CsCheck is a C# random testing library inspired by QuickCheck.
It differs in that generation and shrinking are both based on PCG, a fast random number generator.
This gives the following advantages:
- Automatic shrinking. Gen classes are composable with no need for Arb classes. So less boilerplate.
- Random testing and shrinking are parallelized. This and PCG make it very fast.
- Shrunk cases have a seed value. Simpler examples can easily be reproduced.
- Shrinking can be continued later to give simpler cases for high dimensional problems.
- Parallel testing and random shrinking work well together.
CsCheck also makes parallel, performance and regression testing simple and fast.
CsCheck is a C# random testing library inspired by QuickCheck.
It differs in that generation and shrinking are both based on PCG, a fast random number generator.
This gives the following advantages over tree based shrinking libraries:
Automatic shrinking. Gen classes are composable with no need for Arb classes. So less boilerplate.
Random testing and shrinking are parallelized. This and PCG make it very fast.
Shrunk cases have a seed value. Simpler examples can easily be reproduced.
Shrinking can be continued later to give simpler cases for high dimensional problems.
Parallel testing and random shrinking work well together. Repeat is not needed.
See why you should use it, the comparison with other random testing libraries, or how CsCheck does in the shrinking challenge.
In one shrinking challenge test CsCheck managed to shrink to a new smaller example than was thought possible and is not reached by any other testing library.
CsCheck is the only random testing library that can always shrink to the simplest example (given enough time).
CsCheck also has functionality to make multiple types of testing simple and fast:
The following tests are in xUnit TUnit but could equally be used in any testing framework.
More to see in the Tests. There are also 1,000+ F# tests using CsCheck in MKL.NET.
No Reflection was used in the making of this product. CsCheck is close to being AOT compatible but 'generic recursion is AOT kryptonite'.
Generator Creation Example
Use Gen and its Linq methods to compose generators for any type. Here we create a Gen for json documents.
More often it will simply be composing a few primitives and collections.
Don't worry about shrinking as it's automatic and the best in the business.
Sample is used to perform tests with a generator. Either return false or throw an exception for failure.
Sample will aggressively shrink any failure down to the simplest example.
The default sample size is 100 iterations. Set iter: to change this or time: to run for a number of seconds.
Setting these from the command line can be a good way to run your tests in different ways and in Release mode.
Unit Single
[Test]
public void Single_Unit_Range()
{
Gen.Single.Unit.Sample(f => f is >= 0f and <= 0.9999999f);
}
Long Range
[Test]
public void Long_Range()
{
(from t in Gen.Select(Gen.Long, Gen.Long)
let start = Math.Min(t.V0, t.V1)
let finish = Math.Max(t.V0, t.V1)
from value in Gen.Long[start, finish]
select (value, start, finish))
.Sample(i => i.start <= i.value && i.value <= i.finish);
}
Int Distribution
[Test]
public void Int_Distribution()
{
int buckets = 70;
int frequency = 10;
int[] expected = Enumerable.Repeat(frequency, buckets).ToArray();
Gen.Int[0, buckets - 1].Array[frequency * buckets]
.Select(sample => Tally(buckets, sample))
.Sample(actual => Check.ChiSquared(expected, actual));
}
Serialization Roundtrip
static void TestRoundtrip<T>(Gen<T> gen, Action<Stream, T> serialize, Func<Stream, T> deserialize)
{
gen.Sample(t =>
{
using var ms = new MemoryStream();
serialize(ms, t);
ms.Position = 0;
return deserialize(ms).Equals(t);
});
}
[Test]
public void Varint()
{
TestRoundtrip(Gen.UInt, StreamSerializer.WriteVarint, StreamSerializer.ReadVarint);
}
[Test]
public void Double()
{
TestRoundtrip(Gen.Double, StreamSerializer.WriteDouble, StreamSerializer.ReadDouble);
}
[Test]
public void DateTime()
{
TestRoundtrip(Gen.DateTime, StreamSerializer.WriteDateTime, StreamSerializer.ReadDateTime);
}
Shrinking Challenge
[Test]
public void No2_LargeUnionList()
{
Gen.Int.Array.Array
.Sample(aa =>
{
var hs = new HashSet<int>();
foreach (var a in aa)
{
foreach (var i in a) hs.Add(i);
if (hs.Count >= 5) return false;
}
return true;
});
}
Recursive
record MyObj(int Id, MyObj[] Children);
[Test]
public void RecursiveDepth()
{
int maxDepth = 4;
Gen.Recursive<MyObj>((i, my) =>
Gen.Select(Gen.Int, my.Array[0, i < maxDepth ? 6 : 0], (i, a) => new MyObj(i, a))
)
.Sample(i =>
{
static int Depth(MyObj o) => o.Children.Length == 0 ? 0 : 1 + o.Children.Max(Depth);
return Depth(i) <= maxDepth;
});
}
Classify
Change the return in Sample to a string to produce a summary classification table.
