advanced.md 97.3 KB
Newer Older
Gennadiy Civil's avatar
Gennadiy Civil committed
1
# Advanced googletest Topics
2

Gennadiy Civil's avatar
Gennadiy Civil committed
3
## Introduction
4

5
6
7
8
9
Now that you have read the [googletest Primer](primer.md) and learned how to
write tests using googletest, it's time to learn some new tricks. This document
will show you more assertions as well as how to construct complex failure
messages, propagate fatal failures, reuse and speed up your test fixtures, and
use various flags with your tests.
Gennadiy Civil's avatar
Gennadiy Civil committed
10
11

## More Assertions
12
13
14
15

This section covers some less frequently used, but still significant,
assertions.

Gennadiy Civil's avatar
Gennadiy Civil committed
16
### Explicit Success and Failure
17

Gennadiy Civil's avatar
Gennadiy Civil committed
18
19
20
These three assertions do not actually test a value or expression. Instead, they
generate a success or failure directly. Like the macros that actually perform a
test, you may stream a custom failure message into them.
21

Gennadiy Civil's avatar
Gennadiy Civil committed
22
23
24
```c++
SUCCEED();
```
25

Gennadiy Civil's avatar
Gennadiy Civil committed
26
Generates a success. This does **NOT** make the overall test succeed. A test is
27
28
considered successful only if none of its assertions fail during its execution.

Abseil Team's avatar
Abseil Team committed
29
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
30
31
NOTE: `SUCCEED()` is purely documentary and currently doesn't generate any
user-visible output. However, we may add `SUCCEED()` messages to googletest's
32
33
output in the future.

Gennadiy Civil's avatar
Gennadiy Civil committed
34
35
36
37
38
```c++
FAIL();
ADD_FAILURE();
ADD_FAILURE_AT("file_path", line_number);
```
39

Gennadiy Civil's avatar
Gennadiy Civil committed
40
41
42
43
`FAIL()` generates a fatal failure, while `ADD_FAILURE()` and `ADD_FAILURE_AT()`
generate a nonfatal failure. These are useful when control flow, rather than a
Boolean expression, determines the test's success or failure. For example, you
might want to write something like:
44

Gennadiy Civil's avatar
Gennadiy Civil committed
45
```c++
46
switch(expression) {
Gennadiy Civil's avatar
Gennadiy Civil committed
47
48
49
50
51
52
  case 1:
     ... some checks ...
  case 2:
     ... some other checks ...
  default:
     FAIL() << "We shouldn't get here.";
53
54
55
}
```

Abseil Team's avatar
Abseil Team committed
56
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
57
58
NOTE: you can only use `FAIL()` in functions that return `void`. See the
[Assertion Placement section](#assertion-placement) for more information.
59

Gennadiy Civil's avatar
Gennadiy Civil committed
60
### Exception Assertions
61

Gennadiy Civil's avatar
Gennadiy Civil committed
62
63
These are for verifying that a piece of code throws (or does not throw) an
exception of the given type:
64

Gennadiy Civil's avatar
Gennadiy Civil committed
65
66
67
68
69
Fatal assertion                            | Nonfatal assertion                         | Verifies
------------------------------------------ | ------------------------------------------ | --------
`ASSERT_THROW(statement, exception_type);` | `EXPECT_THROW(statement, exception_type);` | `statement` throws an exception of the given type
`ASSERT_ANY_THROW(statement);`             | `EXPECT_ANY_THROW(statement);`             | `statement` throws an exception of any type
`ASSERT_NO_THROW(statement);`              | `EXPECT_NO_THROW(statement);`              | `statement` doesn't throw any exception
70
71
72

Examples:

Gennadiy Civil's avatar
Gennadiy Civil committed
73
```c++
74
75
76
77
78
79
80
81
ASSERT_THROW(Foo(5), bar_exception);

EXPECT_NO_THROW({
  int n = 5;
  Bar(&n);
});
```

82
**Availability**: requires exceptions to be enabled in the build environment
83

Gennadiy Civil's avatar
Gennadiy Civil committed
84
### Predicate Assertions for Better Error Messages
85

Gennadiy Civil's avatar
Gennadiy Civil committed
86
87
88
89
90
Even though googletest has a rich set of assertions, they can never be complete,
as it's impossible (nor a good idea) to anticipate all scenarios a user might
run into. Therefore, sometimes a user has to use `EXPECT_TRUE()` to check a
complex expression, for lack of a better macro. This has the problem of not
showing you the values of the parts of the expression, making it hard to
91
92
93
94
95
understand what went wrong. As a workaround, some users choose to construct the
failure message by themselves, streaming it into `EXPECT_TRUE()`. However, this
is awkward especially when the expression has side-effects or is expensive to
evaluate.

Gennadiy Civil's avatar
Gennadiy Civil committed
96
googletest gives you three different options to solve this problem:
97

Gennadiy Civil's avatar
Gennadiy Civil committed
98
#### Using an Existing Boolean Function
99

Gennadiy Civil's avatar
Gennadiy Civil committed
100
101
102
If you already have a function or functor that returns `bool` (or a type that
can be implicitly converted to `bool`), you can use it in a *predicate
assertion* to get the function arguments printed for free:
103
104


105
106
107
| Fatal assertion                   | Nonfatal assertion                | Verifies                    |
| --------------------------------- | --------------------------------- | --------------------------- |
| `ASSERT_PRED1(pred1, val1)`       | `EXPECT_PRED1(pred1, val1)`       | `pred1(val1)` is true       |
Abseil Team's avatar
Abseil Team committed
108
| `ASSERT_PRED2(pred2, val1, val2)` | `EXPECT_PRED2(pred2, val1, val2)` | `pred2(val1, val2)` is true |
109
| `...`                             | `...`                             | `...`                       |
110

Gennadiy Civil's avatar
Gennadiy Civil committed
111
112
113
In the above, `predn` is an `n`-ary predicate function or functor, where `val1`,
`val2`, ..., and `valn` are its arguments. The assertion succeeds if the
predicate returns `true` when applied to the given arguments, and fails
114
115
116
117
118
otherwise. When the assertion fails, it prints the value of each argument. In
either case, the arguments are evaluated exactly once.

Here's an example. Given

Gennadiy Civil's avatar
Gennadiy Civil committed
119
```c++
Yi Zheng's avatar
Yi Zheng committed
120
// Returns true if m and n have no common divisors except 1.
121
bool MutuallyPrime(int m, int n) { ... }
Gennadiy Civil's avatar
Gennadiy Civil committed
122

123
124
125
126
127
const int a = 3;
const int b = 4;
const int c = 10;
```

Gennadiy Civil's avatar
Gennadiy Civil committed
128
129
130
131
132
133
134
135
136
137
138
the assertion

```c++
  EXPECT_PRED2(MutuallyPrime, a, b);
```

will succeed, while the assertion

```c++
  EXPECT_PRED2(MutuallyPrime, b, c);
```
139

Gennadiy Civil's avatar
Gennadiy Civil committed
140
will fail with the message
141

Gennadiy Civil's avatar
Gennadiy Civil committed
142
143
144
145
146
```none
MutuallyPrime(b, c) is false, where
b is 4
c is 10
```
147

Abseil Team's avatar
Abseil Team committed
148
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
149
150
151
152
> NOTE:
>
> 1.  If you see a compiler error "no matching function to call" when using
>     `ASSERT_PRED*` or `EXPECT_PRED*`, please see
Krystian Kuzniarek's avatar
Krystian Kuzniarek committed
153
>     [this](faq.md#the-compiler-complains-no-matching-function-to-call-when-i-use-assert_pred-how-do-i-fix-it)
154
155
>     for how to resolve it.

Gennadiy Civil's avatar
Gennadiy Civil committed
156
#### Using a Function That Returns an AssertionResult
157

Gennadiy Civil's avatar
Gennadiy Civil committed
158
159
160
161
While `EXPECT_PRED*()` and friends are handy for a quick job, the syntax is not
satisfactory: you have to use different macros for different arities, and it
feels more like Lisp than C++. The `::testing::AssertionResult` class solves
this problem.
162

Gennadiy Civil's avatar
Gennadiy Civil committed
163
164
165
An `AssertionResult` object represents the result of an assertion (whether it's
a success or a failure, and an associated message). You can create an
`AssertionResult` using one of these factory functions:
166

Gennadiy Civil's avatar
Gennadiy Civil committed
167
```c++
168
169
170
171
172
173
174
175
176
177
178
179
180
namespace testing {

// Returns an AssertionResult object to indicate that an assertion has
// succeeded.
AssertionResult AssertionSuccess();

// Returns an AssertionResult object to indicate that an assertion has
// failed.
AssertionResult AssertionFailure();

}
```

Gennadiy Civil's avatar
Gennadiy Civil committed
181
182
You can then use the `<<` operator to stream messages to the `AssertionResult`
object.
183

Gennadiy Civil's avatar
Gennadiy Civil committed
184
185
186
To provide more readable messages in Boolean assertions (e.g. `EXPECT_TRUE()`),
write a predicate function that returns `AssertionResult` instead of `bool`. For
example, if you define `IsEven()` as:
187

Gennadiy Civil's avatar
Gennadiy Civil committed
188
```c++
Abseil Team's avatar
Abseil Team committed
189
testing::AssertionResult IsEven(int n) {
190
  if ((n % 2) == 0)
Abseil Team's avatar
Abseil Team committed
191
    return testing::AssertionSuccess();
192
  else
Abseil Team's avatar
Abseil Team committed
193
    return testing::AssertionFailure() << n << " is odd";
194
195
196
197
198
}
```

instead of:

Gennadiy Civil's avatar
Gennadiy Civil committed
199
```c++
200
201
202
203
204
205
206
bool IsEven(int n) {
  return (n % 2) == 0;
}
```

the failed assertion `EXPECT_TRUE(IsEven(Fib(4)))` will print:

Gennadiy Civil's avatar
Gennadiy Civil committed
207
208
209
210
211
```none
Value of: IsEven(Fib(4))
  Actual: false (3 is odd)
Expected: true
```
212
213
214

instead of a more opaque

Gennadiy Civil's avatar
Gennadiy Civil committed
215
216
217
218
219
```none
Value of: IsEven(Fib(4))
  Actual: false
Expected: true
```
220

Gennadiy Civil's avatar
Gennadiy Civil committed
221
222
223
224
If you want informative messages in `EXPECT_FALSE` and `ASSERT_FALSE` as well
(one third of Boolean assertions in the Google code base are negative ones), and
are fine with making the predicate slower in the success case, you can supply a
success message:
225

Gennadiy Civil's avatar
Gennadiy Civil committed
226
```c++
Abseil Team's avatar
Abseil Team committed
227
testing::AssertionResult IsEven(int n) {
228
  if ((n % 2) == 0)
Abseil Team's avatar
Abseil Team committed
229
    return testing::AssertionSuccess() << n << " is even";
230
  else
Abseil Team's avatar
Abseil Team committed
231
    return testing::AssertionFailure() << n << " is odd";
232
233
234
235
236
}
```

Then the statement `EXPECT_FALSE(IsEven(Fib(6)))` will print

Gennadiy Civil's avatar
Gennadiy Civil committed
237
238
239
240
241
```none
  Value of: IsEven(Fib(6))
     Actual: true (8 is even)
  Expected: false
```
242

Gennadiy Civil's avatar
Gennadiy Civil committed
243
#### Using a Predicate-Formatter
244
245
246
247

If you find the default message generated by `(ASSERT|EXPECT)_PRED*` and
`(ASSERT|EXPECT)_(TRUE|FALSE)` unsatisfactory, or some arguments to your
predicate do not support streaming to `ostream`, you can instead use the
Gennadiy Civil's avatar
Gennadiy Civil committed
248
249
following *predicate-formatter assertions* to *fully* customize how the message
is formatted:
250

Gennadiy Civil's avatar
Gennadiy Civil committed
251
252
253
254
255
Fatal assertion                                  | Nonfatal assertion                               | Verifies
------------------------------------------------ | ------------------------------------------------ | --------
`ASSERT_PRED_FORMAT1(pred_format1, val1);`       | `EXPECT_PRED_FORMAT1(pred_format1, val1);`       | `pred_format1(val1)` is successful
`ASSERT_PRED_FORMAT2(pred_format2, val1, val2);` | `EXPECT_PRED_FORMAT2(pred_format2, val1, val2);` | `pred_format2(val1, val2)` is successful
`...`                                            | `...`                                            | ...
256

Gennadiy Civil's avatar
Gennadiy Civil committed
257
258
259
The difference between this and the previous group of macros is that instead of
a predicate, `(ASSERT|EXPECT)_PRED_FORMAT*` take a *predicate-formatter*
(`pred_formatn`), which is a function or functor with the signature:
260

Gennadiy Civil's avatar
Gennadiy Civil committed
261
```c++
Abseil Team's avatar
Abseil Team committed
262
263
264
265
266
267
268
269
testing::AssertionResult PredicateFormattern(const char* expr1,
                                             const char* expr2,
                                             ...
                                             const char* exprn,
                                             T1 val1,
                                             T2 val2,
                                             ...
                                             Tn valn);
Gennadiy Civil's avatar
Gennadiy Civil committed
270
```
271

Gennadiy Civil's avatar
Gennadiy Civil committed
272
273
274
275
276
where `val1`, `val2`, ..., and `valn` are the values of the predicate arguments,
and `expr1`, `expr2`, ..., and `exprn` are the corresponding expressions as they
appear in the source code. The types `T1`, `T2`, ..., and `Tn` can be either
value types or reference types. For example, if an argument has type `Foo`, you
can declare it as either `Foo` or `const Foo&`, whichever is appropriate.
277

Gennadiy Civil's avatar
Gennadiy Civil committed
278
279
As an example, let's improve the failure message in `MutuallyPrime()`, which was
used with `EXPECT_PRED2()`:
280

Gennadiy Civil's avatar
Gennadiy Civil committed
281
```c++
282
283
284
285
286
// Returns the smallest prime common divisor of m and n,
// or 1 when m and n are mutually prime.
int SmallestPrimeCommonDivisor(int m, int n) { ... }

// A predicate-formatter for asserting that two integers are mutually prime.
Abseil Team's avatar
Abseil Team committed
287
288
289
290
291
testing::AssertionResult AssertMutuallyPrime(const char* m_expr,
                                             const char* n_expr,
                                             int m,
                                             int n) {
  if (MutuallyPrime(m, n)) return testing::AssertionSuccess();
Caleb Morse's avatar
Caleb Morse committed
292

Abseil Team's avatar
Abseil Team committed
293
  return testing::AssertionFailure() << m_expr << " and " << n_expr
Gennadiy Civil's avatar
Gennadiy Civil committed
294
295
      << " (" << m << " and " << n << ") are not mutually prime, "
      << "as they have a common divisor " << SmallestPrimeCommonDivisor(m, n);
296
297
298
299
300
}
```

With this predicate-formatter, we can use

Gennadiy Civil's avatar
Gennadiy Civil committed
301
302
```c++
  EXPECT_PRED_FORMAT2(AssertMutuallyPrime, b, c);
303
304
305
306
```

to generate the message

Gennadiy Civil's avatar
Gennadiy Civil committed
307
308
309
```none
b and c (4 and 10) are not mutually prime, as they have a common divisor 2.
```
310

Gennadiy Civil's avatar
Gennadiy Civil committed
311
312
As you may have realized, many of the built-in assertions we introduced earlier
are special cases of `(EXPECT|ASSERT)_PRED_FORMAT*`. In fact, most of them are
313
314
indeed defined using `(EXPECT|ASSERT)_PRED_FORMAT*`.

Gennadiy Civil's avatar
Gennadiy Civil committed
315
### Floating-Point Comparison
316

Gennadiy Civil's avatar
Gennadiy Civil committed
317
318
319
320
321
322
Comparing floating-point numbers is tricky. Due to round-off errors, it is very
unlikely that two floating-points will match exactly. Therefore, `ASSERT_EQ` 's
naive comparison usually doesn't work. And since floating-points can have a wide
value range, no single fixed error bound works. It's better to compare by a
fixed relative error bound, except for values close to 0 due to the loss of
precision there.
323
324
325

In general, for floating-point comparison to make sense, the user needs to
carefully choose the error bound. If they don't want or care to, comparing in
Gennadiy Civil's avatar
Gennadiy Civil committed
326
terms of Units in the Last Place (ULPs) is a good default, and googletest
327
328
provides assertions to do this. Full details about ULPs are quite long; if you
want to learn more, see
Gennadiy Civil's avatar
Gennadiy Civil committed
329
330
331
[here](https://randomascii.wordpress.com/2012/02/25/comparing-floating-point-numbers-2012-edition/).

