Optimize BeEquivalentTo

[jnyrup]

In #1932 I noticed that we generate the equivalency summary even when a test does not fail.
That didn't look right to me.

Benchmark 1: Comparing two simple objects

[MemoryDiagnoser]
[SimpleJob(RuntimeMoniker.Net472)]
[SimpleJob(RuntimeMoniker.Net60)]
public class BeEquivalentToBenchmarks
{
    public class MyClass
    {
        public int MyProperty { get; set; }
    }

    private readonly MyClass a = new();
    private readonly MyClass b = new();

    [Benchmark]
    public AndConstraint<ObjectAssertions> BeEquivalentTo() => a.Should().BeEquivalentTo(b);
}

develop

Method	Runtime	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	8.888 us	0.1760 us	0.2524 us	2.0447	0.0153	13 KB
BeEquivalentTo	.NET Framework 4.7.2	11.181 us	0.1033 us	0.0915 us	2.1820	0.0153	13 KB

373a6ad

Method	Runtime	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	7.830 us	0.0923 us	0.0863 us	1.7090	0.0153	11 KB
BeEquivalentTo	.NET Framework 4.7.2	10.427 us	0.1348 us	0.1195 us	1.8463	0.0153	11 KB

6c5db11

Method	Runtime	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	6.972 us	0.0881 us	0.0824 us	1.5564	0.0153	10 KB
BeEquivalentTo	.NET Framework 4.7.2	10.023 us	0.1090 us	0.1020 us	1.6785	0.0153	10 KB

72c2e25

Method	Runtime	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	7.239 us	0.0411 us	0.0385 us	1.4954	0.0153	9 KB
BeEquivalentTo	.NET Framework 4.7.2	9.646 us	0.0978 us	0.0915 us	1.6785	0.0153	10 KB

a50e07a

Method	Runtime	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	6.732 us	0.0667 us	0.0624 us	1.3809	0.0153	8 KB
BeEquivalentTo	.NET Framework 4.7.2	9.559 us	0.1534 us	0.1435 us	1.5717	0.0153	10 KB

8eeb20c

Method	Runtime	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	6.833 us	0.0264 us	0.0247 us	1.3199	0.0153	8 KB
BeEquivalentTo	.NET Framework 4.7.2	9.518 us	0.1333 us	0.1247 us	1.5106	0.0153	9 KB

Benchmark 2: Comparing lists of complex objects

BeEquivalentToWithDeeplyNestedStructures with these modifications

[Params(5)]
public int N { get; set; }

[Params(4)]
public int Depth { get; set; }

[GlobalSetup]
public void GlobalSetup()
{
	subject = Enumerable.Range(0, N).Select(_ => CreateComplex(Depth)).ToList();
	expectation = Enumerable.Range(0, N).Select(_ => CreateComplex(Depth)).ToList();

	AssertionOptions.AssertEquivalencyUsing(e => e.WithoutStrictOrdering());
}

develop

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	5	4	243.9 us	1.60 us	1.49 us	44.9219	1.9531	278 KB
BeEquivalentTo	.NET Framework 4.7.2	5	4	345.5 us	3.87 us	3.62 us	49.3164	1.9531	303 KB

8eeb20c - RequiresConversion

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	5	4	222.2 us	1.87 us	1.46 us	34.6680	0.9766	215 KB
BeEquivalentTo	.NET Framework 4.7.2	5	4	323.7 us	5.21 us	4.88 us	40.0391	1.4648	247 KB

55c79b3 Node.GetHashCode

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	5	4	216.6 us	1.21 us	1.08 us	33.2031	1.2207	205 KB
BeEquivalentTo	.NET Framework 4.7.2	5	4	305.7 us	3.23 us	3.03 us	38.0859	1.4648	235 KB

55c79b3 Node.Equals

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	5	4	216.5 us	1.17 us	1.10 us	32.7148	1.2207	201 KB
BeEquivalentTo	.NET Framework 4.7.2	5	4	306.4 us	4.11 us	3.64 us	37.1094	0.9766	231 KB

55c79b3 Cache PathName

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	5	4	209.4 us	0.99 us	0.92 us	31.2500	1.2207	192 KB
BeEquivalentTo	.NET Framework 4.7.2	5	4	288.2 us	2.05 us	1.92 us	35.6445	0.9766	221 KB

[jnyrup]

In develop the lambda is constructed in each iteration no matter if we enter the if or not.

