[fdio] Improve fdio_pipe_half signature, step 3.
The return value on fdio_pipe_half conflated two things: the error code
on failure (as a zx_status_t) or a file descriptor on success. This
technically worked, because they're both ints, the error code was always
negative, and the file descriptor always positive. However, the stated
return type of zx_status_t was misleading. This changes the signature
such that it always returns an actual zx_status_t, and the file
descriptor is returned through a pointer argument.
Also remove the last argument, since it was always given the same value.
This needs to be done as a soft transition because it's called from the
Dart runtime, from googletest, and from Crashpad, and Crashpad and
Chromium both depend on googletest on Fuchsia. The steps are as follows:
1) Add fdio_pipe_half2.
2) Update Dart to use fdio_pipe_half2.
3) Update googletest to use fdio_pipe_half2.
4) Roll updated googletest into Chronium.
5) Update Crashpad to use fdio_pipe_half2 and roll updated googletest into it.
6) Update fdio_pipe_half to match fdio_pipe_half2.
7) Update Dart to use fdio_pipe_half again.
8) Update googletest to use fdio_pipe_half again.
9) Roll updated googletest into Chronium.
10) Update Crashpad to use fdio_pipe_half again and roll updated googletest into it.
11) Remove fdio_pipe_half2.
This is step 3.
PiperOrigin-RevId: 241957137
Add `-frtti` to the compiler with the base flags case so that RTTI is
enabled by default with clang. Add its inverse analog,
`cxx_no_rtti_flags` in order to test the case with RTTI off, similar to
gcc.
This reduces the amount of testing/support overhead needed in the
non-RTTI case with clang, as the tests currently fail when these two
features are off with version 1.8.1. This something I used in when
investigating test failures on FreeBSD, as the tests that rely on
RTTI were failing with googletest 1.8.1 on the OS platform.
More investigation is being done to determine how this should be fixed
on FreeBSD 11.2-RELEASE with ports, as the package doesn't currently
compile the tests, and when enabled (based on my WIP diff), the tests
fail in similar ways.
Signed-off-by: Enji Cooper <yaneurabeya@gmail.com>
Add HWASan annotations.
These mirror existing ASan annotations.
HWASan uses memory (address) tagging to detect memory errors:
https://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html
It inserts a random tag in the MSB of heap and stack allocation addresses. This tag dominates pointer comparison in StackGrowsDown(), making the result non-deterministic, and entirely unrelated to the actual stack growth direction. The function attribute disables this behavior.
The annotations in gtest-printers are there because the printers are used to basically dump memory. The sanitizers may have ideas why this memory should not be accessed, and that is counter productive. In particular, the test may access only part of an array, but in case of a test failure gtest will dump the entire array which may contain uninitialized bytes - that's what SANITIZE_MEMORY annotation is for. There are similar reasons for ADDRESS and THREAD annotations. HWADDRESS in its current implementation can not cause issues there, I believe, but it falls under the same umbrella of tools whose checking should not apply to test printers because it is not the code under test.
PiperOrigin-RevId: 241379822
Remove support for "global" ::string and ::wstring types.
This support existed for legacy codebases that existed from before namespaces
where a thing. It is no longer necessary.
PiperOrigin-RevId: 241335738
gtest prior to this change would completely ignore `GTEST_SKIP()` if
called in `Environment::SetUp()`, instead of bailing out early, unlike
`Test::SetUp()`, which would cause the tests themselves to be skipped.
The only way (prior to this change) to skip the tests would be to
trigger a fatal error via `GTEST_FAIL()`.
Desirable behavior, in this case, when dealing with
`Environment::SetUp()` is to check for prerequisites on a system
(example, kernel supports a particular featureset, e.g., capsicum), and
skip the tests. The alternatives prior to this change would be
undesirable:
- Failing sends the wrong message to the test user, as the result of the
tests is indeterminate, not failed.
- Having to add per-test class abstractions that override `SetUp()` to
test for the capsicum feature set, then skip all of the tests in their
respective SetUp fixtures, would be a lot of human and computational
work; checking for the feature would need to be done for all of the
tests, instead of once for all of the tests.
For those reasons, making `Environment::SetUp()` handle `GTEST_SKIP()`,
by not executing the testcases, is the most desirable solution.
In order to properly diagnose what happened when running the tests if
they are skipped, print out the diagnostics in an ad hoc manner.
Update the documentation to note this change and integrate a new test,
gtest_skip_in_environment_setup_test, into the test suite.
This change addresses #2189.
Signed-off-by: Enji Cooper <yaneurabeya@gmail.com>
Building all test binaries under their respective subtrees makes
building the project via cmake easier to grok without additional hacks.
In particular, when dealing with the conversion I proposed in
https://reviews.freebsd.org/D19430 (switching from autotools to cmake),
I ran into unexpected gtest prefixing under the googlemock directory, as
opposed to the googletest directory. Example:
Before: `googlemock/gtest/googletest-break-on-failure-unittest_`
After: `googletest/googletest-break-on-failure-unittest_`
The latter form is easier to translate to packaging manifests when
building googlemock is disabled, as well as enabled, as the path remains
consistent when the feature flag is disabled.
Signed-off-by: Enji Cooper <yaneurabeya@gmail.com>
Fix emission of -Wzero-as-null-pointer-constant when comparing integers.
The following code fails to compile:
#pragma clang diagnostic error "-Wzero-as-null-pointer-constant"
void foo() {
EXPECT_EQ(0, 0);
}
This happens because gtest checks the first argument to EXPECT_EQ and
ASSERT_EQ is a null pointer constant. The magic it does to do this causes the
warning to be emitted.
This patch removes that check. It replaces the explicit check with a Compare
overload that can only be selected when 0 or nullptr is passed on the LHS
with a pointer on the right.
This patch does not suppress -Wzero-as-null-pointer-constant when users
are actually using it as NULL.
PiperOrigin-RevId: 236654634
Let embedders customize GTEST_INTERNAL_DEPRECATED().
GTEST_INTERNAL_DEPRECATED is currently used to nudge googletest users to migrate off old TEST_CASE macros to the new TEST_SUITE macros. This move is non-trivial for Chromium (see https://crbug.com/925652), and might be difficult for other big projects with many dependencies.
This CL facilitates moving off of deprecated APIs by making it possible for an embedder to define GTEST_INTERNAL_DEPRECATED() in gtest/internal/custom/gtest-port.h. Example usage:
1) #define GTEST_INTERNAL_DEPRECATED() to nothing, to disable deprecation warnings while migrating off googletest's deprecated APIs. This can be preferable to having to disable all deprecation warnings (-Wno-error=deprecated or -Wno-deprecated-declarations).
2) #define GTEST_INTERNAL_DEPRECATED() for an unsupported compiler.
PiperOrigin-RevId: 236171043
Due to some caveats in the FreeBSD build system and the fact that the
source file is used to compile 2 different death tests with different
flags, I needed (as a shortterm workaround) to copy the test to 2
differently named files.
While this works for compiling the test, as I discovered, this doesn't
work with running `CxxExceptionDeathTest.PrintsMessageForStdException`,
as the testcase hardcodes `googletest-death-test_ex_test.cc`. Use `__FILE__`
when looking for failures, as opposed to looking for the hardcoded name
as it can vary depending on how the test was built.
Signed-off-by: Enji Cooper <yaneurabeya@gmail.com>