#### Floating-Point Macros
332

333

334
335
336
337
| Fatal assertion                 | Nonfatal assertion              | Verifies                                 |
| ------------------------------- | ------------------------------- | ---------------------------------------- |
| `ASSERT_FLOAT_EQ(val1, val2);`  | `EXPECT_FLOAT_EQ(val1, val2);`  | the two `float` values are almost equal  |
| `ASSERT_DOUBLE_EQ(val1, val2);` | `EXPECT_DOUBLE_EQ(val1, val2);` | the two `double` values are almost equal |
338
339


Gennadiy Civil's avatar
Gennadiy Civil committed
340
By "almost equal" we mean the values are within 4 ULP's from each other.
341
342
343

The following assertions allow you to choose the acceptable error bound:

344

345
346
347
| Fatal assertion                       | Nonfatal assertion                    | Verifies                                                                         |
| ------------------------------------- | ------------------------------------- | -------------------------------------------------------------------------------- |
| `ASSERT_NEAR(val1, val2, abs_error);` | `EXPECT_NEAR(val1, val2, abs_error);` | the difference between `val1` and `val2` doesn't exceed the given absolute error |
348
349


Gennadiy Civil's avatar
Gennadiy Civil committed
350
#### Floating-Point Predicate-Format Functions
351

Gennadiy Civil's avatar
Gennadiy Civil committed
352
353
354
355
Some floating-point operations are useful, but not that often used. In order to
avoid an explosion of new macros, we provide them as predicate-format functions
that can be used in predicate assertion macros (e.g. `EXPECT_PRED_FORMAT2`,
etc).
356

Gennadiy Civil's avatar
Gennadiy Civil committed
357
```c++
Abseil Team's avatar
Abseil Team committed
358
359
EXPECT_PRED_FORMAT2(testing::FloatLE, val1, val2);
EXPECT_PRED_FORMAT2(testing::DoubleLE, val1, val2);
360
361
```

Gennadiy Civil's avatar
Gennadiy Civil committed
362
363
Verifies that `val1` is less than, or almost equal to, `val2`. You can replace
`EXPECT_PRED_FORMAT2` in the above table with `ASSERT_PRED_FORMAT2`.
364

Gennadiy Civil's avatar
Gennadiy Civil committed
365
366
### Asserting Using gMock Matchers

Abseil Team's avatar
Abseil Team committed
367
[gMock](gmock_for_dummies.md) comes with
Abseil Team's avatar
Abseil Team committed
368
[a library of matchers](gmock_cheat_sheet.md#MatcherList) for
Abseil Team's avatar
Abseil Team committed
369
370
371
validating arguments passed to mock objects. A gMock *matcher* is basically a
predicate that knows how to describe itself. It can be used in these assertion
macros:
Gennadiy Civil's avatar
Gennadiy Civil committed
372

373

374
375
376
| Fatal assertion                | Nonfatal assertion             | Verifies              |
| ------------------------------ | ------------------------------ | --------------------- |
| `ASSERT_THAT(value, matcher);` | `EXPECT_THAT(value, matcher);` | value matches matcher |
377
378


Gennadiy Civil's avatar
Gennadiy Civil committed
379
380
381
382
383
384
385
386
387
388
For example, `StartsWith(prefix)` is a matcher that matches a string starting
with `prefix`, and you can write:

```c++
using ::testing::StartsWith;
...
    // Verifies that Foo() returns a string starting with "Hello".
    EXPECT_THAT(Foo(), StartsWith("Hello"));
```

389
Read this
Abseil Team's avatar
Abseil Team committed
390
[recipe](gmock_cook_book.md#using-matchers-in-googletest-assertions)
391
in the gMock Cookbook for more details.
Gennadiy Civil's avatar
Gennadiy Civil committed
392
393
394

gMock has a rich set of matchers. You can do many things googletest cannot do
alone with them. For a list of matchers gMock provides, read
Abseil Team's avatar
Abseil Team committed
395
396
[this](gmock_cook_book.md##using-matchers). It's easy to write
your [own matchers](gmock_cook_book.md#NewMatchers) too.
Gennadiy Civil's avatar
Gennadiy Civil committed
397
398

gMock is bundled with googletest, so you don't need to add any build dependency
Abseil Team's avatar
Abseil Team committed
399
in order to take advantage of this. Just include `"gmock/gmock.h"`
Gennadiy Civil's avatar
Gennadiy Civil committed
400
401
402
403
and you're ready to go.

### More String Assertions

Abseil Team's avatar
Abseil Team committed
404
405
(Please read the [previous](#asserting-using-gmock-matchers) section first if
you haven't.)
Gennadiy Civil's avatar
Gennadiy Civil committed
406

407
You can use the gMock
Abseil Team's avatar
Abseil Team committed
408
[string matchers](gmock_cheat_sheet.md#string-matchers) with
409
`EXPECT_THAT()` or `ASSERT_THAT()` to do more string comparison tricks
Gennadiy Civil's avatar
Gennadiy Civil committed
410
411
412
413
414
415
416
417
418
419
420
(sub-string, prefix, suffix, regular expression, and etc). For example,

```c++
using ::testing::HasSubstr;
using ::testing::MatchesRegex;
...
  ASSERT_THAT(foo_string, HasSubstr("needle"));
  EXPECT_THAT(bar_string, MatchesRegex("\\w*\\d+"));
```

If the string contains a well-formed HTML or XML document, you can check whether
421
422
its DOM tree matches an
[XPath expression](http://www.w3.org/TR/xpath/#contents):
Gennadiy Civil's avatar
Gennadiy Civil committed
423
424
425
426

```c++
// Currently still in //template/prototemplate/testing:xpath_matcher
#include "template/prototemplate/testing/xpath_matcher.h"
Abseil Team's avatar
Abseil Team committed
427
using ::prototemplate::testing::MatchesXPath;
Gennadiy Civil's avatar
Gennadiy Civil committed
428
429
430
431
EXPECT_THAT(html_string, MatchesXPath("//a[text()='click here']"));
```

### Windows HRESULT assertions
432
433
434

These assertions test for `HRESULT` success or failure.

Gennadiy Civil's avatar
Gennadiy Civil committed
435
436
437
438
Fatal assertion                        | Nonfatal assertion                     | Verifies
-------------------------------------- | -------------------------------------- | --------
`ASSERT_HRESULT_SUCCEEDED(expression)` | `EXPECT_HRESULT_SUCCEEDED(expression)` | `expression` is a success `HRESULT`
`ASSERT_HRESULT_FAILED(expression)`    | `EXPECT_HRESULT_FAILED(expression)`    | `expression` is a failure `HRESULT`
439

Gennadiy Civil's avatar
Gennadiy Civil committed
440
441
The generated output contains the human-readable error message associated with
the `HRESULT` code returned by `expression`.
442
443
444

You might use them like this:

Gennadiy Civil's avatar
Gennadiy Civil committed
445
446
```c++
CComPtr<IShellDispatch2> shell;
447
448
449
450
451
ASSERT_HRESULT_SUCCEEDED(shell.CoCreateInstance(L"Shell.Application"));
CComVariant empty;
ASSERT_HRESULT_SUCCEEDED(shell->ShellExecute(CComBSTR(url), empty, empty, empty, empty));
```

Gennadiy Civil's avatar
Gennadiy Civil committed
452
### Type Assertions
453
454

You can call the function
Gennadiy Civil's avatar
Gennadiy Civil committed
455
456

```c++
457
458
459
::testing::StaticAssertTypeEq<T1, T2>();
```

Gennadiy Civil's avatar
Gennadiy Civil committed
460
461
to assert that types `T1` and `T2` are the same. The function does nothing if
the assertion is satisfied. If the types are different, the function call will
kuzkry's avatar
kuzkry committed
462
fail to compile, the compiler error message will say that
463
`T1 and T2 are not the same type` and most likely (depending on the compiler)
kuzkry's avatar
kuzkry committed
464
465
show you the actual values of `T1` and `T2`. This is mainly useful inside
template code.
Gennadiy Civil's avatar
Gennadiy Civil committed
466
467
468
469
470
471

**Caveat**: When used inside a member function of a class template or a function
template, `StaticAssertTypeEq<T1, T2>()` is effective only if the function is
instantiated. For example, given:

```c++
472
473
template <typename T> class Foo {
 public:
Abseil Team's avatar
Abseil Team committed
474
  void Bar() { testing::StaticAssertTypeEq<int, T>(); }
475
476
};
```
Gennadiy Civil's avatar
Gennadiy Civil committed
477

478
the code:
Gennadiy Civil's avatar
Gennadiy Civil committed
479
480

```c++
481
482
void Test1() { Foo<bool> foo; }
```
Gennadiy Civil's avatar
Gennadiy Civil committed
483
484
485
486
487

will not generate a compiler error, as `Foo<bool>::Bar()` is never actually
instantiated. Instead, you need:

```c++
488
489
void Test2() { Foo<bool> foo; foo.Bar(); }
```
Gennadiy Civil's avatar
Gennadiy Civil committed
490

491
492
to cause a compiler error.

Gennadiy Civil's avatar
Gennadiy Civil committed
493
### Assertion Placement
494

Gennadiy Civil's avatar
Gennadiy Civil committed
495
496
497
498
499
500
501
502
You can use assertions in any C++ function. In particular, it doesn't have to be
a method of the test fixture class. The one constraint is that assertions that
generate a fatal failure (`FAIL*` and `ASSERT_*`) can only be used in
void-returning functions. This is a consequence of Google's not using
exceptions. By placing it in a non-void function you'll get a confusing compile
error like `"error: void value not ignored as it ought to be"` or `"cannot
initialize return object of type 'bool' with an rvalue of type 'void'"` or
`"error: no viable conversion from 'void' to 'string'"`.
503

Gennadiy Civil's avatar
Gennadiy Civil committed
504
505
If you need to use fatal assertions in a function that returns non-void, one
option is to make the function return the value in an out parameter instead. For
506
507
508
509
510
example, you can rewrite `T2 Foo(T1 x)` to `void Foo(T1 x, T2* result)`. You
need to make sure that `*result` contains some sensible value even when the
function returns prematurely. As the function now returns `void`, you can use
any assertion inside of it.

Gennadiy Civil's avatar
Gennadiy Civil committed
511
512
If changing the function's type is not an option, you should just use assertions
that generate non-fatal failures, such as `ADD_FAILURE*` and `EXPECT_*`.
513

Abseil Team's avatar
Abseil Team committed
514
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
515
516
NOTE: Constructors and destructors are not considered void-returning functions,
according to the C++ language specification, and so you may not use fatal
517
518
519
520
521
assertions in them; you'll get a compilation error if you try. Instead, either
call `abort` and crash the entire test executable, or put the fatal assertion in
a `SetUp`/`TearDown` function; see
[constructor/destructor vs. `SetUp`/`TearDown`](faq.md#CtorVsSetUp)

Abseil Team's avatar
Abseil Team committed
522
{: .callout .warning}
523
WARNING: A fatal assertion in a helper function (private void-returning method)
Abseil Team's avatar
Abseil Team committed
524
525
called from a constructor or destructor does not terminate the current test, as
your intuition might suggest: it merely returns from the constructor or
526
527
528
destructor early, possibly leaving your object in a partially-constructed or
partially-destructed state! You almost certainly want to `abort` or use
`SetUp`/`TearDown` instead.
529

530
531
## Skipping test execution

532
533
Related to the assertions `SUCCEED()` and `FAIL()`, you can prevent further test
execution at runtime with the `GTEST_SKIP()` macro. This is useful when you need
534
535
to check for preconditions of the system under test during runtime and skip
tests in a meaningful way.
536

537
538
539
`GTEST_SKIP()` can be used in individual test cases or in the `SetUp()` methods
of classes derived from either `::testing::Environment` or `::testing::Test`.
For example:
540
541

```c++
542
543
544
545
546
547
TEST(SkipTest, DoesSkip) {
  GTEST_SKIP() << "Skipping single test";
  EXPECT_EQ(0, 1);  // Won't fail; it won't be executed
}

class SkipFixture : public ::testing::Test {
548
549
 protected:
  void SetUp() override {
550
    GTEST_SKIP() << "Skipping all tests for this fixture";
551
552
553
  }
};

554
555
556
// Tests for SkipFixture won't be executed.
TEST_F(SkipFixture, SkipsOneTest) {
  EXPECT_EQ(5, 7);  // Won't fail
557
558
559
}
```

560
As with assertion macros, you can stream a custom message into `GTEST_SKIP()`.
561

Gennadiy Civil's avatar
Gennadiy Civil committed
562
## Teaching googletest How to Print Your Values
563

Gennadiy Civil's avatar
Gennadiy Civil committed
564
565
When a test assertion such as `EXPECT_EQ` fails, googletest prints the argument
values to help you debug. It does this using a user-extensible value printer.
566
567

This printer knows how to print built-in C++ types, native arrays, STL
Gennadiy Civil's avatar
Gennadiy Civil committed
568
569
containers, and any type that supports the `<<` operator. For other types, it
prints the raw bytes in the value and hopes that you the user can figure it out.
570

Gennadiy Civil's avatar
Gennadiy Civil committed
571
572
573
As mentioned earlier, the printer is *extensible*. That means you can teach it
to do a better job at printing your particular type than to dump the bytes. To
do that, define `<<` for your type:
574

Gennadiy Civil's avatar
Gennadiy Civil committed
575
576
```c++
#include <ostream>
577
578
579

namespace foo {

Gennadiy Civil's avatar
Gennadiy Civil committed
580
581
582
class Bar {  // We want googletest to be able to print instances of this.
...
  // Create a free inline friend function.
583
  friend std::ostream& operator<<(std::ostream& os, const Bar& bar) {
Gennadiy Civil's avatar
Gennadiy Civil committed
584
585
586
    return os << bar.DebugString();  // whatever needed to print bar to os
  }
};
587

Gennadiy Civil's avatar
Gennadiy Civil committed
588
589
590
// If you can't declare the function in the class it's important that the
// << operator is defined in the SAME namespace that defines Bar.  C++'s look-up
// rules rely on that.
591
std::ostream& operator<<(std::ostream& os, const Bar& bar) {
592
593
594
595
596
597
  return os << bar.DebugString();  // whatever needed to print bar to os
}

}  // namespace foo
```

Gennadiy Civil's avatar
Gennadiy Civil committed
598
599
600
601
Sometimes, this might not be an option: your team may consider it bad style to
have a `<<` operator for `Bar`, or `Bar` may already have a `<<` operator that
doesn't do what you want (and you cannot change it). If so, you can instead
define a `PrintTo()` function like this:
602

Gennadiy Civil's avatar
Gennadiy Civil committed
603
604
```c++
#include <ostream>
605
606
607

namespace foo {

Gennadiy Civil's avatar
Gennadiy Civil committed
608
609
class Bar {
  ...
610
  friend void PrintTo(const Bar& bar, std::ostream* os) {
Gennadiy Civil's avatar
Gennadiy Civil committed
611
612
613
    *os << bar.DebugString();  // whatever needed to print bar to os
  }
};
614

Gennadiy Civil's avatar
Gennadiy Civil committed
615
616
617
// If you can't declare the function in the class it's important that PrintTo()
// is defined in the SAME namespace that defines Bar.  C++'s look-up rules rely
// on that.
618
void PrintTo(const Bar& bar, std::ostream* os) {
619
620
621
622
623
624
  *os << bar.DebugString();  // whatever needed to print bar to os
}

}  // namespace foo
```

Gennadiy Civil's avatar
Gennadiy Civil committed
625
626
627
628
If you have defined both `<<` and `PrintTo()`, the latter will be used when
googletest is concerned. This allows you to customize how the value appears in
googletest's output without affecting code that relies on the behavior of its
`<<` operator.
629

Gennadiy Civil's avatar
Gennadiy Civil committed
630
631
If you want to print a value `x` using googletest's value printer yourself, just
call `::testing::PrintToString(x)`, which returns an `std::string`:
632

Gennadiy Civil's avatar
Gennadiy Civil committed
633
```c++
634
635
636
vector<pair<Bar, int> > bar_ints = GetBarIntVector();

EXPECT_TRUE(IsCorrectBarIntVector(bar_ints))
Abseil Team's avatar
Abseil Team committed
637
    << "bar_ints = " << testing::PrintToString(bar_ints);
638
639
```

Gennadiy Civil's avatar
Gennadiy Civil committed
640
## Death Tests
641

Gennadiy Civil's avatar
Gennadiy Civil committed
642
643
644
645
646
647
648
In many applications, there are assertions that can cause application failure if
a condition is not met. These sanity checks, which ensure that the program is in
a known good state, are there to fail at the earliest possible time after some
program state is corrupted. If the assertion checks the wrong condition, then
the program may proceed in an erroneous state, which could lead to memory
corruption, security holes, or worse. Hence it is vitally important to test that
such assertion statements work as expected.
649
650
651
652
653
654

Since these precondition checks cause the processes to die, we call such tests
_death tests_. More generally, any test that checks that a program terminates
(except by throwing an exception) in an expected fashion is also a death test.