All other optional parameters work the same but writeLine: is now mandatory.
[Test]
public void AllocatorMany_Classify()
{
Gen.Select(Gen.Int[3, 30], Gen.Int[3, 15]).SelectMany((rows, cols) =>
Gen.Select(
Gen.Int[0, 5].Array[cols].Where(a => a.Sum() > 0).Array[rows],
Gen.Int[900, 1000].Array[rows],
Gen.Int.Uniform))
.Sample((solution,
rowPrice,
seed) =>
{
var rowTotal = Array.ConvertAll(solution, row => row.Sum());
var colTotal = Enumerable.Range(0, solution[0].Length).Select(col => solution.SumCol(col)).ToArray();
var allocation = AllocatorMany.Allocate(rowPrice, rowTotal, colTotal, new(seed), time: 60);
if (!TotalsCorrectly(rowTotal, colTotal, allocation.Solution))
throw new Exception("Does not total correctly");
return $"{(allocation.KnownGlobal ? "Global" : "Local")}/{allocation.SolutionType}";
}, TUnitX.WriteLine, time: 900);
}
Count
%
Median
Lower Q
Upper Q
Minimum
Maximum
Global
458
50.22%
RoundingMinimum
343
37.61%
2.68ms
0.50ms
10.85ms
0.03ms
190.92ms
EveryCombination
87
9.54%
173.99ms
16.80ms
1,199.64ms
0.20ms
42,257.35ms
RandomChange
28
3.07%
59,592.98ms
55,267.94ms
59,901.58ms
38,575.41ms
60,107.64ms
Local
454
49.78%
RoundingMinimum
301
33.00%
60,000.12ms
60,000.04ms
60,003.70ms
60,000.02ms
60,144.84ms
RandomChange
90
9.87%
60,000.06ms
60,000.03ms
60,004.41ms
60,000.02ms
60,136.59ms
EveryCombination
63
6.91%
60,000.10ms
60,000.03ms
60,001.29ms
60,000.01ms
60,019.36ms
Model-based testing
Model-based is the most efficient form of random testing.
Only a small amount of code is needed to fully test functionality.
SampleModelBased generates an initial actual and model and then applies a random sequence of operations to both checking that the actual and model are still equal.
SetSlim Add
[Test]
public void
SetSlim_ModelBased()
{
Gen.Int.Array.Select(a => (new SetSlim<int>(a), new HashSet<int>(a)))
.SampleModelBased(
Gen.Int.Operation<SetSlim<int>, HashSet<int>>(
(ss, i) => ss.Add(i),
(hs, i) => hs.Add(i)
)
// ... other operations
);
}
Metamorphic testing
The second most efficient form of random testing is metamorphic which means doing something two different ways and checking they produce the same result.
SampleMetamorphic generates two identical initial samples and then applies the two functions and asserts the results are equal.
This can be needed when no model can be found that is not just a reimplementation.
More about how useful metamorphic tests can be here: How to specify it!.
CsCheck has support for parallel testing with full shrinking capability.
A number of operations are run sequentially and then a number in parallel on an initial state and the result is compared to all the possible linearized versions.
At least one of these must be equal to the parallel result.
Idea from John Hughes talk and paper. This is easier to implement with CsCheck than QuickCheck because the random shrinking does not need to repeat each step as QuickCheck does (10 times by default) to make shrinking deterministic.
Causal profiling is a technique to investigate the effect of speeding up one or more concurrent regions of code.
It shows which regions are the bottleneck and what overall performance gain could be achieved from each region.
Idea from Emery Berger. My blog posts on this here.
