo=$mid; fi done echo "$hi" } width=6 for label in "${labels[@]}"; do if [ ${#label} -gt $width ]; then width=${#label}; fi done names=() for bin in "${binaries[@]}"; do name=$(basename "$bin") if [ ${#name} -lt 6 ]; then name=$(printf '%6s' "$name"); fi names+=("$name") done printf "%-${width}s" "script" for name in "${names[@]}"; do printf ' %s' "$name"; done printf '\n' for s in "${!scripts[@]}"; do printf "%-${width}s" "${labels[$s]}" for b in "${!binaries[@]}"; do result=$(measure "${kinds[$b]}" "${binaries[$b]}" "${scripts[$s]}") printf " %${#names[$b]}s" "$result" done printf '\n' done