Note that if a piece of code throws an exception, we don't consider it "death"
Gennadiy Civil's avatar
Gennadiy Civil committed
655
656
for the purpose of death tests, as the caller of the code could catch the
exception and avoid the crash. If you want to verify exceptions thrown by your
657
code, see [Exception Assertions](#ExceptionAssertions).
658

Gennadiy Civil's avatar
Gennadiy Civil committed
659
660
If you want to test `EXPECT_*()/ASSERT_*()` failures in your test code, see
Catching Failures
661

Gennadiy Civil's avatar
Gennadiy Civil committed
662
### How to Write a Death Test
663

Gennadiy Civil's avatar
Gennadiy Civil committed
664
googletest has the following macros to support death tests:
665

666
667
668
669
Fatal assertion                                  | Nonfatal assertion                               | Verifies
------------------------------------------------ | ------------------------------------------------ | --------
`ASSERT_DEATH(statement, matcher);`              | `EXPECT_DEATH(statement, matcher);`              | `statement` crashes with the given error
`ASSERT_DEATH_IF_SUPPORTED(statement, matcher);` | `EXPECT_DEATH_IF_SUPPORTED(statement, matcher);` | if death tests are supported, verifies that `statement` crashes with the given error; otherwise verifies nothing
Abseil Team's avatar
Abseil Team committed
670
`ASSERT_DEBUG_DEATH(statement, matcher);`        | `EXPECT_DEBUG_DEATH(statement, matcher);`        | `statement` crashes with the given error **in debug mode**. When not in debug (i.e. `NDEBUG` is defined), this just executes `statement`
671
`ASSERT_EXIT(statement, predicate, matcher);`    | `EXPECT_EXIT(statement, predicate, matcher);`    | `statement` exits with the given error, and its exit code matches `predicate`
672

Gennadiy Civil's avatar
Gennadiy Civil committed
673
674
where `statement` is a statement that is expected to cause the process to die,
`predicate` is a function or function object that evaluates an integer exit
hyuk.myeong's avatar
hyuk.myeong committed
675
status, and `matcher` is either a gMock matcher matching a `const std::string&`
676
677
678
679
680
or a (Perl) regular expression - either of which is matched against the stderr
output of `statement`. For legacy reasons, a bare string (i.e. with no matcher)
is interpreted as `ContainsRegex(str)`, **not** `Eq(str)`. Note that `statement`
can be *any valid statement* (including *compound statement*) and doesn't have
to be an expression.
Gennadiy Civil's avatar
Gennadiy Civil committed
681

682
683
684
As usual, the `ASSERT` variants abort the current test function, while the
`EXPECT` variants do not.

Abseil Team's avatar
Abseil Team committed
685
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
686
687
688
689
690
> NOTE: We use the word "crash" here to mean that the process terminates with a
> *non-zero* exit status code. There are two possibilities: either the process
> has called `exit()` or `_exit()` with a non-zero value, or it may be killed by
> a signal.
>
hyuk.myeong's avatar
hyuk.myeong committed
691
> This means that if *`statement`* terminates the process with a 0 exit code, it
Gennadiy Civil's avatar
Gennadiy Civil committed
692
693
> is *not* considered a crash by `EXPECT_DEATH`. Use `EXPECT_EXIT` instead if
> this is the case, or if you want to restrict the exit code more precisely.
694
695

A predicate here must accept an `int` and return a `bool`. The death test
Gennadiy Civil's avatar
Gennadiy Civil committed
696
succeeds only if the predicate returns `true`. googletest defines a few
697
698
predicates that handle the most common cases:

Gennadiy Civil's avatar
Gennadiy Civil committed
699
```c++
700
701
702
703
704
705
::testing::ExitedWithCode(exit_code)
```

This expression is `true` if the program exited normally with the given exit
code.

Gennadiy Civil's avatar
Gennadiy Civil committed
706
```c++
Abseil Team's avatar
Abseil Team committed
707
testing::KilledBySignal(signal_number)  // Not available on Windows.
708
709
710
711
712
713
714
715
716
```

This expression is `true` if the program was killed by the given signal.

The `*_DEATH` macros are convenient wrappers for `*_EXIT` that use a predicate
that verifies the process' exit code is non-zero.

Note that a death test only cares about three things:

Gennadiy Civil's avatar
Gennadiy Civil committed
717
718
719
720
1.  does `statement` abort or exit the process?
2.  (in the case of `ASSERT_EXIT` and `EXPECT_EXIT`) does the exit status
    satisfy `predicate`? Or (in the case of `ASSERT_DEATH` and `EXPECT_DEATH`)
    is the exit status non-zero? And
hyuk.myeong's avatar
hyuk.myeong committed
721
3.  does the stderr output match `matcher`?
722

Gennadiy Civil's avatar
Gennadiy Civil committed
723
724
725
In particular, if `statement` generates an `ASSERT_*` or `EXPECT_*` failure, it
will **not** cause the death test to fail, as googletest assertions don't abort
the process.
726
727
728
729

To write a death test, simply use one of the above macros inside your test
function. For example,

Gennadiy Civil's avatar
Gennadiy Civil committed
730
```c++
731
732
TEST(MyDeathTest, Foo) {
  // This death test uses a compound statement.
Gennadiy Civil's avatar
Gennadiy Civil committed
733
734
735
736
  ASSERT_DEATH({
    int n = 5;
    Foo(&n);
  }, "Error on line .* of Foo()");
737
}
Gennadiy Civil's avatar
Gennadiy Civil committed
738

739
TEST(MyDeathTest, NormalExit) {
Abseil Team's avatar
Abseil Team committed
740
  EXPECT_EXIT(NormalExit(), testing::ExitedWithCode(0), "Success");
741
}
Gennadiy Civil's avatar
Gennadiy Civil committed
742

743
TEST(MyDeathTest, KillMyself) {
Abseil Team's avatar
Abseil Team committed
744
  EXPECT_EXIT(KillMyself(), testing::KilledBySignal(SIGKILL),
Gennadiy Civil's avatar
Gennadiy Civil committed
745
              "Sending myself unblockable signal");
746
747
748
749
750
}
```

verifies that:

Gennadiy Civil's avatar
Gennadiy Civil committed
751
752
753
754
*   calling `Foo(5)` causes the process to die with the given error message,
*   calling `NormalExit()` causes the process to print `"Success"` to stderr and
    exit with exit code 0, and
*   calling `KillMyself()` kills the process with signal `SIGKILL`.
755
756
757
758

The test function body may contain other assertions and statements as well, if
necessary.

Gennadiy Civil's avatar
Gennadiy Civil committed
759
### Death Test Naming
760

Abseil Team's avatar
Abseil Team committed
761
{: .callout .important}
Gennadiy Civil's avatar
Gennadiy Civil committed
762
IMPORTANT: We strongly recommend you to follow the convention of naming your
763
**test suite** (not test) `*DeathTest` when it contains a death test, as
764
765
demonstrated in the above example. The
[Death Tests And Threads](#death-tests-and-threads) section below explains why.
Gennadiy Civil's avatar
Gennadiy Civil committed
766
767
768

If a test fixture class is shared by normal tests and death tests, you can use
`using` or `typedef` to introduce an alias for the fixture class and avoid
769
duplicating its code:
Gennadiy Civil's avatar
Gennadiy Civil committed
770
771

```c++
Abseil Team's avatar
Abseil Team committed
772
class FooTest : public testing::Test { ... };
773

Gennadiy Civil's avatar
Gennadiy Civil committed
774
using FooDeathTest = FooTest;
775
776
777
778
779
780
781
782
783
784

TEST_F(FooTest, DoesThis) {
  // normal test
}

TEST_F(FooDeathTest, DoesThat) {
  // death test
}
```

Gennadiy Civil's avatar
Gennadiy Civil committed
785
### Regular Expression Syntax
786

Gennadiy Civil's avatar
Gennadiy Civil committed
787
On POSIX systems (e.g. Linux, Cygwin, and Mac), googletest uses the
788
[POSIX extended regular expression](http://www.opengroup.org/onlinepubs/009695399/basedefs/xbd_chap09.html#tag_09_04)
Gennadiy Civil's avatar
Gennadiy Civil committed
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
syntax. To learn about this syntax, you may want to read this
[Wikipedia entry](http://en.wikipedia.org/wiki/Regular_expression#POSIX_Extended_Regular_Expressions).

On Windows, googletest uses its own simple regular expression implementation. It
lacks many features. For example, we don't support union (`"x|y"`), grouping
(`"(xy)"`), brackets (`"[xy]"`), and repetition count (`"x{5,7}"`), among
others. Below is what we do support (`A` denotes a literal character, period
(`.`), or a single `\\ ` escape sequence; `x` and `y` denote regular
expressions.):

Expression | Meaning
---------- | --------------------------------------------------------------
`c`        | matches any literal character `c`
`\\d`      | matches any decimal digit
`\\D`      | matches any character that's not a decimal digit
`\\f`      | matches `\f`
`\\n`      | matches `\n`
`\\r`      | matches `\r`
`\\s`      | matches any ASCII whitespace, including `\n`
`\\S`      | matches any character that's not a whitespace
`\\t`      | matches `\t`
`\\v`      | matches `\v`
`\\w`      | matches any letter, `_`, or decimal digit
`\\W`      | matches any character that `\\w` doesn't match
`\\c`      | matches any literal character `c`, which must be a punctuation
`.`        | matches any single character except `\n`
`A?`       | matches 0 or 1 occurrences of `A`
`A*`       | matches 0 or many occurrences of `A`
`A+`       | matches 1 or many occurrences of `A`
`^`        | matches the beginning of a string (not that of each line)
`$`        | matches the end of a string (not that of each line)
`xy`       | matches `x` followed by `y`

To help you determine which capability is available on your system, googletest
defines macros to govern which regular expression it is using. The macros are:
824
825
826
`GTEST_USES_SIMPLE_RE=1` or `GTEST_USES_POSIX_RE=1`. If you want your death
tests to work in all cases, you can either `#if` on these macros or use the more
limited syntax only.
Gennadiy Civil's avatar
Gennadiy Civil committed
827
828
829
830
831

### How It Works

Under the hood, `ASSERT_EXIT()` spawns a new process and executes the death test
statement in that process. The details of how precisely that happens depend on
Abseil Team's avatar
Abseil Team committed
832
the platform and the variable `::testing::GTEST_FLAG(death_test_style)` (which is
Gennadiy Civil's avatar
Gennadiy Civil committed
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
initialized from the command-line flag `--gtest_death_test_style`).

*   On POSIX systems, `fork()` (or `clone()` on Linux) is used to spawn the
    child, after which:
    *   If the variable's value is `"fast"`, the death test statement is
        immediately executed.
    *   If the variable's value is `"threadsafe"`, the child process re-executes
        the unit test binary just as it was originally invoked, but with some
        extra flags to cause just the single death test under consideration to
        be run.
*   On Windows, the child is spawned using the `CreateProcess()` API, and
    re-executes the binary to cause just the single death test under
    consideration to be run - much like the `threadsafe` mode on POSIX.

Other values for the variable are illegal and will cause the death test to fail.
Abseil Team's avatar
Abseil Team committed
848
849
Currently, the flag's default value is
**`"fast"`**.
Gennadiy Civil's avatar
Gennadiy Civil committed
850
851
852
853
854
855
856
857

1.  the child's exit status satisfies the predicate, and
2.  the child's stderr matches the regular expression.

If the death test statement runs to completion without dying, the child process
will nonetheless terminate, and the assertion fails.

### Death Tests And Threads
858
859
860
861
862
863
864
865

The reason for the two death test styles has to do with thread safety. Due to
well-known problems with forking in the presence of threads, death tests should
be run in a single-threaded context. Sometimes, however, it isn't feasible to
arrange that kind of environment. For example, statically-initialized modules
may start threads before main is ever reached. Once threads have been created,
it may be difficult or impossible to clean them up.

Gennadiy Civil's avatar
Gennadiy Civil committed
866
googletest has three features intended to raise awareness of threading issues.
867

Gennadiy Civil's avatar
Gennadiy Civil committed
868
869
1.  A warning is emitted if multiple threads are running when a death test is
    encountered.
870
871
2.  Test suites with a name ending in "DeathTest" are run before all other
    tests.
Gennadiy Civil's avatar
Gennadiy Civil committed
872
873
874
3.  It uses `clone()` instead of `fork()` to spawn the child process on Linux
    (`clone()` is not available on Cygwin and Mac), as `fork()` is more likely
    to cause the child to hang when the parent process has multiple threads.
875
876
877
878

It's perfectly fine to create threads inside a death test statement; they are
executed in a separate process and cannot affect the parent.

Gennadiy Civil's avatar
Gennadiy Civil committed
879
880
### Death Test Styles

881
882
883
884
The "threadsafe" death test style was introduced in order to help mitigate the
risks of testing in a possibly multithreaded environment. It trades increased
test execution time (potentially dramatically so) for improved thread safety.

Gennadiy Civil's avatar
Gennadiy Civil committed
885
886
The automated testing framework does not set the style flag. You can choose a
particular style of death tests by setting the flag programmatically:
887

Gennadiy Civil's avatar
Gennadiy Civil committed
888
889
```c++
testing::FLAGS_gtest_death_test_style="threadsafe"
890
891
```

Gennadiy Civil's avatar
Gennadiy Civil committed
892
893
894
895
896
897
You can do this in `main()` to set the style for all death tests in the binary,
or in individual tests. Recall that flags are saved before running each test and
restored afterwards, so you need not do that yourself. For example:

```c++
int main(int argc, char** argv) {
Abseil Team's avatar
Abseil Team committed
898
899
  testing::InitGoogleTest(&argc, argv);
  testing::FLAGS_gtest_death_test_style = "fast";
Gennadiy Civil's avatar
Gennadiy Civil committed
900
901
  return RUN_ALL_TESTS();
}
902
903

TEST(MyDeathTest, TestOne) {
Abseil Team's avatar
Abseil Team committed
904
  testing::FLAGS_gtest_death_test_style = "threadsafe";
905
906
907
908
909
910
911
912
913
914
  // This test is run in the "threadsafe" style:
  ASSERT_DEATH(ThisShouldDie(), "");
}

TEST(MyDeathTest, TestTwo) {
  // This test is run in the "fast" style:
  ASSERT_DEATH(ThisShouldDie(), "");
}
```

Gennadiy Civil's avatar
Gennadiy Civil committed
915
### Caveats
916

Gennadiy Civil's avatar
Gennadiy Civil committed
917
918
919
920
921
922
923
924
The `statement` argument of `ASSERT_EXIT()` can be any valid C++ statement. If
it leaves the current function via a `return` statement or by throwing an
exception, the death test is considered to have failed. Some googletest macros
may return from the current function (e.g. `ASSERT_TRUE()`), so be sure to avoid
them in `statement`.

Since `statement` runs in the child process, any in-memory side effect (e.g.
modifying a variable, releasing memory, etc) it causes will *not* be observable
925
926
927
928
in the parent process. In particular, if you release memory in a death test,
your program will fail the heap check as the parent process will never see the
memory reclaimed. To solve this problem, you can

Gennadiy Civil's avatar
Gennadiy Civil committed
929
930
931
1.  try not to free memory in a death test;
2.  free the memory again in the parent process; or
3.  do not use the heap checker in your program.
932

Gennadiy Civil's avatar
Gennadiy Civil committed
933
934
935
Due to an implementation detail, you cannot place multiple death test assertions
on the same line; otherwise, compilation will fail with an unobvious error
message.
936
937
938
939
940
941

Despite the improved thread safety afforded by the "threadsafe" style of death
test, thread problems such as deadlock are still possible in the presence of
handlers registered with `pthread_atfork(3)`.


Gennadiy Civil's avatar
Gennadiy Civil committed
942
943
## Using Assertions in Sub-routines

Abseil Team's avatar
Abseil Team committed
944
{: .callout .note}
krzysio's avatar
krzysio committed
945
946
Note: If you want to put a series of test assertions in a subroutine to check
for a complex condition, consider using
Abseil Team's avatar
Abseil Team committed
947
[a custom GMock matcher](gmock_cook_book.md#NewMatchers)
krzysio's avatar
krzysio committed
948
949
950
instead. This lets you provide a more readable error message in case of failure
and avoid all of the issues described below.

Gennadiy Civil's avatar
Gennadiy Civil committed
951
### Adding Traces to Assertions
952

Gennadiy Civil's avatar
Gennadiy Civil committed
953
954
If a test sub-routine is called from several places, when an assertion inside it
fails, it can be hard to tell which invocation of the sub-routine the failure is
955
956
957
from. You can alleviate this problem using extra logging or custom failure
messages, but that usually clutters up your tests. A better solution is to use
the `SCOPED_TRACE` macro or the `ScopedTrace` utility:
958

Gennadiy Civil's avatar
Gennadiy Civil committed
959
960
```c++
SCOPED_TRACE(message);
Abseil Team's avatar
Abseil Team committed
961
962
```
```c++
Gennadiy Civil's avatar
Gennadiy Civil committed
963
964
ScopedTrace trace("file_path", line_number, message);
```
965

Fedor Trushkin's avatar
Fedor Trushkin committed
966
967
968
969
970
where `message` can be anything streamable to `std::ostream`. `SCOPED_TRACE`
macro will cause the current file name, line number, and the given message to be
added in every failure message. `ScopedTrace` accepts explicit file name and
line number in arguments, which is useful for writing test helpers. The effect
will be undone when the control leaves the current lexical scope.
971
972
973

For example,

Gennadiy Civil's avatar
Gennadiy Civil committed
974
```c++
975
10: void Sub1(int n) {
976
977
11:   EXPECT_EQ(Bar(n), 1);
12:   EXPECT_EQ(Bar(n + 1), 2);
978
13: }
Caleb Morse's avatar
Caleb Morse committed
979
14:
980
981
982
983
984
985
986
987
988
989
990
991
992
15: TEST(FooTest, Bar) {
16:   {
17:     SCOPED_TRACE("A");  // This trace point will be included in
18:                         // every failure in this scope.
19:     Sub1(1);
20:   }
21:   // Now it won't.
22:   Sub1(9);
23: }
```

could result in messages like these:

Gennadiy Civil's avatar
Gennadiy Civil committed
993
```none
994
995
996
997
path/to/foo_test.cc:11: Failure
Value of: Bar(n)
Expected: 1
  Actual: 2
Abseil Team's avatar
Abseil Team committed
998
Google Test trace:
999
1000
1001
1002
1003
1004
1005
1006
path/to/foo_test.cc:17: A

path/to/foo_test.cc:12: Failure
Value of: Bar(n + 1)
Expected: 2
  Actual: 3
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1007
Without the trace, it would've been difficult to know which invocation of
1008
1009
1010
`Sub1()` the two failures come from respectively. (You could add an extra
message to each assertion in `Sub1()` to indicate the value of `n`, but that's
tedious.)
1011
1012
1013

Some tips on using `SCOPED_TRACE`:

Gennadiy Civil's avatar
Gennadiy Civil committed
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1.  With a suitable message, it's often enough to use `SCOPED_TRACE` at the
    beginning of a sub-routine, instead of at each call site.
2.  When calling sub-routines inside a loop, make the loop iterator part of the
    message in `SCOPED_TRACE` such that you can know which iteration the failure
    is from.
3.  Sometimes the line number of the trace point is enough for identifying the
    particular invocation of a sub-routine. In this case, you don't have to
    choose a unique message for `SCOPED_TRACE`. You can simply use `""`.
4.  You can use `SCOPED_TRACE` in an inner scope when there is one in the outer
    scope. In this case, all active trace points will be included in the failure
    messages, in reverse order they are encountered.
5.  The trace dump is clickable in Emacs - hit `return` on a line number and
    you'll be taken to that line in the source file!
1027

Gennadiy Civil's avatar
Gennadiy Civil committed
1028
### Propagating Fatal Failures
1029
1030
1031
1032

A common pitfall when using `ASSERT_*` and `FAIL*` is not understanding that
when they fail they only abort the _current function_, not the entire test. For
example, the following test will segfault:
Gennadiy Civil's avatar
Gennadiy Civil committed
1033
1034

```c++
1035
1036
1037
void Subroutine() {
  // Generates a fatal failure and aborts the current function.
  ASSERT_EQ(1, 2);
Gennadiy Civil's avatar
Gennadiy Civil committed
1038

1039
1040
1041
1042
1043
  // The following won't be executed.
  ...
}

TEST(FooTest, Bar) {
Gennadiy Civil's avatar
Gennadiy Civil committed
1044
1045
1046
  Subroutine();  // The intended behavior is for the fatal failure
                 // in Subroutine() to abort the entire test.

1047
  // The actual behavior: the function goes on after Subroutine() returns.
Abseil Team's avatar
Abseil Team committed
1048
  int* p = nullptr;
Gennadiy Civil's avatar
Gennadiy Civil committed
1049
  *p = 3;  // Segfault!
1050
1051
1052
}
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1053
To alleviate this, googletest provides three different solutions. You could use
1054
1055
either exceptions, the `(ASSERT|EXPECT)_NO_FATAL_FAILURE` assertions or the
`HasFatalFailure()` function. They are described in the following two
1056
1057
subsections.

1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
#### Asserting on Subroutines with an exception

The following code can turn ASSERT-failure into an exception:

```c++
class ThrowListener : public testing::EmptyTestEventListener {
  void OnTestPartResult(const testing::TestPartResult& result) override {
    if (result.type() == testing::TestPartResult::kFatalFailure) {
      throw testing::AssertionException(result);
    }
  }
};
int main(int argc, char** argv) {
  ...
  testing::UnitTest::GetInstance()->listeners().Append(new ThrowListener);
  return RUN_ALL_TESTS();
}
```

This listener should be added after other listeners if you have any, otherwise
they won't see failed `OnTestPartResult`.

Gennadiy Civil's avatar
Gennadiy Civil committed
1080
#### Asserting on Subroutines
1081

Gennadiy Civil's avatar
Gennadiy Civil committed
1082
1083
1084
As shown above, if your test calls a subroutine that has an `ASSERT_*` failure
in it, the test will continue after the subroutine returns. This may not be what
you want.
1085

Gennadiy Civil's avatar
Gennadiy Civil committed
1086
1087
Often people want fatal failures to propagate like exceptions. For that
googletest offers the following macros:
1088

Gennadiy Civil's avatar
Gennadiy Civil committed
1089
1090
1091
Fatal assertion                       | Nonfatal assertion                    | Verifies
------------------------------------- | ------------------------------------- | --------
`ASSERT_NO_FATAL_FAILURE(statement);` | `EXPECT_NO_FATAL_FAILURE(statement);` | `statement` doesn't generate any new fatal failures in the current thread.
1092

Gennadiy Civil's avatar
Gennadiy Civil committed
1093
1094
1095
Only failures in the thread that executes the assertion are checked to determine
the result of this type of assertions. If `statement` creates new threads,
failures in these threads are ignored.
1096
1097
1098

Examples:

Gennadiy Civil's avatar
Gennadiy Civil committed
1099
```c++
1100
1101
1102
1103
1104
1105
1106
1107
ASSERT_NO_FATAL_FAILURE(Foo());

int i;
EXPECT_NO_FATAL_FAILURE({
  i = Bar();
});
```

1108
Assertions from multiple threads are currently not supported on Windows.
1109

Gennadiy Civil's avatar
Gennadiy Civil committed
1110
#### Checking for Failures in the Current Test
1111
1112

`HasFatalFailure()` in the `::testing::Test` class returns `true` if an
Gennadiy Civil's avatar
Gennadiy Civil committed
1113
1114
assertion in the current test has suffered a fatal failure. This allows
functions to catch fatal failures in a sub-routine and return early.
1115

Gennadiy Civil's avatar
Gennadiy Civil committed
1116
```c++
1117
1118
1119
1120
1121
1122
1123
class Test {
 public:
  ...
  static bool HasFatalFailure();
};
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1124
1125
The typical usage, which basically simulates the behavior of a thrown exception,
is:
1126

Gennadiy Civil's avatar
Gennadiy Civil committed
1127
```c++
1128
1129
1130
TEST(FooTest, Bar) {
  Subroutine();
  // Aborts if Subroutine() had a fatal failure.
Gennadiy Civil's avatar
Gennadiy Civil committed
1131
1132
  if (HasFatalFailure()) return;

1133
1134
1135
1136
1137
1138
1139
1140
  // The following won't be executed.
  ...
}
```

If `HasFatalFailure()` is used outside of `TEST()` , `TEST_F()` , or a test
fixture, you must add the `::testing::Test::` prefix, as in:

Gennadiy Civil's avatar
Gennadiy Civil committed
1141
```c++
Abseil Team's avatar
Abseil Team committed
1142
if (testing::Test::HasFatalFailure()) return;
1143
1144
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1145
1146
1147
Similarly, `HasNonfatalFailure()` returns `true` if the current test has at
least one non-fatal failure, and `HasFailure()` returns `true` if the current
test has at least one failure of either kind.
1148

Gennadiy Civil's avatar
Gennadiy Civil committed
1149
## Logging Additional Information
1150

Gennadiy Civil's avatar
Gennadiy Civil committed
1151
1152
In your test code, you can call `RecordProperty("key", value)` to log additional
information, where `value` can be either a string or an `int`. The *last* value
1153
1154
1155
recorded for a key will be emitted to the
[XML output](#generating-an-xml-report) if you specify one. For example, the
test
1156

Gennadiy Civil's avatar
Gennadiy Civil committed
1157
```c++
1158
1159
1160
1161
1162
1163
1164
1165
TEST_F(WidgetUsageTest, MinAndMaxWidgets) {
  RecordProperty("MaximumWidgets", ComputeMaxUsage());
  RecordProperty("MinimumWidgets", ComputeMinUsage());
}
```

will output XML like this:

Gennadiy Civil's avatar
Gennadiy Civil committed
1166
1167
1168
1169
```xml
  ...
    <testcase name="MinAndMaxWidgets" status="run" time="0.006" classname="WidgetUsageTest" MaximumWidgets="12" MinimumWidgets="9" />
  ...
1170
1171
```

Abseil Team's avatar
Abseil Team committed
1172
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
1173
1174
1175
1176
1177
> NOTE:
>
> *   `RecordProperty()` is a static member of the `Test` class. Therefore it
>     needs to be prefixed with `::testing::Test::` if used outside of the
>     `TEST` body and the test fixture class.
hyuk.myeong's avatar
hyuk.myeong committed
1178
> *   *`key`* must be a valid XML attribute name, and cannot conflict with the
Gennadiy Civil's avatar
Gennadiy Civil committed
1179
1180
1181
>     ones already used by googletest (`name`, `status`, `time`, `classname`,
>     `type_param`, and `value_param`).
> *   Calling `RecordProperty()` outside of the lifespan of a test is allowed.
1182
>     If it's called outside of a test but between a test suite's
1183
1184
1185
1186
1187
>     `SetUpTestSuite()` and `TearDownTestSuite()` methods, it will be
>     attributed to the XML element for the test suite. If it's called outside
>     of all test suites (e.g. in a test environment), it will be attributed to
>     the top-level XML element.

1188
## Sharing Resources Between Tests in the Same Test Suite
1189

Gennadiy Civil's avatar
Gennadiy Civil committed
1190
googletest creates a new test fixture object for each test in order to make
1191
1192
1193
1194
tests independent and easier to debug. However, sometimes tests use resources
that are expensive to set up, making the one-copy-per-test model prohibitively
expensive.

Gennadiy Civil's avatar
Gennadiy Civil committed
1195
1196
If the tests don't change the resource, there's no harm in their sharing a
single resource copy. So, in addition to per-test set-up/tear-down, googletest
1197
also supports per-test-suite set-up/tear-down. To use it:
1198

Gennadiy Civil's avatar
Gennadiy Civil committed
1199
1200
1.  In your test fixture class (say `FooTest` ), declare as `static` some member
    variables to hold the shared resources.
1201
2.  Outside your test fixture class (typically just below it), define those
Gennadiy Civil's avatar
Gennadiy Civil committed
1202
    member variables, optionally giving them initial values.
1203
3.  In the same test fixture class, define a `static void SetUpTestSuite()`
1204
1205
    function (remember not to spell it as **`SetupTestSuite`** with a small
    `u`!) to set up the shared resources and a `static void TearDownTestSuite()`
Gennadiy Civil's avatar
Gennadiy Civil committed
1206
1207
    function to tear them down.

1208
1209
1210
That's it! googletest automatically calls `SetUpTestSuite()` before running the
*first test* in the `FooTest` test suite (i.e. before creating the first
`FooTest` object), and calls `TearDownTestSuite()` after running the *last test*
Gennadiy Civil's avatar
Gennadiy Civil committed
1211
1212
in it (i.e. after deleting the last `FooTest` object). In between, the tests can
use the shared resources.
1213
1214

Remember that the test order is undefined, so your code can't depend on a test
Gennadiy Civil's avatar
Gennadiy Civil committed
1215
1216
1217
preceding or following another. Also, the tests must either not modify the state
of any shared resource, or, if they do modify the state, they must restore the
state to its original value before passing control to the next test.
1218

1219
Here's an example of per-test-suite set-up and tear-down:
Gennadiy Civil's avatar
Gennadiy Civil committed
1220
1221

```c++
Abseil Team's avatar
Abseil Team committed
1222
class FooTest : public testing::Test {
1223
 protected:
1224
1225
  // Per-test-suite set-up.
  // Called before the first test in this test suite.
1226
  // Can be omitted if not needed.
1227
  static void SetUpTestSuite() {
1228
1229
1230
    shared_resource_ = new ...;
  }

1231
1232
  // Per-test-suite tear-down.
  // Called after the last test in this test suite.
1233
  // Can be omitted if not needed.
1234
  static void TearDownTestSuite() {
1235
    delete shared_resource_;
Abseil Team's avatar
Abseil Team committed
1236
    shared_resource_ = nullptr;
1237
1238
  }

Gennadiy Civil's avatar
Gennadiy Civil committed
1239
  // You can define per-test set-up logic as usual.
Abseil Team's avatar
Abseil Team committed
1240
  void SetUp() override { ... }
Gennadiy Civil's avatar
Gennadiy Civil committed
1241
1242

  // You can define per-test tear-down logic as usual.
Abseil Team's avatar
Abseil Team committed
1243
  void TearDown() override { ... }
1244
1245
1246
1247
1248

  // Some expensive resource shared by all tests.
  static T* shared_resource_;
};

Abseil Team's avatar
Abseil Team committed
1249
T* FooTest::shared_resource_ = nullptr;
1250
1251

TEST_F(FooTest, Test1) {
Gennadiy Civil's avatar
Gennadiy Civil committed
1252
  ... you can refer to shared_resource_ here ...
1253
}
Gennadiy Civil's avatar
Gennadiy Civil committed
1254

1255
TEST_F(FooTest, Test2) {
Gennadiy Civil's avatar
Gennadiy Civil committed
1256
  ... you can refer to shared_resource_ here ...
1257
1258
1259
}
```

Abseil Team's avatar
Abseil Team committed
1260
{: .callout .note}
1261
NOTE: Though the above code declares `SetUpTestSuite()` protected, it may
Gennadiy Civil's avatar
Gennadiy Civil committed
1262
1263
1264
1265
sometimes be necessary to declare it public, such as when using it with
`TEST_P`.

## Global Set-Up and Tear-Down
1266

1267
Just as you can do set-up and tear-down at the test level and the test suite
1268
1269
1270
1271
1272
level, you can also do it at the test program level. Here's how.

First, you subclass the `::testing::Environment` class to define a test
environment, which knows how to set-up and tear-down:

Gennadiy Civil's avatar
Gennadiy Civil committed
1273
```c++
Abseil Team's avatar
Abseil Team committed
1274
class Environment : public ::testing::Environment {
1275
 public:
Abseil Team's avatar
Abseil Team committed
1276
  ~Environment() override {}
Gennadiy Civil's avatar
Gennadiy Civil committed
1277

1278
  // Override this to define how to set up the environment.
1279
  void SetUp() override {}
Gennadiy Civil's avatar
Gennadiy Civil committed
1280

1281
  // Override this to define how to tear down the environment.
1282
  void TearDown() override {}
1283
1284
1285
};
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1286
Then, you register an instance of your environment class with googletest by
1287
1288
calling the `::testing::AddGlobalTestEnvironment()` function:

Gennadiy Civil's avatar
Gennadiy Civil committed
1289
```c++
1290
1291
1292
1293
Environment* AddGlobalTestEnvironment(Environment* env);
```

Now, when `RUN_ALL_TESTS()` is called, it first calls the `SetUp()` method of
1294
1295
each environment object, then runs the tests if none of the environments
reported fatal failures and `GTEST_SKIP()` was not called. `RUN_ALL_TESTS()`
1296
1297
always calls `TearDown()` with each environment object, regardless of whether or
not the tests were run.
1298

1299
It's OK to register multiple environment objects. In this suite, their `SetUp()`
1300
1301
1302
will be called in the order they are registered, and their `TearDown()` will be
called in the reverse order.

Gennadiy Civil's avatar
Gennadiy Civil committed
1303
Note that googletest takes ownership of the registered environment objects.
1304
1305
Therefore **do not delete them** by yourself.

Gennadiy Civil's avatar
Gennadiy Civil committed
1306
1307
1308
1309
You should call `AddGlobalTestEnvironment()` before `RUN_ALL_TESTS()` is called,
probably in `main()`. If you use `gtest_main`, you need to call this before
`main()` starts for it to take effect. One way to do this is to define a global
variable like this:
1310

Gennadiy Civil's avatar
Gennadiy Civil committed
1311
```c++
Abseil Team's avatar
Abseil Team committed
1312
1313
testing::Environment* const foo_env =
    testing::AddGlobalTestEnvironment(new FooEnvironment);
1314
1315
1316
1317
```

However, we strongly recommend you to write your own `main()` and call
`AddGlobalTestEnvironment()` there, as relying on initialization of global
Gennadiy Civil's avatar
Gennadiy Civil committed
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
variables makes the code harder to read and may cause problems when you register
multiple environments from different translation units and the environments have
dependencies among them (remember that the compiler doesn't guarantee the order
in which global variables from different translation units are initialized).

## Value-Parameterized Tests

*Value-parameterized tests* allow you to test your code with different
parameters without writing multiple copies of the same test. This is useful in a
number of situations, for example:

*   You have a piece of code whose behavior is affected by one or more
    command-line flags. You want to make sure your code performs correctly for
    various values of those flags.
*   You want to test different implementations of an OO interface.
*   You want to test your code over various inputs (a.k.a. data-driven testing).
    This feature is easy to abuse, so please exercise your good sense when doing
    it!

### How to Write Value-Parameterized Tests

To write value-parameterized tests, first you should define a fixture class. It
1340
1341
1342
1343
1344
1345
1346
must be derived from both `testing::Test` and `testing::WithParamInterface<T>`
(the latter is a pure interface), where `T` is the type of your parameter
values. For convenience, you can just derive the fixture class from
`testing::TestWithParam<T>`, which itself is derived from both `testing::Test`
and `testing::WithParamInterface<T>`. `T` can be any copyable type. If it's a
raw pointer, you are responsible for managing the lifespan of the pointed
values.
Gennadiy Civil's avatar
Gennadiy Civil committed
1347

Abseil Team's avatar
Abseil Team committed
1348
{: .callout .note}
1349
NOTE: If your test fixture defines `SetUpTestSuite()` or `TearDownTestSuite()`
Gennadiy Civil's avatar
Gennadiy Civil committed
1350
1351
they must be declared **public** rather than **protected** in order to use
`TEST_P`.
1352

Gennadiy Civil's avatar
Gennadiy Civil committed
1353
1354
```c++
class FooTest :
1355
    public testing::TestWithParam<const char*> {
1356
1357
1358
1359
1360
1361
  // You can implement all the usual fixture class members here.
  // To access the test parameter, call GetParam() from class
  // TestWithParam<T>.
};

// Or, when you want to add parameters to a pre-existing fixture class:
1362
class BaseTest : public testing::Test {
1363
1364
1365
  ...
};
class BarTest : public BaseTest,
1366
                public testing::WithParamInterface<const char*> {
1367
1368
1369
1370
  ...
};
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1371
1372
1373
Then, use the `TEST_P` macro to define as many test patterns using this fixture
as you want. The `_P` suffix is for "parameterized" or "pattern", whichever you
prefer to think.
1374

Gennadiy Civil's avatar
Gennadiy Civil committed
1375
```c++
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
TEST_P(FooTest, DoesBlah) {
  // Inside a test, access the test parameter with the GetParam() method
  // of the TestWithParam<T> class:
  EXPECT_TRUE(foo.Blah(GetParam()));
  ...
}

TEST_P(FooTest, HasBlahBlah) {
  ...
}
```

1388
1389
1390
Finally, you can use `INSTANTIATE_TEST_SUITE_P` to instantiate the test suite
with any set of parameters you want. googletest defines a number of functions
for generating test parameters. They return what we call (surprise!) *parameter
Gennadiy Civil's avatar
Gennadiy Civil committed
1391
1392
generators*. Here is a summary of them, which are all in the `testing`
namespace:
1393

1394

1395
1396
1397
1398
| Parameter Generator                                                                       | Behavior                                                                                                          |
| ----------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------- |
| `Range(begin, end [, step])`                                                              | Yields values `{begin, begin+step, begin+step+step, ...}`. The values do not include `end`. `step` defaults to 1. |
| `Values(v1, v2, ..., vN)`                                                                 | Yields values `{v1, v2, ..., vN}`.                                                                                |
1399
| `ValuesIn(container)` and  `ValuesIn(begin,end)`                                          | Yields values from a C-style array, an  STL-style container, or an iterator range `[begin, end)`                  |
1400
1401
| `Bool()`                                                                                  | Yields sequence `{false, true}`.                                                                                  |
| `Combine(g1, g2, ..., gN)`                                                                | Yields all combinations (Cartesian product) as std\:\:tuples of the values generated by the `N` generators.       |
1402
1403


Gennadiy Civil's avatar
Gennadiy Civil committed
1404
For more details, see the comments at the definitions of these functions.
1405

1406
1407
The following statement will instantiate tests from the `FooTest` test suite
each with parameter values `"meeny"`, `"miny"`, and `"moe"`.
1408

Gennadiy Civil's avatar
Gennadiy Civil committed
1409
```c++
Abseil Team's avatar
Abseil Team committed
1410
INSTANTIATE_TEST_SUITE_P(MeenyMinyMoe,
1411
1412
                         FooTest,
                         testing::Values("meeny", "miny", "moe"));
1413
1414
```

Abseil Team's avatar
Abseil Team committed
1415
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
1416
1417
NOTE: The code above must be placed at global or namespace scope, not at
function scope.
1418

Abseil Team's avatar
Abseil Team committed
1419
1420
1421
The first argument to `INSTANTIATE_TEST_SUITE_P` is a unique name for the
instantiation of the test suite. The next argument is the name of the test
pattern, and the last is the parameter generator.
Abseil Team's avatar
Abseil Team committed
1422

Abseil Team's avatar
Abseil Team committed
1423
1424
1425
1426
You can instantiate a test pattern more than once, so to distinguish different
instances of the pattern, the instantiation name is added as a prefix to the
actual test suite name. Remember to pick unique prefixes for different
instantiations. The tests from the instantiation above will have these names:
Abseil Team's avatar
Abseil Team committed
1427

Abseil Team's avatar
Abseil Team committed
1428
1429
1430
1431
1432
1433
*   `MeenyMinyMoe/FooTest.DoesBlah/0` for `"meeny"`
*   `MeenyMinyMoe/FooTest.DoesBlah/1` for `"miny"`
*   `MeenyMinyMoe/FooTest.DoesBlah/2` for `"moe"`
*   `MeenyMinyMoe/FooTest.HasBlahBlah/0` for `"meeny"`
*   `MeenyMinyMoe/FooTest.HasBlahBlah/1` for `"miny"`
*   `MeenyMinyMoe/FooTest.HasBlahBlah/2` for `"moe"`
1434

1435
You can use these names in [`--gtest_filter`](#running-a-subset-of-the-tests).
Gennadiy Civil's avatar
Gennadiy Civil committed
1436

Abseil Team's avatar
Abseil Team committed
1437
1438
The following statement will instantiate all tests from `FooTest` again, each
with parameter values `"cat"` and `"dog"`:
Gennadiy Civil's avatar
Gennadiy Civil committed
1439
1440

```c++
1441
const char* pets[] = {"cat", "dog"};
Abseil Team's avatar
Abseil Team committed
1442
INSTANTIATE_TEST_SUITE_P(Pets, FooTest, testing::ValuesIn(pets));
1443
1444
1445
1446
```

The tests from the instantiation above will have these names:

Abseil Team's avatar
Abseil Team committed
1447
1448
1449
1450
*   `Pets/FooTest.DoesBlah/0` for `"cat"`
*   `Pets/FooTest.DoesBlah/1` for `"dog"`
*   `Pets/FooTest.HasBlahBlah/0` for `"cat"`
*   `Pets/FooTest.HasBlahBlah/1` for `"dog"`
1451

1452
1453
1454
Please note that `INSTANTIATE_TEST_SUITE_P` will instantiate *all* tests in the
given test suite, whether their definitions come before or *after* the
`INSTANTIATE_TEST_SUITE_P` statement.
1455

Abseil Team's avatar
Abseil Team committed
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
Additionally, by default, every `TEST_P` without a corresponding
`INSTANTIATE_TEST_SUITE_P` causes a failing test in test suite
`GoogleTestVerification`. If you have a test suite where that omission is not an
error, for example it is in a library that may be linked in for other reasons or
where the list of test cases is dynamic and may be empty, then this check can be
suppressed by tagging the test suite:

```c++
GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST(FooTest);
```

Abseil Team's avatar
Abseil Team committed
1467
1468
You can see [sample7_unittest.cc] and [sample8_unittest.cc] for more examples.

Abseil Team's avatar
Abseil Team committed
1469
1470
[sample7_unittest.cc]: https://github.com/google/googletest/blob/master/googletest/samples/sample7_unittest.cc "Parameterized Test example"
[sample8_unittest.cc]: https://github.com/google/googletest/blob/master/googletest/samples/sample8_unittest.cc "Parameterized Test example with multiple parameters"
1471

Gennadiy Civil's avatar
Gennadiy Civil committed
1472
### Creating Value-Parameterized Abstract Tests
1473

Gennadiy Civil's avatar
Gennadiy Civil committed
1474
1475
1476
1477
1478
1479
1480
1481
In the above, we define and instantiate `FooTest` in the *same* source file.
Sometimes you may want to define value-parameterized tests in a library and let
other people instantiate them later. This pattern is known as *abstract tests*.
As an example of its application, when you are designing an interface you can
write a standard suite of abstract tests (perhaps using a factory function as
the test parameter) that all implementations of the interface are expected to
pass. When someone implements the interface, they can instantiate your suite to
get all the interface-conformance tests for free.
1482
1483
1484

To define abstract tests, you should organize your code like this:

Gennadiy Civil's avatar
Gennadiy Civil committed
1485
1486
1487
1.  Put the definition of the parameterized test fixture class (e.g. `FooTest`)
    in a header file, say `foo_param_test.h`. Think of this as *declaring* your
    abstract tests.
1488
2.  Put the `TEST_P` definitions in `foo_param_test.cc`, which includes
Gennadiy Civil's avatar
Gennadiy Civil committed
1489
1490
1491
    `foo_param_test.h`. Think of this as *implementing* your abstract tests.

Once they are defined, you can instantiate them by including `foo_param_test.h`,
1492
1493
invoking `INSTANTIATE_TEST_SUITE_P()`, and depending on the library target that
contains `foo_param_test.cc`. You can instantiate the same abstract test suite
Gennadiy Civil's avatar
Gennadiy Civil committed
1494
multiple times, possibly in different source files.
1495

Gennadiy Civil's avatar
Gennadiy Civil committed
1496
### Specifying Names for Value-Parameterized Test Parameters
1497

1498
The optional last argument to `INSTANTIATE_TEST_SUITE_P()` allows the user to
Gennadiy Civil's avatar
Gennadiy Civil committed
1499
1500
1501
specify a function or functor that generates custom test name suffixes based on
the test parameters. The function should accept one argument of type
`testing::TestParamInfo<class ParamType>`, and return `std::string`.
1502

Gennadiy Civil's avatar
Gennadiy Civil committed
1503
1504
1505
`testing::PrintToStringParamName` is a builtin test suffix generator that
returns the value of `testing::PrintToString(GetParam())`. It does not work for
`std::string` or C strings.
1506

Abseil Team's avatar
Abseil Team committed
1507
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
1508
NOTE: test names must be non-empty, unique, and may only contain ASCII
1509
alphanumeric characters. In particular, they
Abseil Team's avatar
Abseil Team committed
1510
[should not contain underscores](faq.md#why-should-test-suite-names-and-test-names-not-contain-underscore)
1511

Gennadiy Civil's avatar
Gennadiy Civil committed
1512
```c++
1513
class MyTestSuite : public testing::TestWithParam<int> {};
1514

1515
TEST_P(MyTestSuite, MyTest)
Gennadiy Civil's avatar
Gennadiy Civil committed
1516
1517
1518
{
  std::cout << "Example Test Param: " << GetParam() << std::endl;
}
1519

1520
1521
INSTANTIATE_TEST_SUITE_P(MyGroup, MyTestSuite, testing::Range(0, 10),
                         testing::PrintToStringParamName());
1522
```
Gennadiy Civil's avatar
Gennadiy Civil committed
1523

1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
Providing a custom functor allows for more control over test parameter name
generation, especially for types where the automatic conversion does not
generate helpful parameter names (e.g. strings as demonstrated above). The
following example illustrates this for multiple parameters, an enumeration type
and a string, and also demonstrates how to combine generators. It uses a lambda
for conciseness:

```c++
enum class MyType { MY_FOO = 0, MY_BAR = 1 };

Abseil Team's avatar
Abseil Team committed
1534
class MyTestSuite : public testing::TestWithParam<std::tuple<MyType, std::string>> {
1535
1536
1537
1538
1539
};

INSTANTIATE_TEST_SUITE_P(
    MyGroup, MyTestSuite,
    testing::Combine(
Abseil Team's avatar
Abseil Team committed
1540
        testing::Values(MyType::MY_FOO, MyType::MY_BAR),
Abseil Team's avatar
Abseil Team committed
1541
        testing::Values("A", "B")),
1542
    [](const testing::TestParamInfo<MyTestSuite::ParamType>& info) {
Abseil Team's avatar
Abseil Team committed
1543
      std::string name = absl::StrCat(
Abseil Team's avatar
Abseil Team committed
1544
          std::get<0>(info.param) == MyType::MY_FOO ? "Foo" : "Bar",
1545
          std::get<1>(info.param));
Abseil Team's avatar
Abseil Team committed
1546
      absl::c_replace_if(name, [](char c) { return !std::isalnum(c); }, '');
1547
1548
1549
1550
1551
      return name;
    });
```

## Typed Tests
Gennadiy Civil's avatar
Gennadiy Civil committed
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571

Suppose you have multiple implementations of the same interface and want to make
sure that all of them satisfy some common requirements. Or, you may have defined
several types that are supposed to conform to the same "concept" and you want to
verify it. In both cases, you want the same test logic repeated for different
types.

While you can write one `TEST` or `TEST_F` for each type you want to test (and
you may even factor the test logic into a function template that you invoke from
the `TEST`), it's tedious and doesn't scale: if you want `m` tests over `n`
types, you'll end up writing `m*n` `TEST`s.

*Typed tests* allow you to repeat the same test logic over a list of types. You
only need to write the test logic once, although you must know the type list
when writing typed tests. Here's how you do it:

First, define a fixture class template. It should be parameterized by a type.
Remember to derive it from `::testing::Test`:

```c++
1572
template <typename T>
Abseil Team's avatar
Abseil Team committed
1573
class FooTest : public testing::Test {
1574
1575
 public:
  ...
Abseil Team's avatar
Abseil Team committed
1576
  using List = std::list<T>;
1577
1578
1579
1580
1581
  static T shared_;
  T value_;
};
```

1582
Next, associate a list of types with the test suite, which will be repeated for
Gennadiy Civil's avatar
Gennadiy Civil committed
1583
each type in the list:
1584

Gennadiy Civil's avatar
Gennadiy Civil committed
1585
1586
```c++
using MyTypes = ::testing::Types<char, int, unsigned int>;
1587
TYPED_TEST_SUITE(FooTest, MyTypes);
1588
1589
```

1590
The type alias (`using` or `typedef`) is necessary for the `TYPED_TEST_SUITE`
Gennadiy Civil's avatar
Gennadiy Civil committed
1591
1592
macro to parse correctly. Otherwise the compiler will think that each comma in
the type list introduces a new macro argument.
1593

Gennadiy Civil's avatar
Gennadiy Civil committed
1594
Then, use `TYPED_TEST()` instead of `TEST_F()` to define a typed test for this
1595
test suite. You can repeat this as many times as you want:
1596

Gennadiy Civil's avatar
Gennadiy Civil committed
1597
```c++
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
TYPED_TEST(FooTest, DoesBlah) {
  // Inside a test, refer to the special name TypeParam to get the type
  // parameter.  Since we are inside a derived class template, C++ requires
  // us to visit the members of FooTest via 'this'.
  TypeParam n = this->value_;

  // To visit static members of the fixture, add the 'TestFixture::'
  // prefix.
  n += TestFixture::shared_;

  // To refer to typedefs in the fixture, add the 'typename TestFixture::'
  // prefix.  The 'typename' is required to satisfy the compiler.
  typename TestFixture::List values;
Gennadiy Civil's avatar
Gennadiy Civil committed
1611

1612
1613
1614
1615
1616
1617
1618
  values.push_back(n);
  ...
}

TYPED_TEST(FooTest, HasPropertyA) { ... }
```

Abseil Team's avatar
Abseil Team committed
1619
1620
You can see [sample6_unittest.cc] for a complete example.

Abseil Team's avatar
Abseil Team committed
1621
[sample6_unittest.cc]: https://github.com/google/googletest/blob/master/googletest/samples/sample6_unittest.cc "Typed Test example"
1622

Gennadiy Civil's avatar
Gennadiy Civil committed
1623
## Type-Parameterized Tests
1624

Gennadiy Civil's avatar
Gennadiy Civil committed
1625
1626
1627
1628
*Type-parameterized tests* are like typed tests, except that they don't require
you to know the list of types ahead of time. Instead, you can define the test
logic first and instantiate it with different type lists later. You can even
instantiate it more than once in the same program.
1629

Gennadiy Civil's avatar
Gennadiy Civil committed
1630
1631
1632
1633
1634
1635
If you are designing an interface or concept, you can define a suite of
type-parameterized tests to verify properties that any valid implementation of
the interface/concept should have. Then, the author of each implementation can
just instantiate the test suite with their type to verify that it conforms to
the requirements, without having to write similar tests repeatedly. Here's an
example:
1636
1637
1638

First, define a fixture class template, as we did with typed tests:

Gennadiy Civil's avatar
Gennadiy Civil committed
1639
```c++
1640
template <typename T>
Abseil Team's avatar
Abseil Team committed
1641
class FooTest : public testing::Test {
1642
1643
1644
1645
  ...
};
```

1646
Next, declare that you will define a type-parameterized test suite:
1647

Gennadiy Civil's avatar
Gennadiy Civil committed
1648
```c++
1649
TYPED_TEST_SUITE_P(FooTest);
1650
1651
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1652
1653
Then, use `TYPED_TEST_P()` to define a type-parameterized test. You can repeat
this as many times as you want:
1654

Gennadiy Civil's avatar
Gennadiy Civil committed
1655
```c++
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
TYPED_TEST_P(FooTest, DoesBlah) {
  // Inside a test, refer to TypeParam to get the type parameter.
  TypeParam n = 0;
  ...
}

TYPED_TEST_P(FooTest, HasPropertyA) { ... }
```

Now the tricky part: you need to register all test patterns using the
1666
`REGISTER_TYPED_TEST_SUITE_P` macro before you can instantiate them. The first
1667
1668
argument of the macro is the test suite name; the rest are the names of the
tests in this test suite:
1669

Gennadiy Civil's avatar
Gennadiy Civil committed
1670
```c++
1671
REGISTER_TYPED_TEST_SUITE_P(FooTest,
1672
                            DoesBlah, HasPropertyA);
1673
1674
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1675
1676
1677
Finally, you are free to instantiate the pattern with the types you want. If you
put the above code in a header file, you can `#include` it in multiple C++
source files and instantiate it multiple times.
1678

Gennadiy Civil's avatar
Gennadiy Civil committed
1679
```c++
Abseil Team's avatar
Abseil Team committed
1680
using MyTypes = ::testing::Types<char, int, unsigned int>;
1681
INSTANTIATE_TYPED_TEST_SUITE_P(My, FooTest, MyTypes);
1682
1683
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1684
To distinguish different instances of the pattern, the first argument to the
1685
`INSTANTIATE_TYPED_TEST_SUITE_P` macro is a prefix that will be added to the
1686
1687
actual test suite name. Remember to pick unique prefixes for different
instances.
1688

Gennadiy Civil's avatar
Gennadiy Civil committed
1689
1690
In the special case where the type list contains only one type, you can write
that type directly without `::testing::Types<...>`, like this:
1691

Gennadiy Civil's avatar
Gennadiy Civil committed
1692
```c++
1693
INSTANTIATE_TYPED_TEST_SUITE_P(My, FooTest, int);
1694
1695
```

Abseil Team's avatar
Abseil Team committed
1696
You can see [sample6_unittest.cc] for a complete example.
1697

Gennadiy Civil's avatar
Gennadiy Civil committed
1698
## Testing Private Code
1699
1700

If you change your software's internal implementation, your tests should not
Gennadiy Civil's avatar
Gennadiy Civil committed
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
break as long as the change is not observable by users. Therefore, **per the
black-box testing principle, most of the time you should test your code through
its public interfaces.**

**If you still find yourself needing to test internal implementation code,
consider if there's a better design.** The desire to test internal
implementation is often a sign that the class is doing too much. Consider
extracting an implementation class, and testing it. Then use that implementation
class in the original class.

If you absolutely have to test non-public interface code though, you can. There
are two cases to consider:

*   Static functions ( *not* the same as static member functions!) or unnamed
    namespaces, and
*   Private or protected class members

To test them, we use the following special techniques:

*   Both static functions and definitions/declarations in an unnamed namespace
    are only visible within the same translation unit. To test them, you can
    `#include` the entire `.cc` file being tested in your `*_test.cc` file.
1723
    (#including `.cc` files is not a good way to reuse code - you should not do
Gennadiy Civil's avatar
Gennadiy Civil committed
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
    this in production code!)

    However, a better approach is to move the private code into the
    `foo::internal` namespace, where `foo` is the namespace your project
    normally uses, and put the private declarations in a `*-internal.h` file.
    Your production `.cc` files and your tests are allowed to include this
    internal header, but your clients are not. This way, you can fully test your
    internal implementation without leaking it to your clients.

*   Private class members are only accessible from within the class or by
    friends. To access a class' private members, you can declare your test
    fixture as a friend to the class and define accessors in your fixture. Tests
    using the fixture can then access the private members of your production
    class via the accessors in the fixture. Note that even though your fixture
    is a friend to your production class, your tests are not automatically
    friends to it, as they are technically defined in sub-classes of the
    fixture.

    Another way to test private members is to refactor them into an
    implementation class, which is then declared in a `*-internal.h` file. Your
    clients aren't allowed to include this header but your tests can. Such is
    called the
    [Pimpl](https://www.gamedev.net/articles/programming/general-and-gameplay-programming/the-c-pimpl-r1794/)
    (Private Implementation) idiom.

    Or, you can declare an individual test as a friend of your class by adding
    this line in the class body:

    ```c++
1753
        FRIEND_TEST(TestSuiteName, TestName);
Gennadiy Civil's avatar
Gennadiy Civil committed
1754
1755
1756
1757
1758
1759
1760
1761
    ```

    For example,

    ```c++
    // foo.h
    class Foo {
      ...
1762
     private:
Gennadiy Civil's avatar
Gennadiy Civil committed
1763
      FRIEND_TEST(FooTest, BarReturnsZeroOnNull);
1764

Gennadiy Civil's avatar
Gennadiy Civil committed
1765
1766
      int Bar(void* x);
    };
1767

Gennadiy Civil's avatar
Gennadiy Civil committed
1768
1769
1770
1771
    // foo_test.cc
    ...
    TEST(FooTest, BarReturnsZeroOnNull) {
      Foo foo;
1772
      EXPECT_EQ(foo.Bar(NULL), 0);  // Uses Foo's private member Bar().
Gennadiy Civil's avatar
Gennadiy Civil committed
1773
1774
    }
    ```
1775

Abseil Team's avatar
Abseil Team committed
1776
1777
1778
1779
1780
    Pay special attention when your class is defined in a namespace. If you want
    your test fixtures and tests to be friends of your class, then they must be
    defined in the exact same namespace (no anonymous or inline namespaces).

    For example, if the code to be tested looks like:
1781

Gennadiy Civil's avatar
Gennadiy Civil committed
1782
1783
    ```c++
    namespace my_namespace {
1784

Gennadiy Civil's avatar
Gennadiy Civil committed
1785
1786
1787
1788
1789
1790
    class Foo {
      friend class FooTest;
      FRIEND_TEST(FooTest, Bar);
      FRIEND_TEST(FooTest, Baz);
      ... definition of the class Foo ...
    };
1791

Gennadiy Civil's avatar
Gennadiy Civil committed
1792
1793
    }  // namespace my_namespace
    ```
1794

Gennadiy Civil's avatar
Gennadiy Civil committed
1795
    Your test code should be something like:
1796

Gennadiy Civil's avatar
Gennadiy Civil committed
1797
1798
    ```c++
    namespace my_namespace {
1799

Abseil Team's avatar
Abseil Team committed
1800
    class FooTest : public testing::Test {
Gennadiy Civil's avatar
Gennadiy Civil committed
1801
1802
1803
     protected:
      ...
    };
1804

Gennadiy Civil's avatar
Gennadiy Civil committed
1805
1806
    TEST_F(FooTest, Bar) { ... }
    TEST_F(FooTest, Baz) { ... }
1807

Gennadiy Civil's avatar
Gennadiy Civil committed
1808
1809
    }  // namespace my_namespace
    ```
1810

Yi Zheng's avatar
Yi Zheng committed
1811
## "Catching" Failures
1812

Gennadiy Civil's avatar
Gennadiy Civil committed
1813
1814
If you are building a testing utility on top of googletest, you'll want to test
your utility. What framework would you use to test it? googletest, of course.
1815

Gennadiy Civil's avatar
Gennadiy Civil committed
1816
1817
1818
1819
The challenge is to verify that your testing utility reports failures correctly.
In frameworks that report a failure by throwing an exception, you could catch
the exception and assert on it. But googletest doesn't use exceptions, so how do
we test that a piece of code generates an expected failure?
1820

Abseil Team's avatar
Abseil Team committed
1821
`"gtest/gtest-spi.h"` contains some constructs to do this. After #including this header,
Gennadiy Civil's avatar
Gennadiy Civil committed
1822
you can use
1823

Gennadiy Civil's avatar
Gennadiy Civil committed
1824
1825
```c++
  EXPECT_FATAL_FAILURE(statement, substring);
1826
1827
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1828
1829
to assert that `statement` generates a fatal (e.g. `ASSERT_*`) failure in the
current thread whose message contains the given `substring`, or use
1830

Gennadiy Civil's avatar
Gennadiy Civil committed
1831
1832
```c++
  EXPECT_NONFATAL_FAILURE(statement, substring);
1833
1834
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1835
if you are expecting a non-fatal (e.g. `EXPECT_*`) failure.
1836

Gennadiy Civil's avatar
Gennadiy Civil committed
1837
1838
1839
1840
Only failures in the current thread are checked to determine the result of this
type of expectations. If `statement` creates new threads, failures in these
threads are also ignored. If you want to catch failures in other threads as
well, use one of the following macros instead:
1841

Gennadiy Civil's avatar
Gennadiy Civil committed
1842
1843
1844
1845
```c++
  EXPECT_FATAL_FAILURE_ON_ALL_THREADS(statement, substring);
  EXPECT_NONFATAL_FAILURE_ON_ALL_THREADS(statement, substring);
```
1846

Abseil Team's avatar
Abseil Team committed
1847
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
1848
NOTE: Assertions from multiple threads are currently not supported on Windows.
1849
1850
1851

For technical reasons, there are some caveats:

Gennadiy Civil's avatar
Gennadiy Civil committed
1852
1853
1.  You cannot stream a failure message to either macro.

1854
2.  `statement` in `EXPECT_FATAL_FAILURE{_ON_ALL_THREADS}()` cannot reference
Gennadiy Civil's avatar
Gennadiy Civil committed
1855
    local non-static variables or non-static members of `this` object.
1856

1857
3.  `statement` in `EXPECT_FATAL_FAILURE{_ON_ALL_THREADS}()` cannot return a
Gennadiy Civil's avatar
Gennadiy Civil committed
1858
    value.
1859

1860
1861
## Registering tests programmatically

1862
The `TEST` macros handle the vast majority of all use cases, but there are few
Abseil Team's avatar
Abseil Team committed
1863
where runtime registration logic is required. For those cases, the framework
1864
1865
provides the `::testing::RegisterTest` that allows callers to register arbitrary
tests dynamically.
1866
1867

This is an advanced API only to be used when the `TEST` macros are insufficient.
1868
1869
The macros should be preferred when possible, as they avoid most of the
complexity of calling this function.
1870
1871
1872
1873

It provides the following signature:

```c++
1874
template <typename Factory>
1875
TestInfo* RegisterTest(const char* test_suite_name, const char* test_name,
1876
1877
1878
                       const char* type_param, const char* value_param,
                       const char* file, int line, Factory factory);
```
1879
1880

The `factory` argument is a factory callable (move-constructible) object or
1881
1882
1883
function pointer that creates a new instance of the Test object. It handles
ownership to the caller. The signature of the callable is `Fixture*()`, where
`Fixture` is the test fixture class for the test. All tests registered with the
1884
same `test_suite_name` must return the same fixture type. This is checked at
1885
runtime.
1886
1887
1888
1889
1890

The framework will infer the fixture class from the factory and will call the
`SetUpTestSuite` and `TearDownTestSuite` for it.

Must be called before `RUN_ALL_TESTS()` is invoked, otherwise behavior is
1891
undefined.
1892
1893
1894
1895

Use case example:

```c++
Abseil Team's avatar
Abseil Team committed
1896
class MyFixture : public testing::Test {
1897
1898
 public:
  // All of these optional, just like in regular macro usage.
1899
1900
  static void SetUpTestSuite() { ... }
  static void TearDownTestSuite() { ... }
1901
1902
1903
  void SetUp() override { ... }
  void TearDown() override { ... }
};
1904
1905

class MyTest : public MyFixture {
1906
1907
1908
 public:
  explicit MyTest(int data) : data_(data) {}
  void TestBody() override { ... }
1909
1910

 private:
1911
1912
  int data_;
};
1913
1914

void RegisterMyTests(const std::vector<int>& values) {
1915
  for (int v : values) {
Abseil Team's avatar
Abseil Team committed
1916
    testing::RegisterTest(
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
        "MyFixture", ("Test" + std::to_string(v)).c_str(), nullptr,
        std::to_string(v).c_str(),
        __FILE__, __LINE__,
        // Important to use the fixture type as the return type here.
        [=]() -> MyFixture* { return new MyTest(v); });
  }
}
...
int main(int argc, char** argv) {
  std::vector<int> values_to_test = LoadValuesFromConfig();
  RegisterMyTests(values_to_test);
  ...
  return RUN_ALL_TESTS();
}
Gennadiy Civil's avatar
Gennadiy Civil committed
1931
```
Gennadiy Civil's avatar
Gennadiy Civil committed
1932
## Getting the Current Test's Name
1933
1934
1935
1936
1937
1938

Sometimes a function may need to know the name of the currently running test.
For example, you may be using the `SetUp()` method of your test fixture to set
the golden file name based on which test is running. The `::testing::TestInfo`
class has this information:

Gennadiy Civil's avatar
Gennadiy Civil committed
1939
```c++
1940
1941
1942
1943
namespace testing {

class TestInfo {
 public:
1944
  // Returns the test suite name and the test name, respectively.
1945
1946
1947
  //
  // Do NOT delete or free the return value - it's managed by the
  // TestInfo class.
1948
  const char* test_suite_name() const;
1949
1950
1951
  const char* name() const;
};

Gennadiy Civil's avatar
Gennadiy Civil committed
1952
}
1953
1954
```

Gennadiy Civil's avatar
Gennadiy Civil committed
1955
To obtain a `TestInfo` object for the currently running test, call
1956
1957
`current_test_info()` on the `UnitTest` singleton object:

Gennadiy Civil's avatar
Gennadiy Civil committed
1958
1959
1960
```c++
  // Gets information about the currently running test.
  // Do NOT delete the returned object - it's managed by the UnitTest class.
Abseil Team's avatar
Abseil Team committed
1961
1962
  const testing::TestInfo* const test_info =
      testing::UnitTest::GetInstance()->current_test_info();
Gennadiy Civil's avatar
Gennadiy Civil committed
1963

1964
  printf("We are in test %s of test suite %s.\n",
Gennadiy Civil's avatar
Gennadiy Civil committed
1965
         test_info->name(),
1966
         test_info->test_suite_name());
1967
1968
1969
```

`current_test_info()` returns a null pointer if no test is running. In
Abseil Team's avatar
Abseil Team committed
1970
1971
particular, you cannot find the test suite name in `SetUpTestSuite()`,
`TearDownTestSuite()` (where you know the test suite name implicitly), or
1972
1973
functions called from them.

Gennadiy Civil's avatar
Gennadiy Civil committed
1974
## Extending googletest by Handling Test Events
1975

Gennadiy Civil's avatar
Gennadiy Civil committed
1976
1977
googletest provides an **event listener API** to let you receive notifications
about the progress of a test program and test failures. The events you can
1978
listen to include the start and end of the test program, a test suite, or a test
Gennadiy Civil's avatar
Gennadiy Civil committed
1979
1980
1981
method, among others. You may use this API to augment or replace the standard
console output, replace the XML output, or provide a completely different form
of output, such as a GUI or a database. You can also use test events as
1982
1983
checkpoints to implement a resource leak checker, for example.

Gennadiy Civil's avatar
Gennadiy Civil committed
1984
### Defining Event Listeners
1985

Gennadiy Civil's avatar
Gennadiy Civil committed
1986
1987
1988
1989
1990
1991
To define a event listener, you subclass either testing::TestEventListener or
testing::EmptyTestEventListener The former is an (abstract) interface, where
*each pure virtual method can be overridden to handle a test event* (For
example, when a test starts, the `OnTestStart()` method will be called.). The
latter provides an empty implementation of all methods in the interface, such
that a subclass only needs to override the methods it cares about.
1992

Gennadiy Civil's avatar
Gennadiy Civil committed
1993
1994
When an event is fired, its context is passed to the handler function as an
argument. The following argument types are used:
1995

Gennadiy Civil's avatar
Gennadiy Civil committed
1996
*   UnitTest reflects the state of the entire test program,
1997
*   TestSuite has information about a test suite, which can contain one or more
Gennadiy Civil's avatar
Gennadiy Civil committed
1998
1999
2000
    tests,
*   TestInfo contains the state of a test, and
*   TestPartResult represents the result of a test assertion.
2001

Gennadiy Civil's avatar
Gennadiy Civil committed
2002
2003
2004
2005
2006
2007
An event handler function can examine the argument it receives to find out
interesting information about the event and the test program's state.

Here's an example:

```c++
Abseil Team's avatar
Abseil Team committed
2008
  class MinimalistPrinter : public testing::EmptyTestEventListener {
2009
    // Called before a test starts.
Abseil Team's avatar
Abseil Team committed
2010
    void OnTestStart(const testing::TestInfo& test_info) override {
2011
      printf("*** Test %s.%s starting.\n",
2012
             test_info.test_suite_name(), test_info.name());
2013
2014
    }

Gennadiy Civil's avatar
Gennadiy Civil committed
2015
    // Called after a failed assertion or a SUCCESS().
Abseil Team's avatar
Abseil Team committed
2016
    void OnTestPartResult(const testing::TestPartResult& test_part_result) override {
2017
2018
2019
2020
2021
2022
2023
2024
      printf("%s in %s:%d\n%s\n",
             test_part_result.failed() ? "*** Failure" : "Success",
             test_part_result.file_name(),
             test_part_result.line_number(),
             test_part_result.summary());
    }

    // Called after a test ends.
Abseil Team's avatar
Abseil Team committed
2025
    void OnTestEnd(const testing::TestInfo& test_info) override {
2026
      printf("*** Test %s.%s ending.\n",
2027
             test_info.test_suite_name(), test_info.name());
2028
2029
2030
2031
    }
  };
```

Gennadiy Civil's avatar
Gennadiy Civil committed
2032
### Using Event Listeners
2033

Gennadiy Civil's avatar
Gennadiy Civil committed
2034
2035
2036
2037
2038
2039
To use the event listener you have defined, add an instance of it to the
googletest event listener list (represented by class TestEventListeners - note
the "s" at the end of the name) in your `main()` function, before calling
`RUN_ALL_TESTS()`:

```c++
2040
int main(int argc, char** argv) {
Abseil Team's avatar
Abseil Team committed
2041
  testing::InitGoogleTest(&argc, argv);
2042
  // Gets hold of the event listener list.
Abseil Team's avatar
Abseil Team committed
2043
2044
  testing::TestEventListeners& listeners =
      testing::UnitTest::GetInstance()->listeners();
Gennadiy Civil's avatar
Gennadiy Civil committed
2045
  // Adds a listener to the end.  googletest takes the ownership.
2046
2047
2048
2049
2050
  listeners.Append(new MinimalistPrinter);
  return RUN_ALL_TESTS();
}
```

Gennadiy Civil's avatar
Gennadiy Civil committed
2051
2052
2053
2054
2055
2056
There's only one problem: the default test result printer is still in effect, so
its output will mingle with the output from your minimalist printer. To suppress
the default printer, just release it from the event listener list and delete it.
You can do so by adding one line:

```c++
2057
2058
2059
2060
2061
2062
  ...
  delete listeners.Release(listeners.default_result_printer());
  listeners.Append(new MinimalistPrinter);
  return RUN_ALL_TESTS();
```

Gennadiy Civil's avatar
Gennadiy Civil committed
2063
Now, sit back and enjoy a completely different output from your tests. For more
Abseil Team's avatar
Abseil Team committed
2064
2065
details, see [sample9_unittest.cc].

Abseil Team's avatar
Abseil Team committed
2066
[sample9_unittest.cc]: https://github.com/google/googletest/blob/master/googletest/samples/sample9_unittest.cc "Event listener example"
2067

Gennadiy Civil's avatar
Gennadiy Civil committed
2068
2069
2070
2071
2072
2073
2074
You may append more than one listener to the list. When an `On*Start()` or
`OnTestPartResult()` event is fired, the listeners will receive it in the order
they appear in the list (since new listeners are added to the end of the list,
the default text printer and the default XML generator will receive the event
first). An `On*End()` event will be received by the listeners in the *reverse*
order. This allows output by listeners added later to be framed by output from
listeners added earlier.
2075

Gennadiy Civil's avatar
Gennadiy Civil committed
2076
### Generating Failures in Listeners
2077

Gennadiy Civil's avatar
Gennadiy Civil committed
2078
2079
You may use failure-raising macros (`EXPECT_*()`, `ASSERT_*()`, `FAIL()`, etc)
when processing an event. There are some restrictions:
2080

Gennadiy Civil's avatar
Gennadiy Civil committed
2081
2082
1.  You cannot generate any failure in `OnTestPartResult()` (otherwise it will
    cause `OnTestPartResult()` to be called recursively).
2083
2.  A listener that handles `OnTestPartResult()` is not allowed to generate any
Gennadiy Civil's avatar
Gennadiy Civil committed
2084
    failure.
2085

Gennadiy Civil's avatar
Gennadiy Civil committed
2086
2087
2088
2089
When you add listeners to the listener list, you should put listeners that
handle `OnTestPartResult()` *before* listeners that can generate failures. This
ensures that failures generated by the latter are attributed to the right test
by the former.
2090

Abseil Team's avatar
Abseil Team committed
2091
2092
See [sample10_unittest.cc] for an example of a failure-raising listener.

Abseil Team's avatar
Abseil Team committed
2093
[sample10_unittest.cc]: https://github.com/google/googletest/blob/master/googletest/samples/sample10_unittest.cc "Failure-raising listener example"
2094

Gennadiy Civil's avatar
Gennadiy Civil committed
2095
## Running Test Programs: Advanced Options
2096

Gennadiy Civil's avatar
Gennadiy Civil committed
2097
2098
googletest test programs are ordinary executables. Once built, you can run them
directly and affect their behavior via the following environment variables
2099
2100
2101
and/or command line flags. For the flags to work, your programs must call
`::testing::InitGoogleTest()` before calling `RUN_ALL_TESTS()`.

Gennadiy Civil's avatar
Gennadiy Civil committed
2102
2103
To see a list of supported flags and their usage, please run your test program
with the `--help` flag. You can also use `-h`, `-?`, or `/?` for short.
2104

Gennadiy Civil's avatar
Gennadiy Civil committed
2105
2106
If an option is specified both by an environment variable and by a flag, the
latter takes precedence.
2107

Gennadiy Civil's avatar
Gennadiy Civil committed
2108
### Selecting Tests
2109

Gennadiy Civil's avatar
Gennadiy Civil committed
2110
#### Listing Test Names
2111
2112
2113
2114
2115

Sometimes it is necessary to list the available tests in a program before
running them so that a filter may be applied if needed. Including the flag
`--gtest_list_tests` overrides all other flags and lists tests in the following
format:
Gennadiy Civil's avatar
Gennadiy Civil committed
2116
2117

```none
2118
TestSuite1.
2119
2120
  TestName1
  TestName2
2121
TestSuite2.
2122
2123
2124
2125
2126
2127
  TestName
```

None of the tests listed are actually run if the flag is provided. There is no
corresponding environment variable for this flag.

Gennadiy Civil's avatar
Gennadiy Civil committed
2128
#### Running a Subset of the Tests
2129

Gennadiy Civil's avatar
Gennadiy Civil committed
2130
2131
2132
2133
By default, a googletest program runs all tests the user has defined. Sometimes,
you want to run only a subset of the tests (e.g. for debugging or quickly
verifying a change). If you set the `GTEST_FILTER` environment variable or the
`--gtest_filter` flag to a filter string, googletest will only run the tests
2134
whose full names (in the form of `TestSuiteName.TestName`) match the filter.
2135
2136

The format of a filter is a '`:`'-separated list of wildcard patterns (called
Gennadiy Civil's avatar
Gennadiy Civil committed
2137
2138
2139
the *positive patterns*) optionally followed by a '`-`' and another
'`:`'-separated pattern list (called the *negative patterns*). A test matches
the filter if and only if it matches any of the positive patterns but does not
2140
2141
2142
match any of the negative patterns.

A pattern may contain `'*'` (matches any string) or `'?'` (matches any single
2143
2144
character). For convenience, the filter `'*-NegativePatterns'` can be also
written as `'-NegativePatterns'`.
2145

Gennadiy Civil's avatar
Gennadiy Civil committed
2146
For example:
2147

Gennadiy Civil's avatar
Gennadiy Civil committed
2148
2149
2150
*   `./foo_test` Has no flag, and thus runs all its tests.
*   `./foo_test --gtest_filter=*` Also runs everything, due to the single
    match-everything `*` value.
2151
2152
*   `./foo_test --gtest_filter=FooTest.*` Runs everything in test suite
    `FooTest` .
Gennadiy Civil's avatar
Gennadiy Civil committed
2153
2154
2155
2156
*   `./foo_test --gtest_filter=*Null*:*Constructor*` Runs any test whose full
    name contains either `"Null"` or `"Constructor"` .
*   `./foo_test --gtest_filter=-*DeathTest.*` Runs all non-death tests.
*   `./foo_test --gtest_filter=FooTest.*-FooTest.Bar` Runs everything in test
2157
    suite `FooTest` except `FooTest.Bar`.
Gennadiy Civil's avatar
Gennadiy Civil committed
2158
*   `./foo_test --gtest_filter=FooTest.*:BarTest.*-FooTest.Bar:BarTest.Foo` Runs
2159
2160
2161
    everything in test suite `FooTest` except `FooTest.Bar` and everything in
    test suite `BarTest` except `BarTest.Foo`.

Abseil Team's avatar
Abseil Team committed
2162
2163
2164
2165
2166
2167
2168
2169
2170
#### Stop test execution upon first failure

By default, a googletest program runs all tests the user has defined. In some
cases (e.g. iterative test development & execution) it may be desirable stop
test execution upon first failure (trading improved latency for completeness).
If `GTEST_FAIL_FAST` environment variable or `--gtest_fail_fast` flag is set,
the test runner will stop execution as soon as the first test failure is
found.

Gennadiy Civil's avatar
Gennadiy Civil committed
2171
#### Temporarily Disabling Tests
2172
2173
2174
2175
2176
2177

If you have a broken test that you cannot fix right away, you can add the
`DISABLED_` prefix to its name. This will exclude it from execution. This is
better than commenting out the code or using `#if 0`, as disabled tests are
still compiled (and thus won't rot).

2178
If you need to disable all tests in a test suite, you can either add `DISABLED_`
2179
to the front of the name of each test, or alternatively add it to the front of
2180
the test suite name.
2181

Gennadiy Civil's avatar
Gennadiy Civil committed
2182
For example, the following tests won't be run by googletest, even though they
2183
2184
will still be compiled:

Gennadiy Civil's avatar
Gennadiy Civil committed
2185
```c++
2186
2187
2188
// Tests that Foo does Abc.
TEST(FooTest, DISABLED_DoesAbc) { ... }

Abseil Team's avatar
Abseil Team committed
2189
class DISABLED_BarTest : public testing::Test { ... };
2190
2191
2192
2193
2194

// Tests that Bar does Xyz.
TEST_F(DISABLED_BarTest, DoesXyz) { ... }
```

Abseil Team's avatar
Abseil Team committed
2195
{: .callout .note}
Gennadiy Civil's avatar
Gennadiy Civil committed
2196
2197
2198
NOTE: This feature should only be used for temporary pain-relief. You still have
to fix the disabled tests at a later date. As a reminder, googletest will print
a banner warning you if a test program contains any disabled tests.
2199

Abseil Team's avatar
Abseil Team committed
2200
{: .callout .tip}
Abseil Team's avatar
Abseil Team committed
2201
2202
2203
TIP: You can easily count the number of disabled tests you have using
`grep`. This number can be used as a metric for
improving your test quality.
2204

Gennadiy Civil's avatar
Gennadiy Civil committed
2205
#### Temporarily Enabling Disabled Tests
2206

Gennadiy Civil's avatar
Gennadiy Civil committed
2207
2208
2209
2210
2211
To include disabled tests in test execution, just invoke the test program with
the `--gtest_also_run_disabled_tests` flag or set the
`GTEST_ALSO_RUN_DISABLED_TESTS` environment variable to a value other than `0`.
You can combine this with the `--gtest_filter` flag to further select which
disabled tests to run.
2212

Gennadiy Civil's avatar
Gennadiy Civil committed
2213
### Repeating the Tests
2214
2215
2216
2217
2218

Once in a while you'll run into a test whose result is hit-or-miss. Perhaps it
will fail only 1% of the time, making it rather hard to reproduce the bug under
a debugger. This can be a major source of frustration.

Gennadiy Civil's avatar
Gennadiy Civil committed
2219
2220
2221
The `--gtest_repeat` flag allows you to repeat all (or selected) test methods in
a program many times. Hopefully, a flaky test will eventually fail and give you
a chance to debug. Here's how to use it:
2222

Gennadiy Civil's avatar
Gennadiy Civil committed
2223
2224
2225
```none
$ foo_test --gtest_repeat=1000
Repeat foo_test 1000 times and don't stop at failures.
2226

Gennadiy Civil's avatar
Gennadiy Civil committed
2227
2228
$ foo_test --gtest_repeat=-1
A negative count means repeating forever.
2229

Gennadiy Civil's avatar
Gennadiy Civil committed
2230
2231
2232
2233
2234
$ foo_test --gtest_repeat=1000 --gtest_break_on_failure
Repeat foo_test 1000 times, stopping at the first failure.  This
is especially useful when running under a debugger: when the test
fails, it will drop into the debugger and you can then inspect
variables and stacks.
2235

Gennadiy Civil's avatar
Gennadiy Civil committed
2236
2237
2238
2239
$ foo_test --gtest_repeat=1000 --gtest_filter=FooBar.*
Repeat the tests whose name matches the filter 1000 times.
```

2240
2241
2242
2243
If your test program contains
[global set-up/tear-down](#global-set-up-and-tear-down) code, it will be
repeated in each iteration as well, as the flakiness may be in it. You can also
specify the repeat count by setting the `GTEST_REPEAT` environment variable.
Gennadiy Civil's avatar
Gennadiy Civil committed
2244
2245

### Shuffling the Tests
2246
2247

You can specify the `--gtest_shuffle` flag (or set the `GTEST_SHUFFLE`
Gennadiy Civil's avatar
Gennadiy Civil committed
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
environment variable to `1`) to run the tests in a program in a random order.
This helps to reveal bad dependencies between tests.

By default, googletest uses a random seed calculated from the current time.
Therefore you'll get a different order every time. The console output includes
the random seed value, such that you can reproduce an order-related test failure
later. To specify the random seed explicitly, use the `--gtest_random_seed=SEED`
flag (or set the `GTEST_RANDOM_SEED` environment variable), where `SEED` is an
integer in the range [0, 99999]. The seed value 0 is special: it tells
googletest to do the default behavior of calculating the seed from the current
2258
2259
time.

Gennadiy Civil's avatar
Gennadiy Civil committed
2260
2261
2262
2263
If you combine this with `--gtest_repeat=N`, googletest will pick a different
random seed and re-shuffle the tests in each iteration.

### Controlling Test Output
2264

Gennadiy Civil's avatar
Gennadiy Civil committed
2265
#### Colored Terminal Output
2266

Gennadiy Civil's avatar
Gennadiy Civil committed
2267
2268
googletest can use colors in its terminal output to make it easier to spot the
important information:
2269

Abseil Team's avatar
Abseil Team committed
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
<pre>...
<font color="green">[----------]</font> 1 test from FooTest
<font color="green">[ RUN      ]</font> FooTest.DoesAbc
<font color="green">[       OK ]</font> FooTest.DoesAbc
<font color="green">[----------]</font> 2 tests from BarTest
<font color="green">[ RUN      ]</font> BarTest.HasXyzProperty
<font color="green">[       OK ]</font> BarTest.HasXyzProperty
<font color="green">[ RUN      ]</font> BarTest.ReturnsTrueOnSuccess
... some error messages ...
<font color="red">[   FAILED ]</font> BarTest.ReturnsTrueOnSuccess
...
<font color="green">[==========]</font> 30 tests from 14 test suites ran.
<font color="green">[   PASSED ]</font> 28 tests.
<font color="red">[   FAILED ]</font> 2 tests, listed below:
<font color="red">[   FAILED ]</font> BarTest.ReturnsTrueOnSuccess
<font color="red">[   FAILED ]</font> AnotherTest.DoesXyz

 2 FAILED TESTS
</pre>
2289

Gennadiy Civil's avatar
Gennadiy Civil committed
2290
You can set the `GTEST_COLOR` environment variable or the `--gtest_color`
2291
command line flag to `yes`, `no`, or `auto` (the default) to enable colors,
Gennadiy Civil's avatar
Gennadiy Civil committed
2292
2293
2294
disable colors, or let googletest decide. When the value is `auto`, googletest
will use colors if and only if the output goes to a terminal and (on non-Windows
platforms) the `TERM` environment variable is set to `xterm` or `xterm-color`.
2295

Calum Robinson's avatar
Calum Robinson committed
2296
2297
2298
2299
2300
2301
#### Suppressing test passes

By default, googletest prints 1 line of output for each test, indicating if it
passed or failed. To show only test failures, run the test program with
`--gtest_brief=1`, or set the GTEST_BRIEF environment variable to `1`.

Gennadiy Civil's avatar
Gennadiy Civil committed
2302
#### Suppressing the Elapsed Time
2303

Gennadiy Civil's avatar
Gennadiy Civil committed
2304
2305
2306
By default, googletest prints the time it takes to run each test. To disable
that, run the test program with the `--gtest_print_time=0` command line flag, or
set the GTEST_PRINT_TIME environment variable to `0`.
2307

Gennadiy Civil's avatar
Gennadiy Civil committed
2308
2309
#### Suppressing UTF-8 Text Output

Gennadiy Civil's avatar
Gennadiy Civil committed
2310
2311
2312
2313
2314
In case of assertion failures, googletest prints expected and actual values of
type `string` both as hex-encoded strings as well as in readable UTF-8 text if
they contain valid non-ASCII UTF-8 characters. If you want to suppress the UTF-8
text because, for example, you don't have an UTF-8 compatible output medium, run
the test program with `--gtest_print_utf8=0` or set the `GTEST_PRINT_UTF8`
Gennadiy Civil's avatar
Gennadiy Civil committed
2315
2316
environment variable to `0`.

Gennadiy Civil's avatar
Gennadiy Civil committed
2317

2318

Gennadiy Civil's avatar
Gennadiy Civil committed
2319
2320
2321
2322
#### Generating an XML Report

googletest can emit a detailed XML report to a file in addition to its normal
textual output. The report contains the duration of each test, and thus can help
Abseil Team's avatar
Abseil Team committed
2323
you identify slow tests.
2324
2325

To generate the XML report, set the `GTEST_OUTPUT` environment variable or the
Gennadiy Civil's avatar
Gennadiy Civil committed
2326
2327
2328
2329
`--gtest_output` flag to the string `"xml:path_to_output_file"`, which will
create the file at the given location. You can also just use the string `"xml"`,
in which case the output can be found in the `test_detail.xml` file in the
current directory.
2330
2331

If you specify a directory (for example, `"xml:output/directory/"` on Linux or
Gennadiy Civil's avatar
Gennadiy Civil committed
2332
`"xml:output\directory\"` on Windows), googletest will create the XML file in
2333
2334
that directory, named after the test executable (e.g. `foo_test.xml` for test
program `foo_test` or `foo_test.exe`). If the file already exists (perhaps left
Gennadiy Civil's avatar
Gennadiy Civil committed
2335
over from a previous run), googletest will pick a different name (e.g.
2336
2337
`foo_test_1.xml`) to avoid overwriting it.

Gennadiy Civil's avatar
Gennadiy Civil committed
2338
2339
2340
2341
2342
The report is based on the `junitreport` Ant task. Since that format was
originally intended for Java, a little interpretation is required to make it
apply to googletest tests, as shown here:

```xml
2343
<testsuites name="AllTests" ...>
2344
  <testsuite name="test_case_name" ...>
Gennadiy Civil's avatar
Gennadiy Civil committed
2345
    <testcase    name="test_name" ...>
2346
2347
2348
2349
2350
2351
2352
2353
      <failure message="..."/>
      <failure message="..."/>
      <failure message="..."/>
    </testcase>
  </testsuite>
</testsuites>
```

Gennadiy Civil's avatar
Gennadiy Civil committed
2354
*   The root `<testsuites>` element corresponds to the entire test program.
2355
*   `<testsuite>` elements correspond to googletest test suites.
Gennadiy Civil's avatar
Gennadiy Civil committed
2356
*   `<testcase>` elements correspond to googletest test functions.
2357
2358
2359

For instance, the following program

Gennadiy Civil's avatar
Gennadiy Civil committed
2360
```c++
2361
2362
2363
2364
2365
2366
2367
TEST(MathTest, Addition) { ... }
TEST(MathTest, Subtraction) { ... }
TEST(LogicTest, NonContradiction) { ... }
```

could generate this report:

Gennadiy Civil's avatar
Gennadiy Civil committed
2368
```xml
2369
<?xml version="1.0" encoding="UTF-8"?>
Gennadiy Civil's avatar
Gennadiy Civil committed
2370
2371
2372
2373
2374
<testsuites tests="3" failures="1" errors="0" time="0.035" timestamp="2011-10-31T18:52:42" name="AllTests">
  <testsuite name="MathTest" tests="2" failures="1" errors="0" time="0.015">
    <testcase name="Addition" status="run" time="0.007" classname="">
      <failure message="Value of: add(1, 1)&#x0A;  Actual: 3&#x0A;Expected: 2" type="">...</failure>
      <failure message="Value of: add(1, -1)&#x0A;  Actual: 1&#x0A;Expected: 0" type="">...</failure>
2375
    </testcase>
Gennadiy Civil's avatar
Gennadiy Civil committed
2376
    <testcase name="Subtraction" status="run" time="0.005" classname="">
2377
2378
    </testcase>
  </testsuite>
Gennadiy Civil's avatar
Gennadiy Civil committed
2379
2380
  <testsuite name="LogicTest" tests="1" failures="0" errors="0" time="0.005">
    <testcase name="NonContradiction" status="run" time="0.005" classname="">
2381
2382
2383
2384
2385
2386
2387
    </testcase>
  </testsuite>
</testsuites>
```

Things to note:

Gennadiy Civil's avatar
Gennadiy Civil committed
2388
*   The `tests` attribute of a `<testsuites>` or `<testsuite>` element tells how
2389
    many test functions the googletest program or test suite contains, while the
Gennadiy Civil's avatar
Gennadiy Civil committed
2390
2391
    `failures` attribute tells how many of them failed.

2392
*   The `time` attribute expresses the duration of the test, test suite, or
Gennadiy Civil's avatar
Gennadiy Civil committed
2393
2394
2395
2396
    entire test program in seconds.

*   The `timestamp` attribute records the local date and time of the test
    execution.
2397

Gennadiy Civil's avatar
Gennadiy Civil committed
2398
2399
*   Each `<failure>` element corresponds to a single failed googletest
    assertion.
2400

2401
#### Generating a JSON Report
2402

Gennadiy Civil's avatar
Gennadiy Civil committed
2403
2404
2405
googletest can also emit a JSON report as an alternative format to XML. To
generate the JSON report, set the `GTEST_OUTPUT` environment variable or the
`--gtest_output` flag to the string `"json:path_to_output_file"`, which will
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
create the file at the given location. You can also just use the string
`"json"`, in which case the output can be found in the `test_detail.json` file
in the current directory.

The report format conforms to the following JSON Schema:

```json
{
  "$schema": "http://json-schema.org/schema#",
  "type": "object",
  "definitions": {
2417
    "TestCase": {
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
      "type": "object",
      "properties": {
        "name": { "type": "string" },
        "tests": { "type": "integer" },
        "failures": { "type": "integer" },
        "disabled": { "type": "integer" },
        "time": { "type": "string" },
        "testsuite": {
          "type": "array",
          "items": {
            "$ref": "#/definitions/TestInfo"
          }
        }
      }
    },
    "TestInfo": {
      "type": "object",
      "properties": {
        "name": { "type": "string" },
        "status": {
          "type": "string",
          "enum": ["RUN", "NOTRUN"]
        },
        "time": { "type": "string" },
        "classname": { "type": "string" },
        "failures": {
          "type": "array",
          "items": {
            "$ref": "#/definitions/Failure"
          }
        }
      }
    },
    "Failure": {
      "type": "object",
      "properties": {
        "failures": { "type": "string" },
        "type": { "type": "string" }
      }
    }
  },
  "properties": {
    "tests": { "type": "integer" },
    "failures": { "type": "integer" },
    "disabled": { "type": "integer" },
    "errors": { "type": "integer" },
    "timestamp": {
      "type": "string",
      "format": "date-time"
    },
    "time": { "type": "string" },
    "name": { "type": "string" },
    "testsuites": {
      "type": "array",
      "items": {
2473
        "$ref": "#/definitions/TestCase"
2474
2475
2476
2477
2478
2479
      }
    }
  }
}
```

2480
2481
The report uses the format that conforms to the following Proto3 using the
[JSON encoding](https://developers.google.com/protocol-buffers/docs/proto3#json):
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498

```proto
syntax = "proto3";

package googletest;

import "google/protobuf/timestamp.proto";
import "google/protobuf/duration.proto";

message UnitTest {
  int32 tests = 1;
  int32 failures = 2;
  int32 disabled = 3;
  int32 errors = 4;
  google.protobuf.Timestamp timestamp = 5;
  google.protobuf.Duration time = 6;
  string name = 7;
2499
  repeated TestCase testsuites = 8;
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
}

message TestCase {
  string name = 1;
  int32 tests = 2;
  int32 failures = 3;
  int32 disabled = 4;
  int32 errors = 5;
  google.protobuf.Duration time = 6;
  repeated TestInfo testsuite = 7;
}

message TestInfo {
  string name = 1;
  enum Status {
    RUN = 0;
    NOTRUN = 1;
  }
  Status status = 2;
  google.protobuf.Duration time = 3;
  string classname = 4;
  message Failure {
    string failures = 1;
    string type = 2;
  }
  repeated Failure failures = 5;
}
```

For instance, the following program

```c++
TEST(MathTest, Addition) { ... }
TEST(MathTest, Subtraction) { ... }
TEST(LogicTest, NonContradiction) { ... }
```

could generate this report:

```json
{
  "tests": 3,
  "failures": 1,
  "errors": 0,
  "time": "0.035s",
2545
  "timestamp": "2011-10-31T18:52:42Z",
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
  "name": "AllTests",
  "testsuites": [
    {
      "name": "MathTest",
      "tests": 2,
      "failures": 1,
      "errors": 0,
      "time": "0.015s",
      "testsuite": [
        {
          "name": "Addition",
          "status": "RUN",
          "time": "0.007s",
          "classname": "",
          "failures": [
            {
2562
              "message": "Value of: add(1, 1)\n  Actual: 3\nExpected: 2",
2563
2564
2565
              "type": ""
            },
            {
2566
              "message": "Value of: add(1, -1)\n  Actual: 1\nExpected: 0",
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
              "type": ""
            }
          ]
        },
        {
          "name": "Subtraction",
          "status": "RUN",
          "time": "0.005s",
          "classname": ""
        }
      ]
2578
    },
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
    {
      "name": "LogicTest",
      "tests": 1,
      "failures": 0,
      "errors": 0,
      "time": "0.005s",
      "testsuite": [
        {
          "name": "NonContradiction",
          "status": "RUN",
          "time": "0.005s",
          "classname": ""
        }
      ]
    }
  ]
}
```

Abseil Team's avatar
Abseil Team committed
2598
{: .callout .important}
2599
2600
IMPORTANT: The exact format of the JSON document is subject to change.

Gennadiy Civil's avatar
Gennadiy Civil committed
2601
### Controlling How Failures Are Reported
2602

2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
#### Detecting Test Premature Exit

Google Test implements the _premature-exit-file_ protocol for test runners
to catch any kind of unexpected exits of test programs. Upon start,
Google Test creates the file which will be automatically deleted after
all work has been finished. Then, the test runner can check if this file
exists. In case the file remains undeleted, the inspected test has exited
prematurely.

This feature is enabled only if the `TEST_PREMATURE_EXIT_FILE` environment
variable has been set.

Gennadiy Civil's avatar
Gennadiy Civil committed
2615
#### Turning Assertion Failures into Break-Points
2616
2617
2618

When running test programs under a debugger, it's very convenient if the
debugger can catch an assertion failure and automatically drop into interactive
Gennadiy Civil's avatar
Gennadiy Civil committed
2619
mode. googletest's *break-on-failure* mode supports this behavior.
2620
2621

To enable it, set the `GTEST_BREAK_ON_FAILURE` environment variable to a value
2622
other than `0`. Alternatively, you can use the `--gtest_break_on_failure`
2623
2624
command line flag.

Gennadiy Civil's avatar
Gennadiy Civil committed
2625
#### Disabling Catching Test-Thrown Exceptions
2626

Gennadiy Civil's avatar
Gennadiy Civil committed
2627
2628
2629
2630
2631
2632
googletest can be used either with or without exceptions enabled. If a test
throws a C++ exception or (on Windows) a structured exception (SEH), by default
googletest catches it, reports it as a test failure, and continues with the next
test method. This maximizes the coverage of a test run. Also, on Windows an
uncaught exception will cause a pop-up window, so catching the exceptions allows
you to run the tests automatically.
2633

Gennadiy Civil's avatar
Gennadiy Civil committed
2634
2635
2636
2637
2638
When debugging the test failures, however, you may instead want the exceptions
to be handled by the debugger, such that you can examine the call stack when an
exception is thrown. To achieve that, set the `GTEST_CATCH_EXCEPTIONS`
environment variable to `0`, or use the `--gtest_catch_exceptions=0` flag when
running the tests.
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667

### Sanitizer Integration

The
[Undefined Behavior Sanitizer](https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html),
[Address Sanitizer](https://github.com/google/sanitizers/wiki/AddressSanitizer),
and
[Thread Sanitizer](https://github.com/google/sanitizers/wiki/ThreadSanitizerCppManual)
all provide weak functions that you can override to trigger explicit failures
when they detect sanitizer errors, such as creating a reference from `nullptr`.
To override these functions, place definitions for them in a source file that
you compile as part of your main binary:

```
extern "C" {
void __ubsan_on_report() {
  FAIL() << "Encountered an undefined behavior sanitizer error";
}
void __asan_on_error() {
  FAIL() << "Encountered an address sanitizer error";
}
void __tsan_on_report() {
  FAIL() << "Encountered a thread sanitizer error";
}
}  // extern "C"
```

After compiling your project with one of the sanitizers enabled, if a particular
test triggers a sanitizer error, googletest will report that it failed.