make
bash tests/run.sh

On Windows, without WSL/Git Bash:

mingw32-make -f Makefile.win
powershell -File tests\run.ps1

run.ps1 mirrors run.sh (same section headers, same pass/fail summary line, same exit code convention) and works on Windows PowerShell 5.1 as well as PowerShell 7+ on any OS. It decodes output as UTF-8, normalizes line endings, closes stdin for each test, and links the unit tests with -lws2_32 and an 8 MiB stack on Windows, as Makefile.win does.

Runs, in order: tests/unit/ (standalone C++ unit tests, compiled and run directly — currently the regex engine's own test suite, which can exercise it without going through the whole language pipeline), tests/cases/ (exact output diff), tests/errors/ (must fail with a specific error code), examples/ (exact output diff where a .expected exists, otherwise just checks exit 0 — used for the one example with genuinely random output), and examples/error_cases/ (must fail with a specific error code).

Drop a .cpp file with a main() in tests/unit/. It's compiled with -I. from the repo root, so include engine headers by path (#include "engine/regex/RegexEngine.h"). Exit non-zero to signal failure; the runner prints your last line of output either way.

Drop a .cuff file in tests/cases/, then generate its golden output:

./cuffc tests/cases/my_case.cuff > tests/cases/my_case.expected

Read the output once before committing it — this locks in whatever the interpreter did as "correct". Don't add a case whose output is non-deterministic (random numbers, timestamps).

Drop a .cuff file in tests/errors/ that's expected to fail, then record which error code it must produce:

./cuffc tests/errors/my_error.cuff   # note the [E####] in the output
echo "E####" > tests/errors/my_error.expected_code

.expected_code is optional — without one, the runner only checks that the script fails (exit code != 0).