# CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

## Project

SDLamp2 is a simple SDL2-based audio player written in C, designed for a child to use. It plays long m4a/mp3 files (fairy tale cassette rips) with a cassette-player-like UI: rewind, stop, play, fast-forward, next tape. Inspired by Winamp.

See `docs/sdlamp2-fsd.md` for the full functional specification and changelog.
See `docs/TOOLCHAIN.md` for target device details and the arm64 Docker build container.

## Build and Run

**Dependencies:** SDL2, SDL2_image, FFmpeg libraries (libavformat, libavcodec, libavutil, libswresample). Installed via system package manager (native) or via apt inside the arm64 Docker container.

```sh
make                    # native build — uses pkg-config
make clean              # remove build artifacts
./build/sdlamp2 [audio_directory]
```

Building for the arm64 target device via the Docker container (from the `docker-arm64/` directory):

```sh
make build              # one-shot: build sdlamp2 inside arm64 container
make shell              # interactive shell inside the arm64 container
make clean              # remove the Docker image
```

The controls spritesheet (`controls.png`) is embedded in the binary. If an external `controls.png` exists in the current working directory it takes precedence, allowing custom skins. Audio directory defaults to cwd if not specified.

No test suite, no linter.

## Architecture

Single-file C program: `src/sdlamp2.c` (~650 lines). One generated header: `src/controls_png.h` (embedded PNG byte array — regenerate with `./tools/embed_png.py assets/controls.png src/controls_png.h` if the spritesheet changes).

Key sections in order:
- **Decoder struct** — holds all FFmpeg state (format/codec contexts, swr resampler, album art texture)
- **Position persistence** — `save_position()`/`load_position()` read/write a tab-separated `positions.txt` in the audio directory
- **File scanning** — `scan_audio_files()` finds supported files (.m4a, .mp3, .wav, .ogg) in a directory
- **Decoder lifecycle** — `decoder_open()` sets up FFmpeg + swr pipeline (planar float → interleaved stereo 48kHz), `decoder_pump()` feeds decoded audio into an SDL_AudioStream, `decoder_seek()` handles seeking
- **File switching** — `switch_file()` saves position, opens new file, restores saved position
- **Main loop** — event handling (mouse clicks on button rects), decoder pumping, SDL audio queue draining, EOF handling, rendering (album art, progress bar, control buttons from sprite sheet)

Audio pipeline: FFmpeg decoder → libswresample (format conversion) → SDL_AudioStream (FIFO) → SDL audio device queue (push model, no callback).

Uses SDL2 (not SDL3). Uses `#if LIBAVUTIL_VERSION_INT` preprocessor checks to support both old (channel_layout) and new (ch_layout) FFmpeg channel layout APIs.

## Conventions

- C formatted with Google style, `ColumnLimit: 100`
- Keep it simple — Casey Muratori's semantic compression; don't refactor too soon
- Minimize dependencies; discuss with owner before adding any
- Non-fatal errors go to stderr and continue; fatal errors (SDL init failures) abort via `panic_and_abort()`
- Update the changelog in `docs/sdlamp2-fsd.md` when making changes