The DELETE option to ctest_coverage_collect_gcov now properly
removes all the .gcov files that were created by this function.
Previously it left behind any files that were excluded by
CTEST_CUSTOM_COVERAGE_EXCLUDE.
This option now also deletes the following files/directory that
are created by ctest_coverage_collect_gcov:
data.json
coverage_file_list.txt
the uncovered/ directory
2a99fae1 FindMPI: Recognize `.lib` file names for specification of link libraries
c4417b09 FindMPI: Recognize /LIBPATH for specification of linker paths.
GitHub for Windows now installs in
$ENV{LOCALAPPDATA}/Github/PortableGit*/cmd
instead of
$ENV{LOCALAPPDATA}/Github/PortableGit*/bin
Search the new location first but keep the old one in case people have
older versions installed.
a1c40291 GetPrerequisites: Look for VS tools using environment variables
c79f88b0 GetPrerequisites: add path of VS 10,11,12,14 in order to find dumpbin.exe
The `VS*COMNTOOLS` environment variables specify locations of VS tools
and are set during the VS installation. Use them in addition to the
hard-coded default install locations.
We define `NDEBUG` without a space after the `-D` option for most
compilers. Remove the space for MSVC (and Intel Fortran) for
consistency. The MS compiler technically does not document that
the `-D` argument may be separated from its value, though every
version to date supports it.
4d8c988c ExternalProject: Fix file download script with CMAKE_TLS_{VERIFY,CAINFO}
23a71e4e ExternalProject: Tell Git not to verify certs only if TLS_VERIFY is OFF
Teach CTestCoverageCollectGCOV to honor the
CTEST_EXTRA_COVERAGE_GLOB variable. When this variable is set,
this module will glob for matching source files that were not
covered and include them in the resulting tar file.
Since commit 272779ce (ExternalProject: Allow TLS_VERIFY for git clones,
2016-04-01) we pass the `-c http.sslVerify=false` option to `git clone`
even if no explicit `TLS_VERIFY` option was set. This changes behavior
because we used to use the default Git behavior by default. Revise the
logic to preserve the old default behavior by passing the new option
only if `TLS_VERIFY` was explicitly passed as `OFF`.
While at it, also honor `CMAKE_TLS_VERIFY` if the explicit `TLS_VERIFY`
option is not given.
On mingw-w64 the GNU Fortran compiler does not define `__MINGW32__` or
any similar indicator. Fix `CMAKE_Fortran_PLATFORM_ID` detection in
this case by falling back to preprocessing a `.c` source file even
when the compiler id is already detected.
c18d91ad Help: add release notes for topic 'ctest-run-submodule-sync'
7f560743 cmCTestGIT: run `git submodule sync` before updating submodules
06b310b5 cmCTestGIT: add an option to initialize submodules on update
56c1ea40 cmCTestGIT: fix git version references
Add NAMES_PER_DIR to all find_library invocations so that we consider
all possible names in each search directory before moving on to the next
directory. Otherwise we may not find self-built libraries first even if
they appear early in the search path.
Use `CMAKE_<LANG>_COMPILER_LOADED` to detect enabled languages because
`if( _LANGUAGES_ MATCHES C )` is always true on Windows as the RC
language is activated automatically and matches C.
Consistently glob for .gcda files in the binary directory.
Previously the behavior of this function depended on the
current working directory that it was called from.
Currently, CTest will not initialize any submodules within the already
checked out source tree. Add an option to do so. The use case for not
doing so is that some submodules may not be necessary for the current
test and keeping network usage down may be important.
Symbolic links that point to external
location no longer cause cmake to fail
with string out of bounds error but
are instead packaged as non relocatable
symlinks and print out a warning message.
This placeholder was added to the compilation rules for other languages
by commit v3.4.0-rc1~342^2 (Factor an <INCLUDES> placeholder out of
<FLAGS> in rule variables, 2015-07-13) but ASM was incorrectly left out.
Signed-off-by: Gregor Jasny <gjasny@googlemail.com>
RPM supports setting of default user, group,
file and directory permissions that will be
applied for files in package unless specified
per file/dir with attr setting
This is related to bug report 14714
Prior to Xcode 7.3 a XCTest bundle was built like an ordinary
bundle. But starting with Xcode 7.3 it is implicitely built as
a plugin for the test host. It looks like in that case the
'Build Products Path' is ignored.
