xUnit runs tests via dotnet test or Test Explorer using [Fact] and [Theory] attributes
Constructor + IDisposable replace [SetUp]/[TearDown] — each test gets a fresh class instance
IClassFixture shares expensive setup without sharing mutable state
Moq creates fake dependencies; always depend on interfaces in constructors
Default parallel execution per class can cause flaky tests — use [Collection] to isolate
Biggest mistake: writing a single [Theory] with only one InlineData — that's just a verbose [Fact]
✦ Definition~90s read
What is Unit Testing in C# with xUnit?
xUnit is a free, open-source unit testing framework for .NET, created by the original author of NUnit v2. It's the de facto standard for modern .NET testing, used by Microsoft internally and in most production .NET codebases. Unlike MSTest or NUnit, xUnit was designed from scratch to eliminate common anti-patterns like shared mutable state between tests — which is exactly why its IClassFixture and ICollectionFixture features can backfire when misused.
★
Imagine you build a vending machine.
The framework enforces test isolation by default: each test class gets a fresh instance, and setup/teardown happens via constructor and IDisposable rather than [Setup]/[TearDown] attributes. This design forces you to think about state management explicitly, which is both its strength and the source of the 'local pass, CI fail' trap when shared fixtures inadvertently introduce ordering dependencies or resource contention.
Where xUnit shines is in its clean separation of test structure: [Fact] for parameterless, deterministic tests and [Theory] with [InlineData], [MemberData], or [ClassData] for data-driven scenarios. Combined with mocking frameworks like Moq (or NSubstitute), you can isolate dependencies and test logic in true unit-test fashion.
However, xUnit's fixture system — IClassFixture<T> for per-class shared setup and ICollectionFixture<T> for cross-class sharing — is often misunderstood. These fixtures are meant for expensive, read-only resources (like a database connection or HTTP client), not for mutable state.
When you treat them as a cache or shared context, tests that pass sequentially on your dev machine fail under parallel CI execution because xUnit runs test collections in parallel by default. The fix is either to make fixtures immutable, use IClassFixture with Dispose for cleanup, or switch to IAsyncLifetime for async setup that doesn't leak state between tests.
In practice, you should avoid xUnit shared fixtures entirely for most unit tests — constructor injection of mocks and per-test setup is simpler and more reliable. Reserve fixtures for integration tests where you genuinely need a shared database container (via Testcontainers) or an HTTP server (via WebApplicationFactory).
For data-driven tests, MemberData and ClassData give you strongly-typed, reusable test data without the overhead of fixtures. The key insight: xUnit's defaults are designed to catch the exact bugs that 'local pass, CI fail' represents — embrace per-test isolation, and only reach for shared fixtures when you've measured a real performance bottleneck.
Plain-English First
Imagine you build a vending machine. Before shipping it to a hospital, you test every single button — does pressing B3 drop a Snickers? Does it reject a fake coin? Unit testing is exactly that: you write tiny automated 'button-press checks' for every function in your code, so you catch broken buttons before your users do. xUnit is the machine that runs all those checks for you, prints a green tick when everything works, and screams a red X when something breaks.
Every production codebase eventually reaches a tipping point where a developer changes one function and silently breaks three others. Without automated tests, you only find out when a customer tweets at you at 2am. Unit testing is not a 'nice to have' — it is the safety harness that lets your team move fast without falling off a cliff. Companies like Microsoft, Stripe, and Shopify treat untested code as unshippable code, and for good reason.
The specific pain xUnit solves is the chaos of manual regression checking. Without it, every new feature means re-clicking through the entire app to make sure nothing broke. xUnit lets you encode that mental checklist as code, run it in under a second, and get a precise pass/fail report with zero human effort. It also integrates natively into the .NET ecosystem, works beautifully with GitHub Actions and Azure DevOps pipelines, and produces machine-readable output that pull-request bots can act on automatically.
By the end of this article you will know how to structure a real xUnit test project from scratch, write both simple Fact tests and data-driven Theory tests, isolate dependencies using the Moq library, and avoid the three mistakes that waste most beginners' first week with xUnit. You will also understand why each xUnit design decision exists, so you can make smart choices on your own projects rather than blindly copying Stack Overflow snippets.
Here's the honest truth: most developers learn xUnit by copying a template, write a few tests that all pass, and then ship untested code into production because they never hit the awkward edges. This article exists to force you past that plateau. You'll leave with the mental model that separates a junior who writes tests from a senior who designs systems that are testable.
What xUnit Shared Fixtures Actually Do — And Where They Break
xUnit shared fixtures let you create a single object instance that is shared across all test methods in a class (or collection). You implement IClassFixture<T> or ICollectionFixture<T>, and xUnit injects that fixture into the test class constructor once per class, not once per test. This is the core mechanic: one setup, many tests.
In practice, shared fixtures are ideal for expensive resources like database connections, HTTP clients, or service containers. They reduce test runtime by avoiding repeated construction and teardown. But they also introduce state coupling: if a test mutates the fixture, subsequent tests see that mutation. xUnit does not reset the fixture between tests. This is the trap — tests pass locally in isolation but fail in CI when run in a different order or with more parallelism.
Use shared fixtures only for read-only or reset-safe resources. For mutable state, prefer collection fixtures with explicit cleanup in Dispose, or use the newer IAsyncLifetime for async teardown. In real systems, the cost of a shared fixture bug is a flaky CI pipeline that erodes team trust in the test suite.
Shared Fixture ≠ Shared State
A shared fixture is not a singleton — it's scoped to one test class. But if that fixture holds mutable state, tests are no longer isolated.
Production Insight
Team had a shared HttpClient fixture that cached auth tokens — tests passed locally because they ran sequentially, but in CI parallel execution caused token expiry races.
Symptom: intermittent 401 responses from integration tests, only in CI, never locally.
Rule: if a fixture holds any mutable state, treat it as per-test or implement explicit reset in Dispose.
