C

Default: C11, -std=c11 -Wall -Wextra -Wpedantic -O2, one Makefile with all, test, plots (or run) targets, only libm unless a library clearly earns its place.

Layout that stays testable

model.c / model.h   pure state-update functions, no I/O
ui.c                terminal or graphics front end
main.c              argument parsing, wiring
test.c              assertions on model.c, exits non-zero on failure

Public examples of this layout: lift-pilot, current-swimmer, gravity-well, relativity-rider, quantum-wave-hopper, field-ripple (all small C11 physics programs that run in a terminal).

Numerical code

• Pick the integrator for the physics: RK4 for smooth ODEs, symplectic (leapfrog/velocity Verlet) for orbits where energy must not drift, implicit or semi-Lagrangian schemes for stiff or fluid problems.
• Test against analytic solutions, not against previous output: e.g. a phugoid period against Lanchester's formula, a Hohmann transfer delta-v against the textbook value, tunnelling probability against the closed form. Assert a tolerance (e.g. within 0.3 %).
• Check conserved quantities (energy, momentum, probability norm) over long runs; drift reveals step-size or scheme problems.
• Keep units in names or comments (v_mps, t_s).

Headless outputs

• --snapshot prints one frame without a terminal, so tests and docs can capture it.
• --gif out.gif writes a scripted run as an animation; a minimal GIF encoder is a few hundred lines and avoids dependencies.
• make plots writes CSV and SVG charts from a scripted run, so the README images are reproducible.

Command-line utilities

• Parse options with getopt_long; print usage on error and return 2.
• Check every return value of fopen, malloc, read, write; report with perror.
• For programs without libc (raw syscalls), keep the syscall layer in one file and test it through the program's output.

Verification loop

1. Build with zero warnings on GCC and Clang.
2. make test with -fsanitize=address,undefined -g in a debug build.
3. Run the snapshot mode and compare with the expected output.
4. docker build . to prove a clean environment can build and test it.

Pitfalls

• Undefined behaviour (signed overflow, uninitialised reads, out-of-bounds) may only show up at -O2. Sanitizers find most of it.
• float vs double: use double for simulation state; float accumulates error quickly.
• %d with size_t and similar printf mismatches are UB; use %zu, enable -Wformat.
• Terminal state: restore the terminal (raw mode, cursor, colours) on every exit path, including signals.
• ncurses needs -lncurses (and sometimes -ltinfo) and wide-character variants for Unicode; check pkg-config rather than hard-coding flags.
• Integer division truncates toward zero; -7 / 2 == -3.