SLeeLa Source Behavioral Test Suite
Purpose
This suite treats every .sleela file in the repository as a test subject. It does not assume that every Wrapper™ is an executable program: source files are classified as either:
- program — contains a
main()entry point and is checked and executed; - library — has no
main()and is checked as a compilable language unit; - expected-rejection — a deliberately invalid regression/version fixture that must be rejected by
sleela check.
For executable Wrappers™, the suite exercises three behavioral dimensions:
- Input — the Wrapper is supplied a deterministic stdin fixture when its source advertises input/read/recv/scan behavior; otherwise it is executed with EOF.
- Output — execution must terminate successfully and produce no
FAILassertion line. - Procedural operations — the real compiler + VM path is used, with an execution timeout, so control flow, calls, returns, and supported procedural built-ins are exercised by the program itself.
The test is deliberately different from a syntax-only inventory: it drives the SLeeLa executable against the repository's actual source corpus.
Coverage target
The repository currently contains approximately 1,000 SLeeLa source files. The suite discovers the corpus dynamically with find, so newly added .sleela files are automatically included.
The generated TSV manifest records one row per source:
path, class, check, run, input, output, procedural, result
This provides a file-level accounting trail rather than only an aggregate test count.
Running
Build the implementation and run the source suite:
make -C impl build/sleela
bash test-suites/sleela-source-behavior.sh
The test log and manifest are written under:
test-suites/logs/sleela-source/
Set SLEELA_SOURCE_TIMEOUT to change the per-program execution timeout. Set SLEELA_BIN to test a different SLeeLa executable.
Meaning of a pass
A pass means the source file met the contract that can be established automatically from the repository:
- valid sources are accepted by the current SLeeLa checker;
- executable sources start through the real compiler/VM path and terminate successfully;
- deterministic input execution does not fail;
- output does not contain an explicit
FAILassertion; - known negative fixtures are rejected.
It does not claim that arbitrary human-level semantics are proven for every library object. Those require object-specific assertions. This suite establishes the corpus-wide executable/checkable baseline and produces the manifest needed to drive deeper per-object behavioral tests.
Extension point
Object-specific expected inputs, outputs, and procedures can be added later without changing the corpus scanner. The intended next layer is a sidecar contract manifest mapping a Wrapper™ to exact input fixtures, expected output, and expected state transitions.