See the difference on
sharplab

[coveralls]

Pull Request Test Coverage Report for Build 2405223709

Warning: This coverage report may be inaccurate.

We've detected an issue with your CI configuration that might affect the accuracy of this pull request's coverage report.
To ensure accuracy in future PRs, please see these guidelines.
A quick fix for this PR: rebase it; your next report should be accurate.

• 30 of 30 (100.0%) changed or added relevant lines in 4 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.008%) to 96.63%

---

Totals
Change from base Build 2386863243:	-0.008%
Covered Lines:	12243
Relevant Lines:	12509

---

💛 - Coveralls

[dennisdoomen]

Is this really that expensive?

[jnyrup]

Expensiveness might be a relative term, as it has to compared against code maintainability.
The raw numbers in terms of performance are listed in the last benchmark.

The operation of concatenating strings (of course) depends on the length of the member name and how far down the member path we are.
The benchmarks performed have quite short type and member names, so a more realistic example with more descriptive/longer names would benefit even more from this.

Many of the calls to PathAndName are avoided in 8eeb20c as ObjectInfo.ctor calls PathAndName and also by using Path and Name separately in Node.Equals and Node.GetHashCode.

The remaining calls to PathAndName comes from

• Node.Depth
• Node.IsRoot
• EquivalencyValidationContext.IsCyclicReference
• Property.ctor

For this example below the caching of PathAndName reduces the number of allocating calls to PathName from 90 to 20.

var subject = new MyClass[]
{
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
};
var expected = = new MyClass[]
{
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
    new () { Value = 1, Inner = new () { Value = 2 } },
};

subject.Should().BeEquivalentTo(expected);

[dennisdoomen]

Why do you need this?

[jnyrup]

The idea was to use GetOrAdd<TArg> (TKey key, Func<TKey, TArg, TValue> valueFactory, TArg factoryArgument)
and pass in the state as the factoryArgument to avoid creating a capturing lambda expression.
I did some more benchmarks on BeEquivalentToWithDeeplyNestedStructures and it showed a measurable negative effect on the runtime.
As the code is more complex and the memory savings are only moderate, I've reverted that commit.

develop

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	32.98 us	0.623 us	0.692 us	5.9204	0.1221	36 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	51.94 us	0.378 us	0.354 us	6.4087	0.1221	39 KB
BeEquivalentTo	.NET 6.0	1	2	41.43 us	0.686 us	0.642 us	7.5684	0.1831	46 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	62.86 us	0.205 us	0.191 us	8.1787	0.1221	51 KB
BeEquivalentTo	.NET 6.0	1	6	77.23 us	0.749 us	0.701 us	14.2822	0.4883	88 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	110.89 us	0.493 us	0.461 us	15.5029	0.4883	96 KB
BeEquivalentTo	.NET 6.0	10	1	204.92 us	1.006 us	0.941 us	38.5742	1.4648	237 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	281.62 us	3.471 us	3.247 us	41.9922	1.4648	258 KB
BeEquivalentTo	.NET 6.0	10	2	291.63 us	1.432 us	1.340 us	54.6875	1.9531	337 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	396.12 us	6.709 us	5.602 us	59.5703	1.9531	368 KB
BeEquivalentTo	.NET 6.0	10	6	629.06 us	3.224 us	3.016 us	122.0703	6.8359	748 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	884.21 us	11.413 us	10.676 us	132.8125	7.8125	818 KB