Key Takeaway
Shared fixtures are for expensive, read-only resources — never for mutable state.
xUnit does not reset fixtures between tests — state leaks are your responsibility.
When a CI test fails but local passes, suspect shared fixture mutation first.
thecodeforge.io
xUnit Shared Fixture Trap: Local Pass, CI Fail
Unit Testing Csharp Xunit
Setting Up a Real xUnit Project — Structure That Scales
The number one mistake teams make is dumping tests into the same project as production code. That forces your shipping binary to carry test dependencies, and it blurs the line between what you own and what you are testing. The industry-standard layout is a separate .Tests project that references your production project.
Here is exactly how to scaffold this from the terminal. The key insight is that dotnet new xunit gives you a ready-to-run test runner — xUnit's runner is baked in via the xunit.runner.visualstudio package, which is what lets Visual Studio's Test Explorer and dotnet test both work without extra wiring.
Notice that your test project references your production project directly. xUnit discovers test classes by scanning for public classes with methods decorated with [Fact] or [Theory] — no base class, no interface, no ceremony. That is a deliberate philosophy: tests should read like plain C#, not like a framework DSL.
ProjectSetup.shCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
# Run these commands in your terminal to scaffold the solution
# 1. Create a solution folder
mkdir OrderProcessingApp && cd OrderProcessingApp
# 2. Create the production class library
dotnet new classlib -n OrderProcessing.Core
# 3. Create the xUnit test project
dotnet new xunit -n OrderProcessing.Tests
# 4. Create a solution file to hold both
dotnet new sln -n OrderProcessingApp
# 5. Add both projects to the solution
dotnet sln add OrderProcessing.Core/OrderProcessing.Core.csproj
dotnet sln add OrderProcessing.Tests/OrderProcessing.Tests.csproj
# 6. Reference the production project FROM the test project
dotnet add OrderProcessing.Tests/OrderProcessing.Tests.csproj \
reference OrderProcessing.Core/OrderProcessing.Core.csproj
# Your folder structure should look like this:
# OrderProcessingApp/
# ├── OrderProcessingApp.sln
# ├── OrderProcessing.Core/ <-- production code
# │ ├── OrderProcessing.Core.csproj
# │ └── OrderCalculator.cs
# └── OrderProcessing.Tests/ <-- test code (separate project)
# ├── OrderProcessing.Tests.csproj
# └── OrderCalculatorTests.cs
# Run all tests from the solution root
dotnet test
Output
Build succeeded.
Test run for OrderProcessing.Tests.dll (.NETCoreApp,Version=v8.0)
Microsoft (R) Test Execution Command Line Tool Version 17.x
Name your test classes to mirror the class they test — OrderCalculator.cs maps to OrderCalculatorTests.cs, in a matching namespace like OrderProcessing.Tests. This makes navigation instant: when you open a class, you always know exactly where its tests live without hunting.
Production Insight
Forgetting to add xunit.runner.visualstudio breaks Test Explorer.
dotnet test works via CLI, but Visual Studio needs the adapter.
Rule: Always include the runner package in your test project.
Key Takeaway
Keep tests in a separate project.
Mirror namespaces for instant navigation.
Always add the Visual Studio test adapter.
Fact vs Theory — Writing Tests That Actually Prove Something
xUnit gives you two test primitives: [Fact] and [Theory]. Understanding the difference is the key to writing tests that are genuinely useful rather than tests that only prove one lucky path through your code.
A [Fact] is a single, unconditional assertion: 'this is always true, no arguments needed.' Use it for edge cases, boundary conditions, and single-scenario checks. A [Theory] is a parameterised test that says 'this should be true for all of these inputs.' You supply multiple data sets via [InlineData], [MemberData], or [ClassData], and xUnit runs your test method once per set, independently.
Why does this matter? Because a bug in a calculation function usually lives at an edge — zero, negative numbers, null strings, max integer. A single [Fact] with one happy-path number gives you false confidence. A [Theory] with seven representative inputs — including edge cases — is what actually catches real bugs before production does.
Below is a production-realistic example using an OrderCalculator that applies discounts. Notice the test names are descriptive English sentences — that is intentional. When a test fails in CI, the name is your first clue, so 'CalculateTotal_WhenDiscountExceedsHundredPercent_ThrowsArgumentException' tells you exactly what broke without opening the file.