[Test]
public void Fasta()
{
Causal.Profile(() => FastaUtils.Fasta.NotMain(10_000_000, null)).Output(writeLine);
}
static int[] Rnds(int i, int j, ref int seed)
{
var region = Causal.RegionStart("rnds");
var a = intPool.Rent(BlockSize1);
var s = a.AsSpan(0, i);
s[0] = j;
for (i = 1, j = Width; i < s.Length; i++)
{
if (j-- == 0)
{
j = Width;
s[i] = IM * 3 / 2;
}
else
{
s[i] = seed = (seed * IA + IC) % IM;
}
}
Causal.RegionEnd(region);
return a;
}
Regression testing
Portfolio Calculation
Single is used to find, pin and continue to check a suitable generated example e.g. to cover a certain codepath. Hash is used to find and check a hash for a number of results.
It saves a temp cache of the results on a successful hash check and each subsequent run will fail with actual vs expected at the first point of any difference.
Together Single and Hash eliminate the need to commit data files in regression testing while also giving detailed information of any change.
[Test]
public void Portfolio_Small_Mixed_Example()
{
var portfolio = ModelGen.Portfolio.Single(p =>
p.Positions.Count == 5
&& p.Positions.Any(p => p.Instrument is Bond)
&& p.Positions.Any(p => p.Instrument is Equity)
, "0N0XIzNsQ0O2");
var currencies = portfolio.Positions.Select(p => p.Instrument.Currency).Distinct().ToArray();
var fxRates = ModelGen.Price.Array[currencies.Length].Single(a =>
a.All(p => pp is > 0.75 and < 1.5)
, "ftXKwKhS6ec4");
double fxRate(Currency c) => fxRates[Array.IndexOf(currencies, c)];
Check.Hash(h =>
{
h.Add(portfolio.Positions.Select(p => p.Profit));
h.Add(portfolio.Profit(fxRate));
h.Add(portfolio.RiskByPosition(fxRate));
}, 5857230471108592669, decimalPlaces: 2);
}
Performance testing
Faster is used to statistically test that the first method is faster than the second and some condition is satisfied (by default equality of the output of the two methods).
Since it's statistical and relative you can run it as a normal test anywhere e.g. across multiple platforms on a continuous integration server.
It's fast because it runs in parallel and knows when to stop.
It's just what you need to iteratively improve performance while making sure it still produces the correct results.
[Test]
public void Faster_Linq_Random()
{
Gen.Byte.Array[100, 1000]
.Faster(
data => data.Aggregate(0.0, (t, b) => t + b),
data => data.Select(i => (double)i).Sum(),
writeLine: TUnitX.WriteLine
);
}
The performance is raised in an exception if it fails but can also be output if it passes with the above output function.
Tests.CheckTests.Faster_Linq_Random [27ms]
Standard Output Messages:
32.29%[29.47%..36.51%] 1.48x[1.42x..1.58x] faster, sigma = 50.0 (2,551 vs 17), min = 208ns vs 375ns
The first number is the estimated percentage median performance improvement with the interquartile range in the square brackets.
The second number is the estimated times median performance improvement with the interquartile range in the square brackets.
33⅓% faster = 1.5x faster and 90% faster = 10x faster, take your pick.
The counts of faster vs slower and the corresponding sigma (the number of standard deviations of the binomial
distribution for the null hypothesis P(faster) = P(slower) = 0.5) are also shown. The default sigma used is 6.0.
Finally the minimum time taken for faster vs slower is shown as an idea of timing and for additional diagnostics of the result (machine dependent).