This patch simply sets the CMake target output path to the value
Xcode expects internally to get a matching view of both.
The command:
xcodebuild -showBuildSettings -target CocoaExampleTests -configuration Debug
Output with Xcode 7.2:
TARGET_BUILD_DIR = .../Tests/XCTest/Debug
Output with Xcode 7.3
TARGET_BUILD_DIR = .../Tests/XCTest/Debug/CocoaExample.app/Contents/PlugIns
Signed-off-by: Gregor Jasny <gjasny@googlemail.com>
In v2.8.12.1-654-g7621ad6 I added underscore prefixes to
CPACK_ADDCOMP_(STR|UNAME) to prevent them from being automatically
written to CPackConfig.cmake.
Do the same for CPACK_ADDGRP_(STR|UNAME) and CPACK_INSTTYPE_(STR|UNAME).
Since commit v3.5.0-rc1~47^2 (FindCUDA: Support special characters in
path, 2016-01-15) our add_custom_command calls use VERBATIM so that
CMake will automatically quote special characters correctly. Fix the
separate compilation code path to not add its own quoting when the
VERBATIM option will be used.
The fix in commit v3.5.0-rc1~27^2 (FindPkgConfig: set standard variables
in the cache, 2016-01-20) added the wrong variable name to the cache.
The test was only testing that the cache variable existed, not that it
also had the correct value. Update the test to ensure that the cache
value matches the local variable value.
Reported-by: Bernd Lörwald
6a6e5d89 GenerateExportHeader: Allow common NO_DEPRECATED_MACRO_NAME for multiple libs
be5a8973 GenerateExportHeader: Do not define DEFINE_NO_DEPRECATED (#16022)
Previously we allowed this definition to persist outside our header.
This would cause conflicts across multiple such headers because the name
was always the same. Fix this by avoiding the definition altogether.
When system-provided packages are upgraded we must re-compile sources
depending on their headers. Use `-MD` instead of `-MMD` so that the
generated depfiles do not exclude system headers.
Suggested-by: Jussi Judin
If list of valid target architectures is empty for given SDK then there will
be no VALID_ARCHS build setting returned by Xcode. Return "" (empty string)
explicitly in this case. This may happens if CMAKE_IOS_INSTALL_COMBINED is ON
but only one architecture used in target.
Depending upon the configuration, certain components may or may not
be installed, for example Boost.Regex, but other components may
still have header dependencies upon these components which will
obviously fail to work. Since we can't make a sensible
determination with the hardcoded dependency information, we
choose to interpret these dependencies less strictly.
491b41dd Help: Add notes for topic 'vs-clang-cl'
ad6d27ac Tests: do not build PrecompiledHeader on Clang/C2
a0f0541f Tests: fix PDBDirectoryAndName on Clang/C2
3541af67 Tests: fix Plugin building on Clang/C2
1902c293 Tests: fix complexOneConfig building on Clang/C2
cab2ec11 Tests: fix Complex building on Clang/C2
ada3736c Tests: fix Module.GenerateExportHeader building on Clang/C2
123b7e13 Tests: fix AliasTarget building on Clang/C2
445d4d4b VS 14: Add flag map for -std= to CppLanguageStandard tag in project files
0a785eb4 Features: Clang has no cxx_decltype_incomplete_return_type in MSVC sim mode
2c2ec488 VS: in Clang/C2 toolset, setup correct compiler settings
37afe00f CMakeDetermineCompilerId: Add detection of clang.exe bundled with VS
Since commit v3.1.0-rc1~564^2 (OS X: Use -iframework for system
framework directories, 2014-05-05) we test the version of Clang is smaller
that 3.1 to see if it supports -iframework.
Considering that "iframework" support has been added in clang@r142418
(Frontend: Support -iframework.) prior to clang 3.1, this made sense.
That said, considering that support for multiple -iframework parameters
has been added later in clang@r164607 (-iframework should allow separate
arguments. ) prior to clang 3.2, this commit updates the check to enable
framework support only if version is >= 3.2
We originally deprecated this module in commit v3.5.0-rc1~295^2
(CMakeForceCompiler: Deprecate this module and its macros, 2015-10-19).
