The [Premature Exit File](https://google.github.io/googletest/advanced.html#detecting-test-premature-exit)
can be used for more than just death tests. Runners can use it to detect any
kind of unexpected exit.
Presently, its creation is tied to whether a platform supports Death
Tests. This change removes that coupling (though if Death Tests are supported,
then the Premature Exit file is implicitly supported).
PiperOrigin-RevId: 853416224
Change-Id: I5354730b82bc8a20981120f4e1df59b58cb31f6e
Sharded tests interact awkwardly with --gtest_fail_if_no_test_selected, but we
can't speak clearly enough to the use cases to complicate the mental model, so
instead we attempt to clarify the simplest approach to debugging a single test
when sharding and --gtest_fail_if_no_test_selected are both in use: unset the
flag.
PiperOrigin-RevId: 829686145
Change-Id: I9ebbddc6e7537feefe2a3707fd323fc9132b99d1
This allows for simpler migration of function args to smart pointers without needing all changes to tests to be atomic.
PiperOrigin-RevId: 818635600
Change-Id: I9434685d9640f82b98d0b5261701b78c93d3ec1e
Replace an internal link to AbslStringify's documentation with the equivalent public link.
PiperOrigin-RevId: 779085254
Change-Id: I2cbd2b6262a5ced06f166d4fcc7c08796e5a60af
While ConditionalPrintAsText gets a char pointer and the string length, it used to only pass the pointer to operator<<. That does not work for strings that are not zero terminated, because operator<< has to resort to strlen, which happily overflows the string buffer.
This CL wraps the char pointer and the length in a string_view and passes that to operator<< to fix that issue.
PiperOrigin-RevId: 765137769
Change-Id: Ie97067ce9d5b23175a512945fee943a8b95a94ff
Clang has recently added "warnings when mixing different charN_t types" [1].
The rationale is that "charN_t represent code units of different UTF encodings.
Therefore the values of 2 different charN_t objects do not represent the same
characters."
Note that the warning here may be legitimate - from https://github.com/google/googletest/issues/4762:
"[...] This is incorrect for values that do not represent valid codepoints."
For the time being, silence the warning by being more explicit about the
conversion being intentional by using static_cast.
Link: https://github.com/llvm/llvm-project/pull/138708 [1]
PiperOrigin-RevId: 760644157
Change-Id: I2e6cc1871975455cecac8731b2f93fd5beeaf0e1
Previously, those matchers always invoked the child matchers with an IsInterested MatchResultListener, resulting in unnecessary work formatting match results that would be discarded.
PiperOrigin-RevId: 750704295
Change-Id: I1639a3a15d269459f26b3aebc3a6cbdced6896a3
Regexps seem to have nothing in common with death tests, yet their description is planted right in the middle of the death test section.
This CL pulls the regexp section one level up and just before death tests.
PiperOrigin-RevId: 721817710
Change-Id: Idc52f450fb10960a590ceb1a70339f86d4478fe4
The original URL started with https://www.opengroup.org and fails to redirect to the reference page. According to Chrome's dev tools (Network tab), the response to that request is a HTTP 302 Found, with the new location being "balancer://pubs/onlinepubs/009695399/basedefs/xbd_chap09.html". Chrome does not know what to do with that and displays a blank page.
Changing the `www` subdomain to `pubs` results in HTTP 200.
PiperOrigin-RevId: 721804738
Change-Id: I816d08d5baf957c63ac97506d04c7cc49c3a5ce7
With this change, arbitrary property names in the testsuite no longer
cause the produced XML output to be ill-formed.
PiperOrigin-RevId: 721549090
Change-Id: Iedffa89bf914478f563c8f3b82cd50557762a665
If you have a death test that uses mock objects, it's very likely that those mock objects are not properly cleaned up before death. The gMock atexit leak detector will call `_exit(1)`, overriding your expected exit code.
PiperOrigin-RevId: 713722657
Change-Id: I92a326f3923efc9f0e235d825616f3094eca995c
This surfaces useful information about the environment failure in a structured form.
As we can see from the updated test, previously unsurfaced information is now present.
PiperOrigin-RevId: 709892315
Change-Id: I2656294d50c33f995bef5c96195a66cff3c4b907
This allows users to override the fake SDK with a real one using https://bazel.build/rules/lib/globals/module#override_repo.
Without this change, it is impossible for a project that depends on googletest as a `bazel_dep` to build tests using the "real" Fuchsia SDK, because any references to `@fuchsia_sdk` within googletest `BUILD.bazel` files unconditionally resolve to the "fake" Fuchsia SDK. With this change, if you have the real Fuchsia SDK declared in your `MODULE.bazel`, you can add the following lines to coerce googletest to use the real Fuchsia SDK as well:
fake_fuchsia_sdk_extension =
use_extension("@com_google_googletest//:fake_fuchsia_sdk.bzl", "fuchsia_sdk")
override_repo(fake_fuchsia_sdk_extension, "fuchsia_sdk")
PiperOrigin-RevId: 709139784
Change-Id: I4d10d441c76b7a2481f15723a24f11525dba3878
Adds feature test macro for C++20 <compare> header, a pretty-printer, and tests.