373a6ad - Lazily construct the equivalency options summary

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	31.98 us	0.206 us	0.193 us	5.5542	0.1221	34 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	49.69 us	0.961 us	0.852 us	6.0425	0.1221	37 KB
BeEquivalentTo	.NET 6.0	1	2	41.15 us	0.427 us	0.400 us	7.2021	0.1831	44 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	62.28 us	0.495 us	0.463 us	7.8125	0.1221	48 KB
BeEquivalentTo	.NET 6.0	1	6	75.42 us	0.317 us	0.296 us	13.9160	0.4883	86 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	109.53 us	0.823 us	0.643 us	15.1367	0.3662	94 KB
BeEquivalentTo	.NET 6.0	10	1	202.09 us	0.848 us	0.752 us	38.3301	1.2207	235 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	277.09 us	2.755 us	2.577 us	41.5039	0.9766	256 KB
BeEquivalentTo	.NET 6.0	10	2	287.58 us	3.306 us	3.092 us	54.1992	1.9531	334 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	393.33 us	3.889 us	3.637 us	59.0820	1.9531	365 KB
BeEquivalentTo	.NET 6.0	10	6	653.00 us	3.513 us	2.934 us	121.0938	6.8359	746 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	862.93 us	9.103 us	8.515 us	131.8359	6.8359	816 KB

6c5db11 - Avoid repeatedly constructing lambda

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	31.32 us	0.404 us	0.378 us	5.1270	0.1221	32 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	50.09 us	0.462 us	0.409 us	5.6152	0.1221	35 KB
BeEquivalentTo	.NET 6.0	1	2	40.81 us	0.412 us	0.385 us	6.6528	0.1831	41 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	60.81 us	0.487 us	0.407 us	7.2632	0.1831	45 KB
BeEquivalentTo	.NET 6.0	1	6	73.71 us	0.394 us	0.368 us	12.6953	0.3662	78 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	108.38 us	1.447 us	1.354 us	13.9160	0.3662	86 KB
BeEquivalentTo	.NET 6.0	10	1	199.46 us	0.758 us	0.672 us	34.9121	1.2207	214 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	277.28 us	2.033 us	1.901 us	38.0859	0.9766	235 KB
BeEquivalentTo	.NET 6.0	10	2	276.10 us	2.801 us	2.339 us	49.3164	1.9531	303 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	395.81 us	3.764 us	3.521 us	54.1992	1.9531	334 KB
BeEquivalentTo	.NET 6.0	10	6	627.73 us	5.543 us	4.914 us	109.3750	5.8594	675 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	858.80 us	10.349 us	9.680 us	121.0938	6.8359	744 KB

Avoid capture in GetEqualityStrategy

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	31.75 us	0.179 us	0.168 us	5.0049	0.1221	31 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	49.64 us	0.746 us	0.698 us	5.6152	0.1221	35 KB
BeEquivalentTo	.NET 6.0	1	2	41.73 us	0.385 us	0.360 us	6.4697	0.1221	40 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	60.87 us	0.661 us	0.586 us	7.2021	0.1221	45 KB
BeEquivalentTo	.NET 6.0	1	6	75.15 us	0.192 us	0.171 us	12.3291	0.3662	76 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	109.04 us	1.774 us	1.385 us	13.9160	0.3662	86 KB
BeEquivalentTo	.NET 6.0	10	1	202.90 us	1.279 us	1.197 us	33.9355	1.2207	208 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	284.56 us	2.712 us	2.537 us	38.0859	0.9766	235 KB
BeEquivalentTo	.NET 6.0	10	2	292.26 us	1.784 us	1.581 us	47.8516	1.9531	295 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	399.16 us	4.039 us	3.778 us	54.1992	1.9531	334 KB
BeEquivalentTo	.NET 6.0	10	6	643.02 us	2.192 us	1.943 us	106.4453	5.8594	657 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	878.81 us	8.633 us	8.075 us	121.0938	6.8359	744 KB

