Ccusage Rust Profile

ccusage Rust Profile

Use this skill for native CLI performance work. Use bun-cpu-profile only for TypeScript launcher, benchmark, or packaging scripts.

Preparation

Read the relevant local Rust Performance Book pages before non-trivial optimization. Locate the clone instead of assuming a machine-specific path:

set perf_book_dir (ghq list --full-path nnethercote/perf-book | head -n 1)
sed -n '1,220p' "$perf_book_dir/src/profiling.md"
sed -n '1,220p' "$perf_book_dir/src/io.md"
sed -n '1,220p' "$perf_book_dir/src/heap-allocations.md"
sed -n '1,220p' "$perf_book_dir/src/parallelism.md"
sed -n '1,220p' "$perf_book_dir/src/type-sizes.md"

If there is no local clone, use the hosted Rust Performance Book as fallback: https://nnethercote.github.io/perf-book/.

Build release binaries before timing:

direnv exec . cargo build --manifest-path rust/Cargo.toml --release --bin ccusage

Compare End To End

Create a separate main worktree for branch-vs-main comparisons:

command git fetch origin main
command git worktree add /tmp/ccusage-main origin/main
direnv exec . cargo build --manifest-path rust/Cargo.toml --release --bin ccusage
cd /tmp/ccusage-main
direnv exec . cargo build --manifest-path rust/Cargo.toml --release --bin ccusage

Measure real commands with deterministic output settings:

hyperfine --warmup 4 --runs 10 --shell none \
	"env LOG_LEVEL=0 COLUMNS=200 NO_COLOR=1 TZ=UTC rust/target/release/ccusage daily --offline --json" \
	"env LOG_LEVEL=0 COLUMNS=200 NO_COLOR=1 TZ=UTC /tmp/ccusage-main/rust/target/release/ccusage daily --offline --json" \
	--export-json /tmp/ccusage-rust-hyperfine.json

For JSON parity, write both outputs and validate with jq:

env LOG_LEVEL=0 COLUMNS=200 NO_COLOR=1 TZ=UTC rust/target/release/ccusage daily --offline --json >/tmp/head.json
env LOG_LEVEL=0 COLUMNS=200 NO_COLOR=1 TZ=UTC /tmp/ccusage-main/rust/target/release/ccusage daily --offline --json >/tmp/main.json
jq -e . /tmp/head.json >/dev/null
jq -e . /tmp/main.json >/dev/null

What To Check

• I/O count and buffering before CPU-only tweaks.
• Avoid unnecessary String allocation and cloning on hot paths; prefer borrowed &str, Arc<str>, or typed summaries where ownership is needed.
• Avoid returning large intermediate object vectors when aggregation can happen earlier without changing output.
• Use parallelism only when it improves end-to-end command time on real fixture shapes.
• Keep binary size visible when adding dependencies or enabling features.

Validation

Profile before committing an optimization. Validate with end-to-end hyperfine and JSON/table parity, not only microbenchmarks.

When CI performance comments are relevant, inspect options with --help, but do not treat help output as a profiling workload:

nix develop --command pnpm exec bun apps/ccusage/scripts/compare-pr-performance.ts --head-runtime rust --help

To reproduce the workflow shape locally, pass real fixture and worktree inputs:

nix develop --command pnpm exec bun apps/ccusage/scripts/compare-pr-performance.ts \
	--base-dir /tmp/ccusage-main \
	--head-dir "$PWD" \
	--head-runtime rust \
	--fixture-dir apps/ccusage/test/fixtures/claude \
	--codex-fixture-dir apps/ccusage/test/fixtures/codex \
	--runs 1 \
	--warmup 0 \
	--output /tmp/ccusage-rust-perf-comment.md