Inexplicably, these types aren't enums, so can't be handled with a switch.
PiperOrigin-RevId: 704783038
Change-Id: I29688989d18f43520fe610c12a447a20d2f98c95
Some examples of asserts that now pass:
```
ASSERT_NEAR(inf, inf, 0)
ASSERT_NEAR(-inf, inf, inf)
ASSERT_NEAR(inf, x, inf) // x is any finite floating point value
```
PiperOrigin-RevId: 685748133
Change-Id: I7b3af377773e8e0031e4c6b86830cbbf76bf20c6
can cause an out of bounds access in GetCurrentExecutableName(). One way this
can happen is if the user forgets to call InitGoogleTest().
PiperOrigin-RevId: 647740658
Change-Id: Id87692aa3d515b8ae0836e474be477d2aafa3871
Changes the behavior of UnorderedElements()/UnorderedElementsAreArray() to print items-without-matchers and matchers-without-items in the case where the actual and expected are different sizes.
PiperOrigin-RevId: 635451316
Change-Id: I2181bb28a14c14cdb577af9268d403e12e942bea
If death tests were not supported the macro would try
to force the second parameter as a string. The second
parameter can be a matcher or a string.
PiperOrigin-RevId: 633199517
Change-Id: If5e71b4be0b569adb273eb4960202197bb264a8f
`EXPECT_...` statements can be used inside matcher definitions – this is an important option that is glossed over in this documentation. Users should definitely be aware of this option, since writing custom messages to the `result_listener` can be very cumbersome (and unnecessary) sometimes.
This change adds a relevant example and includes the associated error message it provides on failure.
PiperOrigin-RevId: 630206661
Change-Id: Idee00ba77ce3c1245597aa082f9cd0efff16aceb
Google Test performs hidden test registration during process
startup. For test binaries that contain a large number of tests, this
registration can be costly. In this CL, we reduce the overhead of
registration via several tactics:
- Treat CodeLocation and FilePath as value types, using std::move to
pass them around.
- Reduce string copies in various places by either passing std::string
values via std::move, or passing const-refs to std::string instances.
- Use std::to_string to stringify an int in DefaultParamName rather than
a std::stringstream.
- Pull some std::string instances out of nested loops in
ParameterizedTestSuiteInfo::RegisterTests so as to reuse some
allocations, and replace stringstream with ordinary string appends.
- Use std::unordered_map in UnitTestImpl::GetTestSuite and
ParameterizedTestSuiteRegistry::GetTestSuitePatternHolder to spend a
little memory to turn O(N) lookups into constant time lookpus.
- Use range-based for loops in a few places.
- Use emplace-ish methods to add to containers where appropriate.
All together, these changes reduce the overall runtime of a series of 50
death tests in a single Chromium test executable by ~38% due to the
fact that the registration costs are paid in every death test's child
process.
PiperOrigin-RevId: 613833210
Change-Id: I51a262a770edff98ffa1e3b60c4d78a8308f9a9f
Google Test performs hidden test registration during process
startup. For test binaries that contain a large number of tests, this
registration can be costly. In this CL, we reduce the overhead of
registration via several tactics:
- Treat CodeLocation and FilePath as value types, using std::move to
pass them around.
- Reduce string copies in various places by either passing std::string
values via std::move, or passing const-refs to std::string instances.
- Use std::to_string to stringify an int in DefaultParamName rather than
a std::stringstream.
- Pull some std::string instances out of nested loops in
ParameterizedTestSuiteInfo::RegisterTests so as to reuse some
allocations, and replace stringstream with ordinary string appends.
- Use std::unordered_map in UnitTestImpl::GetTestSuite and
ParameterizedTestSuiteRegistry::GetTestSuitePatternHolder to spend a
little memory to turn O(N) lookups into constant time lookpus.
- Use range-based for loops in various places.
- Use emplace-ish methods to add to containers where appropriate.
All together, these changes reduce the overall runtime of a series of 50
death tests in a single Chromium test executable by ~38% due to the
fact that the registration costs are paid in every death test's child
process.
PiperOrigin-RevId: 612763676
Change-Id: I1f46e012ccb9004c009e1027e4f7c38780ffb9e2
Since Fuchsia engineers rarely work within this repo, initialize a lightweight fake @fuchsia_sdk repo rather than distributing the Fuchsia SDK here.