Matrix Multiply
[Test]
public void Faster_Matrix_Multiply_Range()
{
var genDim = Gen.Int[5, 30];
var genArray = Gen.Double.Unit.Array2D;
Gen.SelectMany(genDim, genDim, genDim, (i, j, k) => Gen.Select(genArray[i, j], genArray[j, k]))
.Faster(
MulIKJ,
MulIJK
);
}
MapSlim Increment
[Test]
public void MapSlim_Performance_Increment()
{
Gen.Byte.Array
.Select(a => (a, new MapSlim<byte, int>(), new Dictionary<int, int>()))
.Faster(
(items, mapslim, _) =>
{
foreach (var b in items)
mapslim.GetValueOrNullRef(b)++;
},
(items, _, dict) =>
{
foreach (var b in items)
{
dict.TryGetValue(b, out int c);
dict[b] = c + 1;
}
},
repeat: 100,
writeLine: TUnitX.WriteLine);
}
Tests.SlimCollectionsTests.MapSlim_Performance_Increment [27 s]
Standard Output Messages:
66.02%[56.48%..74.81%] 2.94x[2.30x..3.97x] faster, sigma = 200.0 (72,690 vs 13,853), min = 11.983ns vs 34.996ns
Tests.ReverseComplementTests.ReverseComplement_Faster [27s 870ms]
Standard Output Messages:
25.15%[20.58%..31.60%] 1.34x[1.26x..1.46x] faster, sigma = 6.0 (36 vs 0), min = 1,453ms vs 1,942ms
Tests.ArraySerializerTests.Varint_Faster [45 ms]
Standard Output Messages:
10.94%[-3.27%..25.81%] 1.12x[0.97x..1.35x] faster, sigma = 10.0 (442 vs 190), min = 7.082ns vs 7.332ns
Debug utilities
The Dbg module is a set of utilities to collect, count and output debug info, time, classify generators, define and remotely call functions, and perform in code regression during testing.
CsCheck can temporarily be added as a reference to run in non test code.
Note this module is only for temporary debug use and the API may change between minor versions.
Count, Info, Set, Get, CallAdd, Call
public void Normal_Code(int z)
{
Dbg.Count();
var d = Calc1(z).DbgSet("d");
Dbg.Call("helpful");
var c = Calc2(d).DbgInfo("c");
Dbg.CallAdd("test cache", () =>
{
Dbg.Info(Dbg.Get("d"));
Dbg.Info(cacheItems);
});
}
[Test]
public void Test()
{
Dbg.CallAdd("helpful", () =>
{
var d = (double)Dbg.Get("d");
// ...
Dbg.Set("d", d);
});
Normal_Code(z);
Dbg.Call("test cache");
Dbg.Output(writeLine);
}
Regression
public double[] Calculation(InputData input)
{
var part1 = CalcPart1(input);
// Add items to the regression on first pass, throw/break here if different on subsequent.
Dbg.Regression.Add(part1);
var part2 = CalcPart2(part1).DbgTee(Dbg.Regression.Add); // Tee can be used to do this inline.
// ...
return CalcFinal(partN).DbgTee(Dbg.Regression.Add);
}
[Test]
public void Test()
{
// Remove any previously saved regression data.
Dbg.Regression.Delete();
Calculation(InputSource1());
// End first pass save mode (only needed if second pass is in this process run).
Dbg.Regression.Close();
// Subsequent pass could be now or a code change and rerun (without the Delete).
Calculation(InputSource2());
// Check full number of items have been reconciled (optional).
Dbg.Regression.Close();
}
Time
public Result CalcPart2(InputData input)
{
using var time = Dbg.Time();
// Calc
time.Line();
// Calc more
time.Line();
// ...
return result;
}
public void LongProcess()
{
using var time = Dbg.Time();
var part1 = CalcPart1(input);
time.Line();
var part2 = new List<Result>();
foreach(var item in part1)
part2.Add(CalcPart2(item));
time.Line();
// ...
return CalcFinal(partN);
}
[Test]
public void Test()
{
LongProcess();
Dbg.Output(writeLine);
}
Logging
CsCheck now supports logging types and pass and fail results for analysis in Sample. We include a Tyche logging implementation.
Configuration
Check functions accept configuration optional parameters e.g. iter: 100_000, seed: "0N0XIzNsQ0O2", print: t => string.Join(", ", t):
iter - The number of iterations to run in the sample (default 100).
time - The number of seconds to run the sample.
seed - The seed to use for the first iteration.
threads - The number of threads to run the sample on (default number logical CPUs).
timeout - The timeout in seconds to use for Faster (default 60 seconds).
print - A function to convert the state to a string for error reporting (default Check.Print).
equal - A function to check if the two states are the same (default Check.Equal).
sigma - For Faster sigma is the number of standard deviations from the null hypothesis (default 6).
replay - The number of times to retry the seed to reproduce a SampleParallel fail (default 100).
Global defaults can also be set via environment variables:
