VS_SETTINGS is a list of key value pairs that get written as item
metadata for the associated non-built file. The entire value of
VS_SETTINGS is evaluated as a generator expression.
Introducing a new DISCOVERY_MODE mode option, which provides greater control
over when gtest_discover_tests perforsm test discovery.
It has two supported modes:
* POST_BUILD
* PRE_TEST
POST_BUILD is the default behavior, which adds a POST_BUILD command
to perform test discovery after the test has been built.
PRE_TEST is a new mode, which delays test discovery until test execution.
DISCOVERY_MODE can be controlled in two ways:
1. Setting the DISCOVERY_MODE when calling gtest_discover_tests
2. Setting the global CMAKE_GTEST_DISCOVER_TESTS_DISCOVERY_MODE
prior to calling gtest_discover_tests
By delaying test discovery until ctest runtime,
we can provide better cross compile support,
because we're more likely to be in an environment that can run the test executable.
This was achieved by moving the command for test discovery
into the generated ctest include file.
In PRE_TEST mode, the generated include file now has the form:
if(EXISTS "path/to/test.exe")
if("path/to/test.exe" IS_NEWER_THAN "path/to/ctest_file")
// command to discover tests
// and generate ctest_file
endif()
include("path/to/ctest_file")
endif()
elseif()
// test not built
endif()
Which generates the appropriate CTest files at runtime
and regenerates the CTest files when the executable is updated.
In the old approach, test discovery was always added as POST_BUILD step
and the executable was ran in order to scan for tests.
If the test executable failed to run for any reason,
this resulted in a link failure.
This failure could more commonly occur when cross compiling,
because your build environment might not have
the appropriate runtime environment dlls required to run the test executable.
I also ran into the issue when compiling a Qt application using Qt Creator.
Qt Creator manages its build and runtime environments separately
and only adds the Qt dlls to the environment during runtime -- not during build.
So when gtest_discover_tests ran my test executable,
it promptly crashed because the environment wasn't correct,
which resulted in a build failure.
Setting the DISCOVERY_MODE to PRE_TEST fixed my build failure
by delaying test discovery until runtime,
because this allowed the test exe to run in the proper environment.
A few non-trivial implementation details worth noting:
1. bracket_arguments
In the PRE_TEST side, parameters whose contents might contain special characters,
need to be handled with great care.
To aid in escaping these list arguments, backslashes, and other special characters,
and ensuring that they are preserved in the generated file,
bracket arguments (i.e. [== <...> ==]) are used.
For example:
gtest_discover_tests(
...
EXTRA_ARGS how now "\"brown\" cow"
)
Generates a file with the following call:
gtest_discover_tests_impl(
...
TEST_EXTRA_ARGS [==[how;now;"brown" cow]==]
)
This way the arguments are forwarded correctly.
2. multi-config generators
Multi-Config generators (e.g. MSBuild) now work more correctly in
PRE_TEST than POST_BUILD.
PRE_TEST is more correct because it will generate a unique CTest file for each configuration.
It generates a per config file responsible for test discovery:
foo[1]_include-Debug.cmake
foo[1]_include-MinSizeRel.cmake
foo[1]_include-Release.cmake
foo[1]_include-RelWithDebInfo.cmake
A per config file for containing the google tests:
foo[1]_tests-Debug.cmake
And an outer ctest file:
foo[1]_include-Debug.cmake
That is generically written to include the correct configuration file
by looking at the value of ${CTEST_CONFIGURATION_TYPE} when CTest runs.
POST_BUILD, in contrast, preserves the existing functionality.
Tests are disocvered based on the last configuration that was chosen
and only a single file is produced:
foo[1]_include.cmake
foo[1]_tests.cmake
But it runs with whatever executable requested by ctest --config,
which means it's possible to run the wrong tests
if some tests were enabled in one configuration but not another.
Move test discovery logic into new gtest_discover_tests_impl method
and make GoogleTestAddTests aware of whether it is being launched in
CMake's script mode.
When launched in script mode, gtest_discover_tests_impl is called
passing arguments obtained from the definitions passed into the call to cmake.
(i.e. cmake -P GoogleTestAddTests -D <arg1> -D <arg2> ...)
This preserves the existing behavior assumed by GoogleTest.cmake.
Unit tests are unchanged and still pass.
Looking ahead, it also allows GoogleTestAddTests to be included in generated files
and call gtest_discover_tests_impl to perform test discovery at test runtime
with the new PRE_TEST discovery mode introduced later in this branch.
My original approach attempted to call execute_process(cmake -P ...) in
the generated file, the same way POST_BUILD is doing, but I ran into
difficulties serializing the command arguments correctly.
By exposing a way to call gtest_discover_tests_impl directly from our generated file,
we remove a layer of shell quoting / parsing that our arguments have to survive,
which simplifies the act of producing a generated file that behaves the
same as its POST_BUILD counterpart.
1994f950ff cmake: List valid values for --trace-format on the command line
e39766d84a Help: Fix documentation of --trace-format parameter
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !4500
1994f950ff cmake: List valid values for --trace-format on the command line
e39766d84a Help: Fix documentation of --trace-format parameter
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !4500
CMake 3.16 and below allow multiple `export()` calls with the same output
file even without using `APPEND`. The implementation worked by accident
by leaking memory. Refactoring in commit 5444a8095d (cmGlobalGenerator:
modernize memrory managemenbt, 2019-12-29, v3.17.0-rc1~239^2) cleaned up
that memory leak and converted it to a use-after-free instead.
The problem is caused by using the `cmGlobalGenerator::BuildExportSets`
map to own `cmExportBuildFileGenerator` instances. It can own only
one instance per output FILE name at a time, so repeating use of the
same file now frees the old `cmExportBuildFileGenerator` instance
and leaves the pointer in the `cmMakefile::ExportBuildFileGenerators`
vector dangling. Move ownership of the instances into `cmMakefile`'s
vector since its entries are not replaced on a repeat output FILE.
In future work we should introduce a policy to error out on this case.
For now simply fix the use-after-free to restore CMake <= 3.16 behavior.
Fixes: #20469
The check_language sets internal variables with a common name in the
caller's scope: `result`, `output` and `content`.
They are now prefixed with `_cl_`,
inspired by the CheckLibraryExists module.
46064c8193 FindRuby: Add support for versions up to 2.7
675eaf3bd0 FindRuby: Update MSVC runtime library selection
b970e25d98 FindRuby: Remove extra whitespace
ecdace4d61 FindRuby: Include FPHSA closer to where it is used
f52f496138 FindRuby: Provide Ruby_LIBRARIES result variable
b00d736a0b FindRuby: Add dedicated tests
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !4481
In commit 3dab4682f6 (FindPython: reduces consumption of resources,
2020-02-10, v3.17.0-rc1~11^2) we accidentally broke the python
executable version validation when the "LOCATION" strategy is used
with the plain `FindPython` module. Fix the logic and add test
cases covering those combinations.
Fixes: #20465
Use the `MSVC_TOOLSET_VERSION` variable computed by CMake to get the
matching Ruby library name component.
Inspired-by: Julien Marrec <julien.marrec@gmail.com>
Matt Davies