Tested locally via `bazel query --[no]enable_bzlmod "deps(set(//googletest/test:gtest_all_test))"` (#4472)
PiperOrigin-RevId: 610826859
Change-Id: I7d41b1dbe9e7f133fe535d7337dc5bff5bf97d3a
The Fuchsia team has been using either fork branches
(https://fuchsia.googlesource.com/third_
party/github.com/google/googletest/+/refs/heads/sandbox/fuchsia_bazel_sdk)
or patch files
(https://cs.opensource.google/fuchsia/fuchsia/+/main:build/bazel/patches/googletest/fuchsia-support.bundle)
to support googletest-based tests that run against a Fuchsia target device.
As our SDK is maturing and @platforms//os:fuchsia constraint has been added
for a while now, upstream Fuchsia gtest support to reduce technical debt and
overhead.
This change is noop for non-fuchsia platform builds.
PiperOrigin-RevId: 606843739
Change-Id: I61e0eb9c641f288d7ae57354fc0b484fce013223
_Exit is standardized since C99, whereas _exit is POSIX-only.
Fixes: #4447
PiperOrigin-RevId: 605000352
Change-Id: Ibfa84edaa043bd003a21383e8148bf45be7217f6
Destruction in static teardown causes issues for Environments which own threads and try to join them in their destruction.
This may be a breaking change for users who call RUN_ALL_TESTS multiple times in the same main function if they also install environments, or those who access registered environments after RUN_ALL_TESTS.
The easiest fix is to only call RUN_ALL_TESTS once as the last line of the main function. Another potential fix is to re-register new instances of the Environment once before each call to RUN_ALL_TESTS.
PiperOrigin-RevId: 604800795
Change-Id: I37c44d4aca4a238052649f45a4b6b9cfb5355b71
Stack traces in assertion failures are an extremely useful tool for
developers tasked with investigating failing tests. It's difficult to
understate this. In contrast to ordinary test assertions (e.g.,
ASSERT_TRUE or EXPECT_FALSE), SUCCEED() is a developer-authored
directive that indicates a success codepath. In fact, the documentation
states that this directive doesn't generate any output. Generating stack
traces for uses of SUCCEED() is wasted work since they are never
printed. If this were to change one day in the future, they still would
not be useful since any emitted message would include the file and line
number where SUCCEED was used. In addition to being noise in the output
in this case, symbolization of stack traces is not free. In some
Chromium configurations, symbolization for use of SUCCEED() can incur a
cost in excess of 25 seconds for a test that otherwise takes 0-1ms; see
https://crbug.com/1517343.
In this CL, we suppress generation and emission of stack traces for
kSuccess messages to reduce the overhead of SUCCEED().
PiperOrigin-RevId: 602832162
Change-Id: I557dd6a1d3e6ed6562daf727d69fd01fe914827b
When -Wdouble-promotion is enabled, the templatized function AppropriateResolution fails to compile since its float instantiation promotes floats to doubles when doing arithmetic and comparisons. Add static casts to resolve these errors.
PiperOrigin-RevId: 600776333
Change-Id: Ia530b4bbca6ddce27caf0a817196d87efef711cb
This enables automatic color output without filesystem support, as that only requires testing to see if the output is a terminal.
Fixes: #4439
PiperOrigin-RevId: 598929397
Change-Id: Idca7490e6e090951a78cd1cdd710f41d756a68b4
Stack traces in assertion failures are an extremely useful tool for
developers tasked with investigating failing tests. It's difficult to
understate this. In contrast to ordinary test assertions (e.g.,
ASSERT_TRUE or EXPECT_FALSE), GTEST_SKIP is a developer-authored
directive to skip one or more tests. Stack traces emitted in skip
messages do not give the developer useful information, as the skip
message itself contains the code location where GTEST_SKIP was used. In
addition to being noise in the output, symbolization of stack traces is
not free. In some Chromium configurations, symbolization in a skipped
test can incur a cost in excess of 25 seconds for a test that otherwise
takes 0-1ms; see https://crbug.com/1517343#c9.
In this CL, we suppress generation and emission of stack traces for
kSkip messages to reduce the output noise and overhead of GTEST_SKIP().
PiperOrigin-RevId: 598899010
Change-Id: I46926fed452c8d7edcb3d636d8fed42cb6c0a9e9
TSAN identified a data race between updates to the states_ map (ex. in Mock::UnregisterLocked) and the iteration done in this destructor. Writes to the map use g_gmock_mutex, but the destructor does not acquire it. Acquiring the lock here fixes the data race.
It should only be possible to trigger this TSAN finding in cases where a mock object is deleted by a thread other than the main thread.
PiperOrigin-RevId: 591935393
Change-Id: I9dd1faa40058d78e165a91333346514b4b73365c
Abseil Team