Previously this excluded callables that return non-moveable types. This is the
same as the
[libc++ std::is_invocable_r bug](https://github.com/llvm/llvm-project/issues/55346)
fixed by
[this commit](https://github.com/llvm/llvm-project/commit/c3a24882903d): it's
wrong to use std::is_convertible for checking the return type, since (despite
its name) that doesn't check the standard-defined notion of "implicitly
convertible". Instead we must base the check on whether the source type can be
used as an argument to a function that accepts the destination type.
PiperOrigin-RevId: 451341205
Change-Id: I2530051312a0361ea7a2ce26993ae973c9242089
In order to make the diff more readable in an upcoming commit that requires the
method to be templated on the action's result type.
PiperOrigin-RevId: 451157029
Change-Id: I57beb7544efccd0459efb3a1f039ea45cd7c7602
`Return(x)` can now be used directly with `WillOnce` (the only place it makes
sense in the type system), without using `ByMove`.
PiperOrigin-RevId: 448380066
Change-Id: Ia71cc60ccbc3b99720662731a2d309735a5ce7c8
It doesn't make semantic sense for the conversion to modify the input, and the
fact that it's allowed to do so appears to have just been a historical accident.
PiperOrigin-RevId: 448135555
Change-Id: Id10f17af38cf3947ee25fe10654d97527173ebfc
The way temporary directories were discovered from
environment variables in different operating systems
was inconsistent.
On Windows, only the system-specific environment variable
TEMP was considered, but not the bazel-common TEST_TMPDIR.
On Unix, only TEST_TMPDIR was considered, but not
the otherwise typical system specific TMPDIR.
Now, always consider TEST_TMPDIR first, followed by the
system-typical environment variable (TEMP on Windows, TMPDIR on Unix)
before falling back to a default on that particular system.
Also: the value for the temporary directory coming from the
environment variable was only checked for a trailing directory
separator on Windows (and appended if needed), not on the other
platforms.
Make this also consistent accross the systems: Now always apply
the same logic to all of these (to accomodate
tests that just concatenate TempDir() without directory separator).
PiperOrigin-RevId: 447909830
Change-Id: I9ea17acdf5944eb2a965615fd9cf142878c33a58
Better document requirements, API decisions, and historical accidents. Make an
implicit conversion easier and in a more appropriate place, and ease the burden
of some assertions in the conversion operator. Stop using the legacy
ActionInterface style for defining the action.
PiperOrigin-RevId: 447894892
Change-Id: I179e23ec2abdd9bf05c204ab18dbb492f1372e8e
So that it can be referenced in conversion operators for actions that need to
know the concrete return type.
PiperOrigin-RevId: 447889344
Change-Id: I643d3298bc8effd08741282a956c221f9d67d378
This CL adds a couple missing overrides in the googletest sources.
These were found downstream when -Wsuggest-override and
-Wsuggest-destructor-override were enabled.
PiperOrigin-RevId: 447754883
Change-Id: I7bf35a8757cbc5ae157827037aa3d13f47392406
Currently MacOS falls back to generic /tmp, but
for all intents and purposes it should behave like
other Unixes using the TEST_TMPDIR environment variable
if available (this environment variable is set in bazel,
which sets up a unique temp directory for the test
process).
While at it, remove an incorrect #endif comment, that
looks like a leftover from some older implementation.
PiperOrigin-RevId: 446108391
Change-Id: I118eacf6e86a41d26cb81a130f7c54cccc0c5665
Rather than branching on whether the return type is ByMoveWrapper within
ReturnAction itself, hoist the distinction to outside. This allows the main
class template to be modified without worrying about this special case, which
means we can stop using a shared pointer to the value (introduced as a
linked_ptr in commit 3d1c78b2bf to support ByMove) in this commit and simplify
the class template further in a future commit with the eventual aim of directly
supporting move-only result types.
PiperOrigin-RevId: 445938943
Change-Id: I7bc71ea301d5e493ac6ecbe57d62738a48a2721a
Commit a070cbd91c added an implicit cast to this path but didn't leave a very
clear explanation for why it was desirable, a clear example, or even test
coverage. Add a better comment and a test that fails when the implicit cast is
removed.
PiperOrigin-RevId: 444871311
Change-Id: I127982fa8d5bce9b6d1b68177c12dc0709449164
bazel build --define=absl=1 ...
A dependency on RE2 is now required when building GoogleTest with Abseil.
Using RE2 will provide a consistent cross-platform regex experience.
Users will need to add the com_googlesource_code_re2, bazel_skylib,
and platforms repository to their WORKSPACE files. See our WORKSPACE
file in the root directory of this project for an example of how to
add the dependencies.
Please note that the com_googlesource_code_re2 dependency must use a
commit from the `abseil` branch of the project:
https://github.com/google/re2/tree/abseil
PiperOrigin-RevId: 444650118
Change-Id: I45c55b26684c0c50d721a05b81c5f8a0c092400f
This is necessary for generic support of these actions, since `DoAll` is a
frequently-used action wrapper.
PiperOrigin-RevId: 444561964
Change-Id: I02edb55e35ab4207fbd71e371255a319c8253136
This provides a type-safe way for an action to express that it wants to be
called only once, or to capture move-only objects. It is a generalization of
the type system-evading hack in ByMove, with the improvement that it works for
_any_ action (including user-defined ones), and correctly expresses that the
action can only be used with WillOnce. I'll make existing actions benefit in a
future commit.
PiperOrigin-RevId: 440496139
Change-Id: I4145d191cca5655995ef41360bb126c123cb41d3
Avoid instantiating functions like std::get<index> for an out of range index
when doing SFINAE on the invocability of the action itself.
PiperOrigin-RevId: 439415110
Change-Id: Ifc20285a6d526c34830870cd1910c2b2b92e1e81
When built with `--define=absl=1` under Bazel, GoogleTest
flags use ABSL_FLAG instead of GoogleTest's own implementation.
There are some minor behavior differences in this mode.
The most notable difference is that unrecognized flags result
in a flag parsing error, and are not returned to the user though
a modified argc/argv, unless they appear after the positional
argument delimiter ("--").
For example, to pass a non-Abseil flag, you would have to do
./mytest --gtest_color=false -- --myflag=myvalue
The documentation at https://abseil.io/docs/cpp/guides/flags
may be helpful in understanding the behavior.
There are some other minor differences. For example,
passing --help results in the program returning 1 instead of 0.
https://github.com/google/googletest/issues/3646
PiperOrigin-RevId: 439312700
Change-Id: Id696a25f50f24a5b1785c45ca8fa59794f86fd5c
The one large test does not link in limited memory environments
Closes#3653
PiperOrigin-RevId: 436753193
Change-Id: Idd59b6509994fc642147b88279ee791cd1d7bdd0
Currently, the "[ DISABLED ]" banner is printed for every test in a suite.
When iterating on a single test gtest_filter this is very noisy.
PiperOrigin-RevId: 436489088
Change-Id: If337087a7a0986b073fabf2b0a55d26485eb5c37
A few tests are examining code locations and looking af the resulting line
numbers to verify that GoogleTest shows those to users correctly. Some of those
locations change when clang-format is run. For those locations, I've wrapped
portions in:
// clang-format off
...
// clang-format on
There may be other locations that are currently not tickled by running
clang-format.
PiperOrigin-RevId: 434844712
Change-Id: I3a9f0a6f39eff741c576b6de389bef9b1d11139d