OrderCalculatorTests.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
// OrderProcessing.Core/OrderCalculator.cs — the class we are testingnamespaceOrderProcessing.Core
{
publicclassOrderCalculator
{
// Applies a percentage discount to the subtotal and returns the final pricepublicdecimalCalculateTotal(decimal subtotal, decimal discountPercent)
{
if (subtotal < 0)
thrownewArgumentOutOfRangeException(nameof(subtotal), "Subtotal cannot be negative.");
if (discountPercent < 0 || discountPercent > 100)
thrownewArgumentOutOfRangeException(nameof(discountPercent), "Discount must be between 0 and 100.");
var discountAmount = subtotal * (discountPercent / 100m);
return subtotal - discountAmount;
}
// Returns true only if the order qualifies for free shippingpublicboolQualifiesForFreeShipping(decimal orderTotal, string customerTier)
{
return customerTier == "Gold" && orderTotal >= 50m
|| customerTier == "Standard" && orderTotal >= 100m;
}
}
}
// ─────────────────────────────────────────────────────────// OrderProcessing.Tests/OrderCalculatorTests.csusingXunit;
usingOrderProcessing.Core;
namespaceOrderProcessing.Tests
{
publicclassOrderCalculatorTests
{
// Create one instance of the calculator — it has no state, so it's safe to shareprivatereadonlyOrderCalculator _calculator = newOrderCalculator();
// ── [Fact] tests: single, unconditional scenarios ──
[Fact]
publicvoid CalculateTotal_WhenSubtotalIsNegative_ThrowsArgumentOutOfRangeException()
{
// Arrange: a clearly invalid subtotaldecimal negativeSubtotal = -10.00m;
decimal validDiscount = 10m;
// Act + Assert: xUnit's Assert.Throws captures the exception cleanlyvar exception = Assert.Throws<ArgumentOutOfRangeException>(
() => _calculator.CalculateTotal(negativeSubtotal, validDiscount)
);
// Also verify the exception message mentions the right parameter nameAssert.Equal("subtotal", exception.ParamName);
}
[Fact]
publicvoid CalculateTotal_WhenDiscountIsZero_ReturnFullSubtotal()
{
// Arrangedecimal subtotal = 200.00m;
decimal noDiscount = 0m;
// Actdecimal result = _calculator.CalculateTotal(subtotal, noDiscount);
// Assert: no discount means we pay full priceAssert.Equal(200.00m, result);
}
// ── [Theory] tests: same logic, many data sets ──
[Theory]
// Format: subtotal, discountPercent, expectedTotal
[InlineData(100.00, 10, 90.00)] // standard 10% off
[InlineData(200.00, 25, 150.00)] // 25% off a larger order
[InlineData(50.00, 100, 0.00)] // 100% off — free item promotion
[InlineData(99.99, 0, 99.99)] // zero discount — pay full price
[InlineData(0.00, 50, 0.00)] // zero subtotal — result must also be zeropublicvoid CalculateTotal_WithVariousDiscounts_ReturnsCorrectTotal(
decimal subtotal,
decimal discountPercent,
decimal expectedTotal)
{
// Act: run the same logic for every [InlineData] row abovedecimal actualTotal = _calculator.CalculateTotal(subtotal, discountPercent);
// Assert.Equal on decimals — xUnit compares value, not referenceAssert.Equal(expectedTotal, actualTotal);
}
[Theory]
[InlineData(50.00, "Gold", true)] // Gold hits threshold of 50
[InlineData(49.99, "Gold", false)] // Gold just under threshold
[InlineData(100.00, "Standard", true)] // Standard hits threshold of 100
[InlineData(99.99, "Standard", false)] // Standard just under threshold
[InlineData(500.00, "Bronze", false)] // Unknown tier never qualifiespublicvoid QualifiesForFreeShipping_WithVariousTiersAndTotals_ReturnsExpected(
decimal orderTotal,
string customerTier,
bool expectedResult)
{
bool actualResult = _calculator.QualifiesForFreeShipping(orderTotal, customerTier);
Assert.Equal(expectedResult, actualResult);
}
}
}
Output
Test run for OrderProcessing.Tests.dll (.NETCoreApp,Version=v8.0)
Interview Gold: Why xUnit Uses [Fact] and [Theory] Instead of [Test]
xUnit's authors deliberately chose different attribute names from NUnit/MSTest to make the intent of the test explicit in the code. A [Fact] is a universal truth. A [Theory] is a hypothesis you are validating across a range of inputs. This naming forces developers to think about what they are actually claiming, which leads to better test design.
Production Insight
A [Theory] with only one InlineData is just a verbose [Fact].
The real value comes from edge cases: zero, null, boundary.
Rule: Every [Theory] should have at least 3 data sets including failure cases.
Key Takeaway
Use [Fact] for single scenarios, [Theory] for data-driven.
Descriptive test names are your first failure clue.
Test boundary conditions, not just happy paths.
Mocking Dependencies With Moq — Testing Code in Isolation
Real services talk to databases, payment gateways, and email providers. If your unit test actually hits a database, it is not a unit test — it is a slow, flaky integration test that fails whenever the DB is unreachable. The solution is dependency injection plus mocking: you inject a fake version of the dependency that behaves exactly as you dictate, so your test owns the scenario completely.
Moq is the most widely used mocking library in the .NET ecosystem. You install it into your test project only — production code never sees it. The mental model is simple: Mock<IEmailService> creates a stand-in actor who plays the role of IEmailService. You script its lines with Setup(...), run the test, then verify it delivered those lines with Verify(...).
This pattern only works if your production class accepts its dependencies through a constructor (constructor injection). If a class creates its own new EmailService() internally, you cannot mock it. This is why dependency injection is not just an architectural nicety — it is a testability requirement.
Below we test an OrderService that sends a confirmation email after a successful order. We want to verify the email is sent exactly once with the right address, without firing off a real email during our test run.
OrderServiceTests.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
// First, add Moq to the test project:// dotnet add OrderProcessing.Tests package Moq// ─── OrderProcessing.Core/IEmailService.cs ───namespaceOrderProcessing.Core
