FR
importPour l'utiliser : import Data.Test

Data.Test — the Property monad (do-style property tests)

Synopsis 
data Trace = MkTrace (List String) (List String) (Maybe Int)   -- inputs, labels, trials asked

What a single run accumulates: the rendered forAll inputs (for the failure report), the statistics labels (for the distribution summary), and the number of trials the property ASKS FOR (withTests), when it asks.

data Outcome a
  = Done a
  | Failed String

The result of running a property's assertions: a value, or the first failure.

Done

the trial finished with a value

Failed

an assertion failed, with its message

data Property a = MkProp (Gen (Pair Trace (Outcome a)))

Property a = Gen (Trace, Outcome a): a generator (carrying the shrink tree) of a traced outcome.

forAll : Show a => Gen a -> Property a

Draw a value from a generator, recording it (via show) for the failure report. The generator's shrink tree becomes the property's, so the drawn value shrinks toward its generator's origin on failure.

withTests : Int -> Property ()

Ask for n trials of this property, whatever the suite's default (oppi test --tests, [test] property_tests, or the pinned 100). A statement anywhere in the block; the last one written wins. n must be at least 1: a property is checked by running it.

(===) : Eq a => Show a => a -> a -> Property ()

Assert two values equal; on failure the report shows both sides.

claimHolds : String -> a -> a -> Property ()

Judge the claim x ≡ y of a specification (conn: the connector as written, quoted when the two sides cannot be compared). On failure the report shows both sides, printed by the runner.

claimInput : Gen a -> Property a

Draw the input of a specification's ∀, recorded for the failure report by the runner's printer — like forAll, without asking for Show.

sameSet : Ord a => Eq a => Show a => List a -> List a -> Property ()

Equal up to permutation ("à permutation près"): same elements, same multiplicities, order ignored. Use when a function's result order is unspecified (e.g. the set of colours / vertices of a graph).

withinAbs : Double -> Double -> Double -> Property ()

Equal within an ABSOLUTE tolerance ε: |a - b| ≤ ε.

withinRel : Double -> Double -> Double -> Property ()

Equal within a RELATIVE tolerance p of the reference: |a - b| ≤ p·|b| (so withinRel 0.01 is "within 1 %").

satisfies : Show a => a -> (a -> Bool) -> Property ()

The value satisfies a predicate (for specs with no unique answer). Reads well infix: result satisfies isSorted.

classify : String -> Bool -> Property ()

Tag this run with lbl when cond holds — the summary reports the % of runs so tagged (so you can see the distribution your generators produce).

collect : Show a => a -> Property ()

Tag this run with show x — collect a distribution of values.

data Check =
  Running Int (List (Pair String Int)) (Maybe Int)
  | Finished CheckResult

The state of a check between two trials: how many trials have passed, the label tally so far, and the number of trials the property asked for (Nothing until a trial says, via withTests) — or the final result.

Running

trials passed so far, label tally, trials asked

Finished

the check is over

checkStart : Check

Where every check starts.

checkStep : Int -> Int -> Int -> Property a -> Check -> Check

ONE trial of prop — the single authority for what a trial is: draw the inputs at size from the seed of trial t (mixed from base), run the assertions, tally the labels, and on the first falsifying trial shrink it to a minimal failing input. checkProperty below is nothing but the iteration of this step; the test runner (oppi test, the browser engine) iterates the SAME step from outside, one evaluation per trial, so the evaluator's divergence guard (step and time budgets) applies PER TRIAL: a trial that loops costs one budget, never the budget of all the trials.

numTests is the SUITE's count (the runner's cascade); a property that asks for its own (withTests) is heard from its first trial on, and its count replaces the suite's for the rest of the check.

checkProperty : Int -> Int -> Int -> Property a -> CheckResult

Run prop up to numTests times at size, deterministically seeded from base. On the first falsifying trial, shrink it to a minimal failing input and stop; otherwise report how the labels were distributed.