Binary Size Comparison

Binary Size Comparison Workflow

This skill provides the workflow to compare binary sizes between your current branch and a baseline (like the master fork-point).

1. Prerequisites & Tool Installation

A. Containerized Build Setup

You MUST use the podman-vscode-build skill to handle container setup and build execution.

• macOS Note: As per podman-vscode-build, use docker instead of podman on macOS due to stability and proper Rosetta 2 support for the x86 toolchains.

B. Rust Tools Installation (macOS & Linux)

These tools must be installed on your Host machine. They require the Rust toolchain.

If you don't have Rust (cargo) installed, install it first:

# This official installer works for both macOS and Linux
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"

Once Rust is available, install the size comparison tools:

cargo install elf_bloat asm-annotate

2. Platform ELF Locations

Identify your target's ELF path relative to the repo root before you start:

• Linux / POSIX: out/<target>/<app-name>
• ESP32: out/<target>/<app-name>.elf
• Silicon Labs EFR32: out/<target>/<app-name>.out
• nRF Connect (script): examples/<app>/nrfconnect/build/nrfconnect/zephyr/zephyr.elf
• nRF/Telink (build_examples.py): out/<target>/zephyr/zephyr.elf

3. Comparison Workflow

A. Build the Baseline (Master Fork-Point)

Determine the fork-point to isolate only your changes.

CURRENT_COMMIT=$(git rev-parse HEAD)
FORK_POINT=$(git merge-base master HEAD)

git checkout $FORK_POINT
scripts/checkout_submodules.py --shallow --force

# BUILD STEP: Use 'podman-vscode-build' skill here
# Example: docker exec -w /workspace bld_vscode bash -c "source scripts/activate.sh && scripts/examples/nrfconnect_example.sh all-clusters-app nrf52840dk/nrf52840"

mkdir -p out/size-compares/master-baseline
cp <PATH_TO_ELF> out/size-compares/master-baseline/baseline.elf

git checkout $CURRENT_COMMIT
scripts/checkout_submodules.py --shallow --force

B. Build the PR Branch

# BUILD STEP: Use 'podman-vscode-build' skill here
# Example: docker exec -w /workspace bld_vscode bash -c "source scripts/activate.sh && scripts/examples/nrfconnect_example.sh all-clusters-app nrf52840dk/nrf52840"

mkdir -p out/size-compares/pr-branch
cp <PATH_TO_ELF> out/size-compares/pr-branch/updated.elf

C. Run and Analyze Comparison

Run elf-bloat on your Host machine to generate the report.

elf-bloat --compare-base out/size-compares/master-baseline/baseline.elf \
  --viewer none \
  out/size-compares/pr-branch/updated.elf \
  > out/size-compares/size_diff_report.csv

# View top 20 growth contributors (skips CSV header)
tail -n +2 out/size-compares/size_diff_report.csv | sort -t',' -k3 -rn | head -n 20

D. Deep Dive Analysis (Optional)

To find the largest functions in a single binary non-interactively (without TUI), use elf-bloat with the custom viewer to dump a CSV:

elf-bloat out/size-compares/pr-branch/updated.elf --viewer custom:cat > out/size-compares/single_size_report.csv
head -n 20 out/size-compares/single_size_report.csv

If a specific function grew significantly and you want to see exactly which C++ lines generated the bloated assembly, use asm-annotate. It maps the compiled assembly instructions back to your original source code.

CRITICAL RULE FOR AI AGENTS: ALWAYS use the --dump flag when running asm-annotate.

# Dump the assembly to a file for analysis or for pasting into a PR description
asm-annotate out/size-compares/pr-branch/updated.elf "chip::app::CodegenDataModelProvider::InitDataModelForTesting()" --dump > out/size-compares/annotated_function.txt
cat out/size-compares/annotated_function.txt