{
publicinterfaceIEmailService
{
// Sends a confirmation to the given address and returns true on successboolSendOrderConfirmation(string recipientEmail, int orderId);
}
}
// ─── OrderProcessing.Core/OrderService.cs ───namespaceOrderProcessing.Core
{
publicclassOrderService
{
privatereadonlyIEmailService _emailService;
privatereadonlyOrderCalculator _calculator;
// Dependencies injected through constructor — this is what makes mocking possiblepublicOrderService(IEmailService emailService, OrderCalculator calculator)
{
_emailService = emailService;
_calculator = calculator;
}
// Places an order, calculates final price, and sends a confirmation email// Returns the final order total, or throws if email failspublicdecimalPlaceOrder(string customerEmail, decimal subtotal, decimal discountPercent)
{
if (string.IsNullOrWhiteSpace(customerEmail))
thrownewArgumentException("Customer email is required.", nameof(customerEmail));
decimal finalTotal = _calculator.CalculateTotal(subtotal, discountPercent);
// Generate a deterministic order ID (simplified for the example)int orderId = Math.Abs(customerEmail.GetHashCode() % 100000);
bool emailSent = _emailService.SendOrderConfirmation(customerEmail, orderId);
if (!emailSent)
thrownewInvalidOperationException("Order confirmation email could not be sent.");
return finalTotal;
}
}
}
// ─── OrderProcessing.Tests/OrderServiceTests.cs ───usingXunit;
usingMoq;
usingOrderProcessing.Core;
namespaceOrderProcessing.Tests
{
publicclassOrderServiceTests
{
// Moq creates a fake IEmailService — no real emails, no SMTP server neededprivatereadonlyMock<IEmailService> _mockEmailService;
privatereadonlyOrderService _orderService;
publicOrderServiceTests()
{
_mockEmailService = newMock<IEmailService>();
var calculator = new OrderCalculator(); // real calculator — no need to mock pure math// Inject the MOCK email service into OrderService
_orderService = newOrderService(_mockEmailService.Object, calculator);
}
[Fact]
publicvoid PlaceOrder_WhenEmailSucceeds_ReturnsFinalTotal()
{
// Arrange: script the fake email service to succeed for ANY string and int
_mockEmailService
.Setup(emailSvc => emailSvc.SendOrderConfirmation(
It.IsAny<string>(), // match any email addressIt.IsAny<int>())) // match any order ID
.Returns(true); // always say 'yes, email was sent'// Act: place a real order through the real OrderServicedecimal finalTotal = _orderService.PlaceOrder(
customerEmail: "alice@example.com",
subtotal: 150.00m,
discountPercent: 10m
);
// Assert 1: the math came out right (10% off 150 = 135)Assert.Equal(135.00m, finalTotal);
// Assert 2: verify the email service was called exactly once// This proves we didn't forget to send the confirmation
_mockEmailService.Verify(
emailSvc => emailSvc.SendOrderConfirmation(
"alice@example.com", // must be this exact addressIt.IsAny<int>()), // any order ID is fineTimes.Once() // called exactly once — not zero, not twice
);
}
[Fact]
publicvoid PlaceOrder_WhenEmailServiceFails_ThrowsInvalidOperationException()
{
// Arrange: script the fake email service to FAIL
_mockEmailService
.Setup(emailSvc => emailSvc.SendOrderConfirmation(
It.IsAny<string>(),
It.IsAny<int>()))
.Returns(false); // simulate an SMTP outage// Act + Assert: placing the order should throw because email failedvar exception = Assert.Throws<InvalidOperationException>(
() => _orderService.PlaceOrder("bob@example.com", 200.00m, 0m)
);
Assert.Contains("confirmation email", exception.Message);
}
[Fact]
publicvoid PlaceOrder_WhenEmailIsEmpty_ThrowsArgumentException_BeforeCallingEmailService()
{
// Act + Assert: an empty email should fail immediatelyAssert.Throws<ArgumentException>(
() => _orderService.PlaceOrder("", 100.00m, 0m)
);
// Verify the email service was NEVER called — we should have thrown before reaching it
_mockEmailService.Verify(
emailSvc => emailSvc.SendOrderConfirmation(
It.IsAny<string>(),
It.IsAny<int>()),
Times.Never() // this line proves our guard clause fired correctly
);
}
}
}
Output
Test run for OrderProcessing.Tests.dll (.NETCoreApp,Version=v8.0)
Moq can mock concrete classes, but only if their methods are marked virtual. If you try to mock a non-virtual method on a concrete class, Moq silently ignores your Setup and calls the real method — your test passes for the wrong reason. The clean solution is always to depend on interfaces, not concrete types. If you own neither, use a wrapper interface.
Production Insight
Mocking concrete classes without virtual methods leads to false positives.
Moq silently calls the real method if a method is not virtual.
Rule: Depend on interfaces, never concrete classes for mocking.
Key Takeaway
Constructor injection enables mocking.
Use Moq's Setup and Verify for behaviour testing.
Always verify the mock's interactions after Act.
Test Lifecycle with IClassFixture — Shared Setup Without Shared State
xUnit creates a fresh instance of your test class for every single test method. This is a deliberate design decision that eliminates a whole class of bugs caused by tests accidentally sharing state. NUnit and MSTest use [SetUp]/[TearDown] methods, which run before and after each test but on the same object — that means a dirty field from test A can corrupt test B if your setup is incomplete.
xUnit's answer is simpler: constructor and IDisposable. Anything you put in the test class constructor runs before each test. Anything you put in Dispose() runs after. No magic attributes — just C# you already know.
But sometimes setup is genuinely expensive — spinning up an in-memory database, loading a large config file — and you do not want to repeat it for every test. That is what IClassFixture<T> is for. It creates the expensive resource once per test class, shares it across all tests in that class, then disposes it when the class is done. Crucially, each test still gets a fresh test class instance — only the fixture is shared.
Here is a pattern you will see in real .NET projects using an in-memory database fixture.
OrderRepositoryTests.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
// dotnet add OrderProcessing.Tests package Microsoft.EntityFrameworkCore.InMemory// ─── OrderProcessing.Core/Order.cs ───namespaceOrderProcessing.Core
{
publicclassOrder
{
publicintId { get; set; }
publicstringCustomerEmail { get; set; } = string.Empty;
publicdecimalTotal { get; set; }
publicDateTimePlacedAt { get; set; }
}
}
// ─── OrderProcessing.Core/AppDbContext.cs ───usingMicrosoft.EntityFrameworkCore;
namespaceOrderProcessing.Core
{
publicclassAppDbContext : DbContext
{
publicAppDbContext(DbContextOptions<AppDbContext> options) : base(options) { }
publicDbSet<Order> Orders => Set<Order>();
}
}
// ─── OrderProcessing.Tests/DatabaseFixture.cs ───// This class is created ONCE and shared across all tests in OrderRepositoryTestsusingMicrosoft.EntityFrameworkCore;
usingOrderProcessing.Core;
namespaceOrderProcessing.Tests
{
publicclassDatabaseFixture : IDisposable
{
publicAppDbContextContext { get; }
publicDatabaseFixture()
{
// Build an in-memory database — fast, isolated, no file system neededvar options = newDbContextOptionsBuilder<AppDbContext>()
.UseInMemoryDatabase(databaseName: "OrderTests_" + Guid.NewGuid()) // unique name prevents bleed-over
.Options;
Context = newAppDbContext(options);
// Seed with known data that tests can rely onContext.Orders.AddRange(
newOrder { Id = 1, CustomerEmail = "alice@example.com", Total = 99.99m, PlacedAt = DateTime.UtcNow },
newOrder { Id = 2, CustomerEmail = "bob@example.com", Total = 249.00m, PlacedAt = DateTime.UtcNow }
);
Context.SaveChanges();
}
// Called automatically by xUnit after all tests in the class have runpublicvoidDispose()
{
Context.Database.EnsureDeleted(); // clean up the in-memory databaseContext.Dispose();
}
}
}
// ─── OrderProcessing.Tests/OrderRepositoryTests.cs ───usingXunit;
usingOrderProcessing.Core;
namespaceOrderProcessing.Tests
{
// IClassFixture<DatabaseFixture> tells xUnit: create DatabaseFixture once, share it with all tests herepublicclassOrderRepositoryTests : IClassFixture<DatabaseFixture>
{
privatereadonlyAppDbContext _context;
// xUnit injects the shared fixture through the constructorpublicOrderRepositoryTests(DatabaseFixture fixture)
{
_context = fixture.Context;
}
[Fact]
publicvoid GetOrder_WhenIdExists_ReturnsCorrectOrder()
{
// Act: query the seeded in-memory databasevar order = _context.Orders.Find(1);
// Assert: the seeded data is exactly what we expectAssert.NotNull(order);
Assert.Equal("alice@example.com", order.CustomerEmail);
Assert.Equal(99.99m, order.Total);
}
[Fact]
publicvoid GetOrder_WhenIdDoesNotExist_ReturnsNull()
{
// Act: look for an order that was never seededvar order = _context.Orders.Find(999);
// Assert: EF Core returns null for a missing key — not an exceptionAssert.Null(order);
}
[Fact]
publicvoid GetAllOrders_ReturnsExactlySeedCount()
{
// The fixture seeded exactly 2 ordersint orderCount = _context.Orders.Count();
Assert.Equal(2, orderCount);
}
}
}
Output
Test run for OrderProcessing.Tests.dll (.NETCoreApp,Version=v8.0)
Pro Tip: Give Each In-Memory Database a Unique Name
Notice databaseName: "OrderTests_" + Guid.NewGuid() in the fixture. If you use a fixed name like 'TestDb', parallel test runs share the same in-memory store and corrupt each other's data. A GUID suffix costs nothing and makes your tests perfectly isolated even under parallel execution.
Production Insight
In-memory databases with fixed names cause parallel test corruption.
Each test class or fixture needs a unique database name.
Rule: Use Guid.NewGuid() for in-memory database names.
Key Takeaway
IClassFixture shares one instance across tests.
Use IDisposable for cleanup.
Unique names prevent parallel test pollution.
Data-Driven Testing with MemberData and ClassData — Beyond InlineData
While [InlineData] is the quickest way to parameterise a test, it has limitations: you cannot reuse data across methods, and the data is hardcoded in the attribute. For real-world scenarios where test data comes from a file, a database, or a computed set, xUnit provides [MemberData] and [ClassData].
[MemberData] points to a static property or method that returns IEnumerable<object[]>. This lets you reuse the same data source across multiple test methods and even compute data dynamically (e.g., reading from a CSV file). [ClassData] points to a separate class that implements IEnumerable<object[]>. This is useful when the data source is complex enough to warrant its own class, perhaps with caching or lazy loading.
A common real-world pattern: load test data from a JSON or CSV file via a static MemberData method. This decouples test logic from data, making your tests easier to read and update without recompiling. Below is an example that reads order test cases from a static method.
Mental Model: Data Sources Are Just Test Iterators
InlineData: quick, simple, but couples data to the test method.
MemberData: reusable, composable, supports lazy evaluation and external files.
ClassData: best for complex or shared data sources that need their own class.
Performance: MemberData and ClassData execute once per test class, not per method — but yield returns data on demand.
Production Insight
Using InlineData with dozens of rows makes test files bloated and hard to review.
MemberData allows data to live in a separate static method or even an external file.
Rule: If you need more than 5 data sets, switch to MemberData.
Key Takeaway
InlineData for quick cases, MemberData for reusable logic.
ClassData for data sources worth a separate abstraction.
Yielding test data lazily keeps memory usage low.
Choosing the Right Data Source for Theory Tests
IfYou have 1-5 fixed data sets used only in this test method
→
UseUse [InlineData] — simplest and most readable.
IfData is reused across multiple test methods or computed at runtime
→
UseUse [MemberData] with a static property or method.
IfData source is complex (e.g., file parsing, caching) and deserves its own class
→
UseUse [ClassData] with a dedicated class implementing IEnumerable<object[]>.
Parallel Test Execution — Why Your CI Pipeline Slows to a Crawl (And How to Fix It)
Here's the dirty secret most tutorials skip: xUnit runs your tests in parallel by default. Sounds great until your integration tests trash the same database and start failing in ways that make zero sense. The WHY is critical here — parallel execution is not free.
When xUnit fires up multiple test classes, each one gets its own collection unless you tell 'em to share. That's fine for pure unit tests with zero shared state. The moment you touch a file, a network socket, or — god forbid — a SQL database, you need to think about isolation.
The fix isn't to disable parallelism entirely (that's amateur hour). Use [CollectionDefinition] to group tests that must run sequentially. Mark slow infrastructure-heavy tests with [Trait("Category", "Integration")] and filter them out of your fast feedback loop. Your CI will thank you.
Senior Shortcut: Keep unit tests parallel. Sequentialize only the minimum needed. Your developer loop shouldn't wait on the same integration tests your build server runs.
ParallelCollectionExample.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
// io.thecodeforge — csharp tutorialusingXunit;
// Define a collection that forces sequential execution
[CollectionDefinition("DatabaseTests", DisableParallelization = true)]
publicclassDatabaseTestCollection { }
// All test classes in this collection run one at a time
[Collection("DatabaseTests")]
publicclassOrderRepositoryTests
{
[Fact]
publicvoid SaveOrder_ReturnsConfirmedId()
{
// This test won't run while other "DatabaseTests" are running// Prevents connection pool exhaustion and row locksAssert.True(true);
}
}
[Collection("DatabaseTests")]
publicclassInvoiceRepositoryTests
{
[Fact]
publicvoid CreateInvoice_CreatesRecord()
{
// Same sequential guaranteeAssert.True(true);
}
}
Output
No output — tests pass sequentially; avoid race condition failures on shared DB state.
Production Trap:
Never assume parallel safety. Your test harness must mirror production concurrency or fail fast. A CI that passes locally but flakes on the build server is a bug, not a mystery.
Key Takeaway
Use CollectionDefinition to isolate shared-state tests. Parallelize everything else. Your pipeline will be faster and flaky-tests-free.
Custom Test Runners — When Built-in Assertions Fail You
Standard assertions are fine until they aren't. Ever debug a failed test that says "Expected: true, Actual: false" with zero context? That's not useful. That's a waste of time.
The answer is a custom test runner. Not for everything — you're not building a framework. But when you have a repetitive assertion pattern (e.g., verifying JSON response shapes, checking business rule violations), write your own extension method that returns a meaningful message.
WHY it matters: in production incidents, every second counts. A custom assertion that spits out "Expected response code 200, got 503. Response body: {\"error\":\"timeout\"}" is worth its weight in gold. Moq's callback hell? Same thing. Wrap it in a helper that tells you exactly which setup failed.
Don't abstract for the sake of it. Write one, maybe two custom assertion methods per project. If you have more, your test design is wrong. Assertions should read like plain English, not regex obfuscation.
CustomAssertionExample.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
// io.thecodeforge — csharp tutorialusingXunit.Sdk;
publicstaticclassCustomAssertions
{
publicstaticvoidHttpStatusOk(HttpResponseMessage response)
{
if (response.StatusCode != System.Net.HttpStatusCode.OK)
{
var body = response.Content.ReadAsStringAsync().GetAwaiter().GetResult();
thrownewXunitException(
$"Expected 200 OK, got {(int)response.StatusCode} {response.ReasonPhrase}.\n" +
$"Response body: {body}");
}
}
}
publicclassPaymentApiTests
{
[Fact]
publicasyncTask ProcessPayment_ReturnsOk()
{
var response = await new HttpClient().PostAsync("http://localhost/api/pay", null);CustomAssertions.HttpStatusOk(response); // precise failure message every time
}
}
Output
Expected 200 OK, got 500 Internal Server Error.
Response body: {"error":"insufficient funds"}
Senior Shortcut:
Never write an assertion that only says 'true/false'. Embed the actual values into the failure message. Your future self debugging at 2 AM will buy you a beer.
Key Takeaway
Custom assertions give you instant failure context. One per project. Any more and you're over-engineering.
Shared Context Without IClassFixture — Using Lazy and Constructor Injection
IClassFixture is great for shared setup but it comes with a catch: xUnit creates one instance of your fixture per test class. If you need a singleton across multiple test classes, you're stuck.
Explicitly: When you have a test that depends on a database connection pool or a configuration object that's expensive to build, you want it created once for the entire test run. IClassFixture doesn't do that — it creates a new instance per concrete test class that implements it.
Here's the production trick: use Lazy<T> in a static context combined with collection fixtures. Define a collection-level fixture that holds a Lazy<YourExpensiveResource>. The first test to hit it pays the cost. Every subsequent test gets the cached instance. No IClassFixture overhead.
WHY this matters: Integration test suites that take 10 minutes can drop to 2. The lazy initialization means you don't pay for what you don't use. Just make sure your resource is thread-safe (read: connection pools are fine, file handles are not).
Both UserTests and OrderTests share the same Database instance, initialized only once per test run.
Senior Shortcut:
Lazy<T> + collection fixtures = shared resource without fixture explosion. Avoids the per-class instantiation overhead of IClassFixture for truly expensive stuff.
Key Takeaway
Use Lazy<T> in collection fixtures to share expensive resources across test classes. Cheaper than per-class fixtures.
Why Your Test Fixtures Are Leaking State — And How to Kill It
Every senior dev has seen this mess: a test passes in isolation but fails when the whole suite runs. Nine times out of ten, it's shared state in a fixture that someone thought was read-only. xUnit reuses fixture instances across tests in the same class. If your fixture holds a mutable collection, a counter, or god forbid a static cache, you're debugging ghosts.
The fix is brutal and simple: make your fixture immutable after construction. If you must mutate, use IClassFixture with disposal that resets state. Or better yet, push mutable dependencies into mocks that you control per test. Production systems don't tolerate hidden state — neither should your test harness.
When you see a test that relies on fixture state being "just right" from a previous run, you're looking at tech debt that will bite you at 3 AM before a release. Kill it now.
LeakyFixture.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
// io.thecodeforge — csharp tutorialpublicclassLeakyFixture
{
publicList<string> Items { get; } = new();
}
publicclassFragileTests : IClassFixture<LeakyFixture>
{
[Fact]
publicvoid Add_Item_First()
{
var fixture = newLeakyFixture();
fixture.Items.Add("A");
Assert.Single(fixture.Items);
}
[Fact]
publicvoid Add_Item_Second_Fails()
{
// ❌ This test sees "A" from Add_Item_First!var fixture = newLeakyFixture();
fixture.Items.Add("B");
Assert.Single(fixture.Items); // Fails: 2 items
}
}
Output
Fail: Expected collection length 1, actual 2
Production Trap:
Never expose a mutable List<T> from a fixture. Use IReadOnlyList<T> or AsReadOnly() to enforce immutability at the type level.
Key Takeaway
Fixture state is shared state — treat it as immutable or you will chase heisenbugs.
Skipping the Fixture Factory — When IClassFixture Slows You Down
IClassFixture is great for expensive setup, but it's overkill when you just need a connection string or a config object. I've seen teams thread a fixture through every test class just to hold an IOptions<T> that never changes. That's ceremony, not architecture.
For cheap, read-only dependencies, ditch the fixture. Use Lazy<T> in a static constructor or inject your settings directly via a test constructor. No fixture registration, no IClassFixture interface, no garbage. Just a static cache that's thread-safe by default.
But watch the trap:Lazy<T> is process-wide, not test-class-wide. If your test suite runs in parallel (and it should), a Lazy<T> that initializes a database connection will fight itself. Use Lazy<T> only for truly immutable data — like config strings or compiled regex patterns. For anything that touches I/O, stick with IClassFixture and accept the minor overhead.
Test passes — config instantiated only once per test run.
Senior Shortcut:
Use Lazy<T> for config or static mocks that are read-only. For any fixture that allocates resources (DB, HTTP client), stick with IClassFixture so disposal is guaranteed.
Key Takeaway
Don't use a fixture when a static Lazy<T> will do — less code, same speed, zero ceremony.
Anti-Pattern: Manually Iterating Over Test Data
Manually looping through test data inside a [Fact] test is a common anti-pattern. Developers write a foreach or for loop, run assertions inside the body, and assume all iterations pass. When the loop breaks early on the first failure, the test stops — leaving later data points untested. Worse, you lose visibility into which specific input failed because exceptions hide the iteration index. This approach violates the principle of “one assertion per test” and makes debugging a guessing game. Instead of manual iteration, use xUnit's [Theory] with [InlineData], [MemberData], or [ClassData]. Each data row runs as an independent test, giving you clear pass/fail per case, parallel execution, and immediate visibility into the exact failing input. Manual loops also prevent accurate test reporting in CI — a single red test can mask five failures. Stop looping; let the framework handle iteration with theories for reliable, granular results.
AntiPatternTest.csCSHARP
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
// io.thecodeforge — csharp tutorialpublicclassCalculatorTests
{
// BAD: manual loop — first failure stops all
[Fact]
publicvoid Add_ManualIteration_Bad()
{
var data = new[] { (1, 2, 3), (4, 5, 9), (0, 0, 0) };
foreach (var (a, b, expected) in data)
{
var result = newCalculator().Add(a, b);
Assert.Equal(expected, result);
}
}
// GOOD: each case is a separate test
[Theory]
[InlineData(1, 2, 3)]
[InlineData(4, 5, 9)]
[InlineData(0, 0, 0)]
publicvoid Add_WithTheory_Good(int a, int b, int expected)
=> Assert.Equal(expected, newCalculator().Add(a, b));
}
Output
Test run: 3 passed, 0 failed (each theory runs independently)
Production Trap:
A manual loop in a single test hides partial failures from your CI dashboard. If the third data point fails, the first two might have passed, but your report shows one red test — losing actionable data. Theories give each input a dedicated pass/fail status.
Key Takeaway
Never iterate test data manually in a [Fact]. Use [Theory] data sources to get independent, debuggable, and parallel-friendly tests.
● Production incidentPOST-MORTEMseverity: high
Test Passes Locally but Fails in CI — The Shared Fixture Trap
Symptom
A test that checks the count of orders in an in-memory database passes when run alone but fails when the full test suite runs in CI. Error: "Expected: 2. Actual: 3."
Assumption
Developers assumed the in-memory database fixture was isolated per class. They had named it "TestDb" without a unique suffix.
Root cause
The DatabaseFixture used a fixed in-memory database name. xUnit runs test classes in parallel by default; two instances of the fixture shared the same database name, causing tests from different classes to see each other's seeded data.
Fix
Changed the in-memory database name to include a GUID: UseInMemoryDatabase("OrderTests_" + Guid.NewGuid()).
Also added a [Collection] attribute to force those tests to run sequentially, preventing any shared state bleed across unrelated tests.
Key lesson
Always give in-memory databases a unique name per fixture instance using Guid.NewGuid().
Understand xUnit's default parallel execution model — shared resources need explicit isolation.
A test that passes solo but fails in a batch is almost always a shared-state problem, not a logic bug.
Production debug guideDiagnose and fix the most common test reliability issues4 entries
Symptom · 01
Test passes on my machine but fails in CI
→
Fix
Check for shared static state, file path differences, and database name collisions.
Add a [Collection] attribute to force sequential execution and see if the failure pattern changes.
Symptom · 02
Test fails intermittently — sometimes passes, sometimes fails
→
Fix
Likely a race condition or timing issue.
Run the test in a loop: for (int i=0; i<100; i++) { RunTest(); } inside a single [Fact].
Then check for missing awaits, shared mutable state, or non-thread-safe mocks.
Symptom · 03
Test fails when run with dotnet test but passes in Test Explorer
→
Fix
Test Explorer often runs tests sequentially by default, while dotnet test runs them parallel.
Add dotnet test --settings sequential.runsettings to confirm.
Fix by adding [Collection] to isolate tests that share resources.
Symptom · 04
Test output shows 'Test Discovery Skipped'
→
Fix
Check that the test project references both xunit and xunit.runner.visualstudio packages.
Run dotnet test --list-tests to verify xUnit sees your tests.
If not, rebuild the project.
★ Quick xUnit Debug Cheat SheetCommands and actions for the most common xUnit test debugging scenarios
dotnet test --verbosity detailed | findstr TestClass
Fix now
Rebuild the test project. If still missing, remove and re-add the packages.
Test fails only in parallel run+
Immediate action
Force sequential execution to confirm
Commands
dotnet test --settings sequential.runsettings
dotnet test --filter "FullyQualifiedName~OrderServiceTests" --no-parallel
Fix now
Create a runsettings file with <ParallelizeTestCollections>false</ParallelizeTestCollections> and use dotnet test --settings.
Theory test shows generic failure message+
Immediate action
Find which data row failed
Commands
dotnet test --verbosity normal (shows parameters for each failed row)
dotnet test --filter "TestName" --verbosity detailed
Fix now
Add a custom display name using [Theory(DisplayName = "...")] or use MemberData with named parameters.
Feature / Aspect
xUnit
NUnit
MSTest
Test attribute
[Fact] / [Theory]
[Test] / [TestCase]
[TestMethod] / [DataTestMethod]
Parameterised tests
[InlineData], [MemberData], [ClassData]
[TestCase], [TestCaseSource]
[DataRow], [DynamicData]
Setup per test
Constructor + IDisposable
[SetUp] / [TearDown]
[TestInitialize] / [TestCleanup]
Shared setup across tests
IClassFixture<T>
[OneTimeSetUp] / [OneTimeTearDown]
[ClassInitialize] / [ClassCleanup]
Test isolation
New class instance per test (safest)
Same instance, reset via [SetUp]
Same instance, reset via [TestInitialize]
Parallel execution
Enabled by default per test class
Opt-in via attribute
Opt-in via settings file
Microsoft .NET team uses
Yes (runtime, ASP.NET Core repos)
No
No
Output capture
Built-in ITestOutputHelper
Console capture built-in
Console capture built-in
NuGet package
xunit + xunit.runner.visualstudio
nunit + NUnit3TestAdapter
MSTest.TestFramework + MSTest.TestAdapter
Key takeaways
1
xUnit's fresh class instance per test eliminates state leak bugs
embrace it, don't fight it.
2
Mocking requires constructor injection and interfaces
if your code uses new you can't unit test it in isolation.
3
IClassFixture shares expensive setup without sharing mutable state; always use unique names for database fixtures.
4
MemberData and ClassData separate test data from test logic and scale better than hundreds of InlineData rows.
5
Run tests sequentially first if you suspect flakiness; then inspect shared resources for concurrent access issues.
Common mistakes to avoid
3 patterns
×
Not using a separate test project
Symptom
Test dependencies (Moq, xunit) are packaged into the production build; the shipping binary is larger and may expose internal test-only types.
Fix
Always create a separate .Tests project that references the production project. Use dotnet new xunit to get the correct framework dependencies.
×
Using [Fact] for parameterised scenarios instead of [Theory]
Symptom
You end up writing five identical test methods that differ only in input values, making the test file long and hard to maintain.
Fix
Use [Theory] with [InlineData] for multiple inputs. If the data set is large, switch to [MemberData] or [ClassData].
×
Sharing mutable state via static fields across tests
Symptom
Tests that pass individually fail when run together. Debugging reveals that static state from one test leaks into another.
Fix
Avoid static mutable fields in test classes. Use constructor + instance fields instead. For shared resources, use IClassFixture<T> but ensure the fixture is immutable or provides isolated copies.
INTERVIEW PREP · PRACTICE MODE
Interview Questions on This Topic
Q01JUNIOR
Explain the difference between [Fact] and [Theory] in xUnit. When would ...
Q02SENIOR
How does xUnit prevent test pollution compared to NUnit or MSTest?
Q03SENIOR
You have a flaky test that sometimes passes and sometimes fails in a par...
Q01 of 03JUNIOR
Explain the difference between [Fact] and [Theory] in xUnit. When would you use each?
ANSWER
[Fact] is used for a single, unconditional test scenario — it takes no parameters. [Theory] is used for parameterised tests that run the same logic against multiple sets of data supplied via [InlineData], [MemberData], or [ClassData]. Use [Fact] for edge cases and single assertions. Use [Theory] when you want to verify behaviour across a range of inputs, like testing a discount calculator with different percentages.
Q02 of 03SENIOR
How does xUnit prevent test pollution compared to NUnit or MSTest?
ANSWER
xUnit creates a new instance of the test class for every single test method. In NUnit and MSTest, the same test class instance is reused across tests (with [SetUp]/[TearDown] called per test). This means mutable instance fields in NUnit can carry state from one test to another if the setup doesn't reset them properly. xUnit's model eliminates that class of bugs entirely — each test starts from a clean slate.
Q03 of 03SENIOR
You have a flaky test that sometimes passes and sometimes fails in a parallel test run. How do you diagnose and fix it?
ANSWER
First, confirm it's a parallel issue: run the test sequentially using a runsettings file with <ParallelizeTestCollections>false</ParallelizeTestCollections> or run the specific test class with dotnet test --no-parallel. If it passes sequentially, the problem is shared state between test classes. Check for static mutable fields, in-memory databases with fixed names, or non-thread-safe mocks. Fix by ensuring each test class has its own isolated resources — use unique database names (GUID), avoid statics, and use [Collection] attributes to group tests that must not run in parallel with other collections.
01
Explain the difference between [Fact] and [Theory] in xUnit. When would you use each?
JUNIOR
02
How does xUnit prevent test pollution compared to NUnit or MSTest?
SENIOR
03
You have a flaky test that sometimes passes and sometimes fails in a parallel test run. How do you diagnose and fix it?
SENIOR
FAQ · 4 QUESTIONS
Frequently Asked Questions
01
How do I run a specific test method with xUnit?
Use dotnet test --filter "FullyQualifiedName~TestMethodName". You can also filter by test class name, category, or traits. In Test Explorer, right-click the test and select 'Run'.
Was this helpful?
02
Can I skip a test conditionally?
Yes, use [Fact(Skip = "Reason")] to skip a test unconditionally. For conditional skipping, use SkipUnless or implement a custom skip condition via ITestCondition (requires third-party library like xunit.skip).
Was this helpful?
03
How do I ensure tests run sequentially and not in parallel?
Create a runsettings file with <RunSettings><RunConfiguration><MaxCpuCount>1</MaxCpuCount></RunConfiguration></RunSettings> and pass it to dotnet test: dotnet test --settings sequential.runsettings. Alternatively, disable parallelisation per test class using [Collection] attributes.
Was this helpful?
04
What is the ITestOutputHelper for?
It's xUnit's built-in output capture mechanism. Inject ITestOutputHelper into your test constructor to write messages that appear in the test output, helping debug test failures without writing to console. Example: _output.WriteLine("Processing order {0}", orderId);