981570b - Avoid allocations in BecauseOf

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	30.45 us	0.243 us	0.227 us	4.7607	0.0610	29 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	49.10 us	0.930 us	0.824 us	5.3711	0.1221	33 KB
BeEquivalentTo	.NET 6.0	1	2	39.17 us	0.352 us	0.329 us	6.1035	0.1221	38 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	60.24 us	0.823 us	0.729 us	6.9580	0.1221	43 KB
BeEquivalentTo	.NET 6.0	1	6	72.41 us	0.417 us	0.390 us	11.5967	0.3662	71 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	107.78 us	0.234 us	0.195 us	13.1836	0.3662	81 KB
BeEquivalentTo	.NET 6.0	10	1	192.19 us	1.045 us	0.926 us	31.7383	0.9766	195 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	280.39 us	2.284 us	2.136 us	35.6445	0.9766	221 KB
BeEquivalentTo	.NET 6.0	10	2	277.87 us	1.155 us	1.081 us	44.4336	1.4648	274 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	390.35 us	3.603 us	3.370 us	50.7813	1.4648	313 KB
BeEquivalentTo	.NET 6.0	10	6	604.42 us	6.993 us	6.199 us	98.6328	5.8594	608 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	858.32 us	6.407 us	5.993 us	112.3047	5.8594	696 KB

8da904f - Avoid allocations in ConversionSelector

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	29.98 us	0.214 us	0.200 us	4.5776	0.0610	28 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	47.50 us	0.629 us	0.558 us	5.1880	0.0610	32 KB
BeEquivalentTo	.NET 6.0	1	2	39.67 us	0.315 us	0.295 us	5.8594	0.1221	36 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	58.46 us	0.445 us	0.394 us	6.6528	0.1221	41 KB
BeEquivalentTo	.NET 6.0	1	6	70.57 us	0.320 us	0.300 us	10.9863	0.3662	68 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	104.00 us	0.939 us	0.832 us	12.5732	0.3662	78 KB
BeEquivalentTo	.NET 6.0	10	1	187.72 us	0.860 us	0.805 us	30.2734	0.9766	185 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	265.62 us	2.762 us	2.584 us	34.1797	0.9766	212 KB
BeEquivalentTo	.NET 6.0	10	2	264.87 us	1.372 us	1.283 us	42.4805	1.4648	262 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	379.77 us	4.020 us	3.563 us	48.8281	1.4648	300 KB
BeEquivalentTo	.NET 6.0	10	6	579.72 us	1.538 us	1.438 us	93.7500	4.8828	579 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	834.86 us	10.682 us	9.469 us	108.3984	5.8594	666 KB

370c674 - Avoid/cache string allocations

Method	Runtime	N	Depth	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
BeEquivalentTo	.NET 6.0	1	1	28.93 us	0.293 us	0.274 us	4.3945	0.0610	27 KB
BeEquivalentTo	.NET Framework 4.7.2	1	1	46.16 us	0.377 us	0.353 us	4.9438	0.0610	31 KB
BeEquivalentTo	.NET 6.0	1	2	37.23 us	0.341 us	0.319 us	5.4932	0.1221	34 KB
BeEquivalentTo	.NET Framework 4.7.2	1	2	56.37 us	0.399 us	0.354 us	6.2256	0.1221	39 KB
BeEquivalentTo	.NET 6.0	1	6	69.50 us	0.335 us	0.313 us	9.8877	0.2441	61 KB
BeEquivalentTo	.NET Framework 4.7.2	1	6	97.00 us	1.051 us	0.878 us	11.3525	0.2441	70 KB
BeEquivalentTo	.NET 6.0	10	1	176.22 us	1.161 us	1.086 us	28.0762	0.9766	172 KB
BeEquivalentTo	.NET Framework 4.7.2	10	1	252.86 us	1.836 us	1.718 us	31.7383	0.9766	196 KB
BeEquivalentTo	.NET 6.0	10	2	254.10 us	1.439 us	1.346 us	38.5742	1.4648	239 KB
BeEquivalentTo	.NET Framework 4.7.2	10	2	350.46 us	3.729 us	3.306 us	44.4336	1.4648	274 KB
BeEquivalentTo	.NET 6.0	10	6	548.07 us	2.091 us	1.956 us	82.0313	4.8828	507 KB
BeEquivalentTo	.NET Framework 4.7.2	10	6	759.56 us	11.574 us	9.036 us	94.7266	4.8828	586 KB

[jnyrup]

The change from two to three branches is the cause of the remaining decrease in branch coverage.

[dennisdoomen]

Nice work!