In gtest_discover_tests(), the TIMEOUT keyword was making it
impossible to set the TIMEOUT test property via the PROPERTIES
keyword. This would be a frequent case, but it doesn't complain
and instead silently does something different to what would
normally be expected. The TIMEOUT keyword has been renamed
to DISCOVERY_TIMEOUT, thereby removing the clash.
This is a breaking change. 3.10.1 and 3.10.2 were the only versions
that supported the TIMEOUT keyword and uses of it were likely
not working as intended.
Fixes: #17801
Until CMake 3.10 a list of source files that had the AUTOUIC_OPTIONS property
populated was kept in `cmMakefile::QtUiFilesWithOptions`. In the process to
remove all AUTOUIC related code from `cmMakefile` for CMake 3.10, the pre
filtered list was replaced by a loop in `cmQtAutoGeneratorInitializer` over
all source files in the `cmMakefile`. This loop introduced the problem that
file paths were computed for source files that weren't in the target's sources
and that might not even have existed. If the path for an unused and not
existing file was computed a `cmake::FATAL_ERROR` with the error message
"Cannot find source file:" was thrown nevertheless.
This caused some projects to fail in CMake 3.10.
This patch adds a test for path errors in the loops in
`cmQtAutoGeneratorInitializer` that iterate over all source files in a
`cmMakefile`. If a path error appears, the file is silently ignored.
If the file is part of the target's sources, the path error will still be
caught in the loop over all the target's sources.
This is the fix for CMake 3.10.1.
Backport KWIML commit 6fc81d8816 (abi.h: Update RISC-V support for
revised macro names, 2017-12-30) to the CMake 3.10 release branch.
The upstream RISC-V project changed some compiler symbols so `__riscv__`
must now be renamed `__riscv`.
Patch-by: Richard W.M. Jones <rjones@redhat.com>
In the AutogenInfo.cmake file the separator for nested lists was
`@LSEP@` which led to a speed regression because the `@` character
triggered an (unsuccessful) expression evaluation.
By setting the policy version of the CMake instance in the
`_autogen` target to 3.9, the OLD `@` evaluating behavior
controlled by policy CMP0053 is disabled.
Also the nested lists separator string is changed to `<<<S>>>`,
which solves the problem twofold.
Issue: #17570
This MR fixes a number of bugs stemming from legacy variable handling,
partially introduced by commit v3.10.0-rc1~103^2 (FindMPI: Add support
for FreeBSD and SLES, 2017-08-09).
* `MPI_COMPILE_OPTIONS` and `MPI_COMPILE_DEFINITIONS` legacy hints were not passed correctly.
* Legacy variables could be parsed after they were set by `FindMPI` cause erroneous hinting on subsequent runs.
* In the case of the compiler supporting MPI implicitly:
* not all consumption variables would be correctly reset.
* `MPI_<LANG>_HEADER_DIR` would be set unintentionally.
* `MPI_<LANG>_LIB_NAMES` appeared in cache despite being ignored.
* on subsequent passes, `MPI_<LANG>_COMPILER` could be corrupted.
* the C-to-CXX copy logic would take place even if `MPI_SKIP_GUESSING` was turned on
* if only some language compilers were given, inconsistent search behavior could be observed
Issue: #17538
The IAR compiler test works when a project specifies LANGUAGES C CXX
but fails if CXX comes before C. This change makes it work regardless
of the order.
Shared libraries linked with CUDA now properly setup booth the correct
install_name but also specify headerpad_max_install_names so that the names
can be changed post build.
Revert the library target dependency forwarding to the `_autogen` target
to the 3.9.6 way, only now using `GetLinkImplementationLibraries`.
Fixes: #17278
The backtrace information is very repetitive and hugely increases the
size of the codemodel object. We need to remove it until an alternative
representation can be developed. Revert commit v3.10.0-rc1~393^2
(server: Report backtraces in codemodel response, 2017-06-20), except
for the protocol version number (because it indicates other new things).
Unfortunately this is incompatible with clients that expect the
"crossReferences" field in targets. However, the regression in memory
usage is quite serious, especially on large projects, and therefore
breaks even older clients that do not use backtraces. Since the
"crossReferences" field was only provided by one release (3.10.0), it is
simplest to revert it outright for 3.10.1.
Fixes: #17502
Add a test to verify expected behavior when a test executable on which
gtest_discover_tests is invoked was not built. (The test for the newly
added timeout conveniently sets up this situation, so this new test was
almost trivial to add.)
Add a TIMEOUT option to gtest_discover_tests. This provides a
work-around in case a test goes out to lunch, rather than causing the
build to simply hang. (Although this is still a problem with the user's
project, hanging the build is not cool, especially in the case of
automatically running CI builds. It is much preferred that the build
should actively fail in this case, and it is trivially easy for us to
implement that.)
Due to a erroneous logical dependency on a temporary variable, FindMPI
would depend its FOUND status on the set of flags needed to support
compiler-implicit MPI support after the first configure run.
This behavior is especially causing a misreported NOT FOUND status after
the first run on MPI implementations without MPI compiler like MS-MPI or
MPICH2 on Windows.
Due to a minor logic error, MPI_LIBRARY, MPI_EXTRA_LIBRARY and
MPI_<LANG>_LIBRARIES hints are currently not correctly handled and
discarded.
Additionally, if either C or CXX aren't enabled, there subsequent calls
to FindMPI will cause cache variables of the sort "MPI_MPI_EXTRA_LIBRARY-NOTFOUND"
to appear due to an error in the compatibility handling. (Doesn't affect
functionality)
Refactoring in commit v3.7.0-rc1~523^2 (CPack/PackageMaker: port to
cmXMLWriter, 2016-06-08) accidentally broke `distribution.dist`
generation when there are dependencies between components and the pkg
(or the derived productbuild) generator are used. The resulting
package cannot be installed and instead the Apple Installer app
shows a "Javascript error" message.
The problem is that a `std::ostringstream` is constructed with some
content, but the stream pointer still points to the beginning of the
stream, so subsequent writes overwrite the initial content instead of
appending.
Some versions of clang 5 (with libc++) have a problem with
`unordered_map` under `-std=gnu++1z`:
/usr/include/c++/__hash_table:1134:43: error: conflicting types for '__hash_table<_Tp, _Hash, _Equal, _Alloc>'
Include `unordered_map` in our test source so that we reject this
combination and fall back to an older C++ standard flag.
Fixes: #17526
The check added by commit v3.10.0-rc2~2^2 (Clang: Diagnose unsupported
GNU-like clang targeting MSVC ABI, 2017-10-10) is incorrectly detecting
clang-cl 6.0 as GNU-like. Currently cmake is testing if the clang
compiler accepts `--version` to see if it accepts GNU style flags.
However, with the latest llvm snapshot this also works for clang-cl:
> clang-cl --version
clang version 6.0.0 (trunk)
Target: x86_64-pc-windows-msvc
Thread model: posix
InstalledDir: C:\Program Files\LLVM\bin
So instead we should use the `/?` flag which fails with clang but
works with clang-cl:
> clang-cl /? &> /dev/null; echo $?
0
> clang /? &> /dev/null; echo $?
1
Fixes: #17518
According to the documentation, tests can be discovered for a target
multiple times by using a different prefix and/or suffix to ensure name
uniqueness. However, while this worked for gtest_add_tests, it did not
work with gtest_discover_tests because the generated file that sets up
the tests was named based only on the target name, and so subsequent
discovery from the same target would clobber earlier discovery.
Fix this by introducing a counter that records how many times discovery
has been used on a target, and use this to generate unique names of the
generated test list files.
Brad King