Remove obsolete argument limit documentation.
Combine uses variadic templates now, so there is no inherent limit on the number of arguments.
PiperOrigin-RevId: 352580160
Print unique_ptr/shared_ptr recursively.
Given that they are smart pointers, it is unlikely that the inner object is
invalid.
PiperOrigin-RevId: 351586888
Launder buffer before reference
In GCC, directly casting the Buffer reference to another type results in
strict-aliasing violation errors. This launders the reference using an
intermediate pointer prior to creating the new reference.
PiperOrigin-RevId: 350809323
Change Matcher<T> to allow binding an implementation by value directly:
- Drop the requirement of MatcherInterface. Doing manual type erasure avoid
extra layers in many cases.
- Avoid the adaptor for `MatcherInterface<T>` and `MatcherInterface<const T&>` mismatch.
- Use a small object optimization when possible. This makes things like
`_` and `Eq(1)` really cheap and do not require memory allocations.
- Migrate some matchers to the new model to speed them up and to test the new framework. More matchers to come in future changes.
PiperOrigin-RevId: 350580998
This fixes build issues with GoogleTest when built with
-Wformat-nonliteral and unblocks updating GoogleTest in BoringSSL.
It was added in 53c478d639, which caught
some bugs. Then it was moved to the header and accidentally dropped in
482ac6ee63.
Affects macros {ASSERT|EXPECT}_{EQ|NE|LE|LT|GE|GT}.
According to removed comments, these overloads were supposed to reduce
code bloat and allow anonymous enums on GCC 4.
However, the way it works on GCC 4 and the latest GCC (10.2 by now) is
that having:
template <typename T1, typename T2>
void foo(T1, T2);
using BiggestInt = long long;
void foo(BiggestInt, BiggestInt);
the template version takes precedence for almost every combination of
integral types except for two long long integers - i.e. implicit
promotion to long long is a worse match than generating a specific
template function.
Tested on GCC 4.8.1 (as GoogleTest requires C++11 and this was
the first C++11 feature-complete release of GCC),
GCC 4.8.5 (last of 4.8.x series) and the latest GCC (10.2.0).
Remove ZX_WAIT_ASYNC_ONCE flag
It is deprecated and will soon be removed in fuchsia source.
It's currently defined as 0 and references should be changed to 0.
PiperOrigin-RevId: 346787585
Remove `status` from our internal Subprocess utility.
This facility is unused, so better to just remove it than figure out what its cross-OS semantics should be.
PiperOrigin-RevId: 345279290
Check for a high bit to see if a return value is a signal or an error code in googletest/test.
This is needed because for subprocess under python3 windows, a return value representing a C signal (such as 0x80000003) is represented as a large positive integer rather than a negative one.
PiperOrigin-RevId: 345270460
Include TargetConditionals.h before checking TARGET_OS_IPHONE
New versions of Clang require that you include TargetConditionals.h before checking the TARGET_OS_* macros. Include it where appropriate to keep the compiler happy.
PiperOrigin-RevId: 345106443
Use environment variable TEST_TMPDIR in Linux environments
for temporary directory if available otherwise use /tmp/.
Bazel sets the environment variable TEST_TMPDIR when launching tests.
PiperOrigin-RevId: 342058921
Use a tagged constructor for FlatTuple instead.
Some versions of MSVC are getting confused with that constructor and generating invalid code.
PiperOrigin-RevId: 342050957
Fix *_THROW tests under MSVC, which produce slightly different type
names from std::type_info (e.g. "class std::runtime_error" instead of
"std::runtime_error")
PiperOrigin-RevId: 341874178
Rollback change from
https://github.com/google/googletest/pull/1836. This change generates
a script on Windows to actually run each test, but the script itself
doesn't correctly report if the test passed.
This change will "break tests" that were already broken on Windows,
but weren't being reported as such.
PiperOrigin-RevId: 341850671
On a diskless system you cannot get the current directory. So if
death tests are disabled anyway, there is no point trying to
get current directory.
Without this fix, running tests on diskless systems will fail,
even when death tests are disabled.
Improve lookup of operator<< for user types
Without this fix, trying to use this class with googletest
struct Foo {};
template <typename OutputStream>
OutputStream& operator<<(OutputStream& os, const Foo&) {
os << "TemplatedStreamableInFoo";
return os;
}
results in an ambiguity error between the class' operator<< and the
operator<< in gtest-printers.h removed in this CL.
This fix also enables implicit conversions to happen, so that e.g.
we will find the base class operator<< if a subclass has no
operator<< of its own.
PiperOrigin-RevId: 336261221
Revision of recent DoubleNearPredFormat change to support more toolchains.
isnan() is a macro in C99, and std::isnan() is a function in C++11. The previous change used `isnan` directly, and broke some tests in open source.
This CL changes it to follow the practice in gmock-matchers.h, and spell uses of isnan as
(std::isnan)(f)
. The parens around `std::isnan` prevent it from being recognized as a macro in the preprocessor.
PiperOrigin-RevId: 333374377
Improve DoubleNearPredFormat output on bad epsilons
DoubleNearPredFormat will happily accept epsilon values (abs_error) that
are so small that they are meaningless. This turns EXPECT_NEAR into a
complicated and non-obvious version of EXPECT_EQ.
This change modifies DoubleNearPredFormat) so that when there is a
failure it calculates the smallest meaningful epsilon value, given the
input values, and then prints a message which explains what happened.
If a true equality test is wanted either pass a literal 0.0 as abs_error
or use EXPECT_EQ. If a check for being almost equal is wanted consider
using EXPECT_DOUBLE_EQ which, contrary to its name, verifies that the
two numbers are *almost* equal (within four ULPs).
With this change the flaky test mentioned in crbug.com/786046 gives this
output:
The difference between 4.2934311416234112e+18 and 4.2934311416234107e+18 is 512, where
4.2934311416234112e+18 evaluates to 4.2934311416234112e+18,
4.2934311416234107e+18 evaluates to 4.2934311416234107e+18.
The abs_error parameter 1.0 evaluates to 1 which is smaller than the minimum distance between doubles for numbers of this magnitude which is 512, thus making this EXPECT_NEAR check equivalent to EXPECT_EQUAL. Consider using EXPECT_DOUBLE_EQ instead.
Tested:
I confirmed that this change detects the bad epsilon value that caused
crbug.com/786046 in Chromium and added a test for the desired output.
PiperOrigin-RevId: 332946880
gtest.cc: make ColoredPrintf static
the prototype was removed from gtest.h in cl/301446904; quiets a
-Wmissing-declarations warning
PiperOrigin-RevId: 329569020
Add millisecond precision to start timestamp in XML/JSON output
- Previous timestamp had format YYYY-MM-DDThh:mm:ss, now YYYY-MM-DDThh:mm:ss.sss
- This conforms to the ISO 8601 standard
PiperOrigin-RevId: 329503623
Mention matchers as an alternative to assertions in subroutines.
Matchers are often the better choice - they can provide more informative error messages and circumvent all of the complexity described in this section.
PiperOrigin-RevId: 326332149
Updated documentation for Value-Parameterized Tests: Per-default, a TEST_P without a corresponding INSTANTIATE_TEST_SUITE_P now causes a failing test.
PiperOrigin-RevId: 323622468
