cpl stress
run against a brute on random inputs; shrink the failure
usage: cpl stress <file.cp> [options]
Runs the solution and a brute on random inputs made from the file's
`read()` types, stops at the first case they disagree on, shrinks it, and
saves it to <file>.tests/ as a regression test.
The brute, the checker and the interactor are the file's own `@brute(main)`
(or `@alt(program)`), `@checker` (or `@check`) and `@interactor` fns unless
an option names a program: a .cpp file (built with g++), a .py file, or an
executable.
options:
--brute PROG the slow, obviously right solution to compare with
--fast PROG run this instead of building the file
--gen PROG make inputs with `PROG <seed> <size>` instead of from the
file's `@gen fn` or its types; such a case is not shrunk
--checker PROG judge with `PROG <input> <output> <answer>`; exit 0 accepts
--interactor PROG judge by talking to `PROG <input>`; exit 0 accepts
--cases N how many cases to run (default 1000)
--size N the largest size (n, q) a case uses (default 8)
--seed N the first seed (default 1)
--time-limit MS per run, in milliseconds (default 5000)
--jobs N cases run at once (default: one per core)
--trace on a failure, run the solution and the brute on the
shrunk case in the debug interpreter and save the two
traces side by side as <file>.tests/stress-<seed>.trace.html
(the solution's alone, when a `@checker` is the judge)
--explain on a failure, run the solution and its `@brute(main)` on
the shrunk case in the debug interpreter and name the
first value they give a variable differently, with its line
--format json print the result as JSON, for an editorGenerated from cpl 0.2.2 (dev, a0d0078e).