Then a use case was found to still require the module so the deprecation
was reverted for CMake 3.5 by commit v3.5.0-rc3~4^2 (CMakeForceCompiler:
De-deprecate until more use cases have alternatives, 2016-02-17). Since
then `CMAKE_TRY_COMPILE_TARGET_TYPE` was introduced to provide an
alternative solution for that use case. Deprecate the module again.
Canonicalize the input paths so we treat them both consistently,
in particular when comparing them via string operations. This
is needed for calls like
fixup_bundle("${CMAKE_INSTALL_PREFIX}/../test" ...)
Suggested-by: Benjamin Ballet <bballet@ivsweb.com>
Use recommended case for variable names. i.e. matching name of the
module as passed to `find_package`.
For backwards compatibility, the upper case versions of both input and
output variables are used and defined when appropriate. Skip this for
the _FOUND variable because FPHSA already does it. Skip this for the
_VERSION variable because that was recently added and never available
with the old name in a release of CMake.
Also detect the library version number. Provide results as variables
and as an imported target, LTTng::UST.
Signed-off-by: Philippe Proulx <eeppeliteloop@gmail.com>
When we check for a working compiler we print a message of the form:
Check for working <LANG> compiler: ...
At one time CMAKE_<LANG>_COMPILER was not well-defined for all
generators so we printed the generator name instead of the path to
the compiler. Nowadays we always know the compiler, so update the
message to print it unconditionally. This is more informative than
the generator name, especially when a toolset (cmake -T) is used.
Suggested-by: Gregor Jasny <gjasny@googlemail.com>
This is done the same way as for Eclipse: cmake tries to determine
the number of CPUs, and then adds the respective -jN to the make
invocations in the project file.
Alex
If CMAKE_INSTALL_RPATH is set and contains $ORIGIN then
dpkg-shlibdeps searches for the DEBIAN directory in order
to resolve $ORIGIN in the rpath to a directory. We need to
create the DEBIAN directory for this to work.
Since commit v3.5.0-rc1~47^2 (FindCUDA: Support special characters in
path, 2016-01-15) our add_custom_command calls use VERBATIM so that
CMake will automatically quote special characters correctly. However,
this breaks the special `$(VCInstallDir)` placeholder used with Visual
Studio generators. Since we do not support preservation of such
placeholders with VERBATIM (see issue #15001) we must fall back to not
using VERBATIM when the placeholder is used.
A better fix would be to stop using `$(VCInstallDir)` and use the value
of `CMAKE_${CUDA_C_OR_CXX}_COMPILER` instead, but that will require
additional semantic and documentation changes. For now simply fix the
regression with the above approach.
Reported-by: Stephen Sorley <Stephen.Sorley@jhuapl.edu>
We deprecated this module in commit v3.5.0-rc1~295^2
(CMakeForceCompiler: Deprecate this module and its macros, 2015-10-19)
in order to determine whether anyone still has use cases that require
it. Indeed we still need to provide a way to work with toolchains that
cannot link binaries without special flags. Remove the deprecation
warnings until we can provide an alternative to the module for this use
case.
We use `sh wx-config` to launch the `wx-config` tool so that it can run
even on Windows. Since it is always a shell script its output may use
POSIX paths even on Windows. Use `cygpath` to convert to Windows paths.
Check found libraries version to match user required version.
Protobuf compiler executable version is checked to be aligned with found
libraries, raising a warning message otherwise.
CPACK_* variables expect component name in upper case.
CPACK_RPM_* variables expected component name to be
in same case as component name.
This patch adds support for CPACK_RPM_* variables with
upper case component names to match the convention with
CPACK_* variables and also preserves same case component
names for back compatibility.
The specific cause is when e.g.
/Library/Frameworks/GStreamer.framework/Versions/1.0/lib/libgio-2.0.0.dylib
is detected by fixup_bundle. set_bundle_key_values() interprets this as
a framework, thus doing a string replace that creates an embedded_item
that is equal to the original path, i.e. it is not embedded.
Before this change, the variable PROTOBUF_LIBRARIES might get redundant
value for debug and optimized configurations, e.g.
`optimized;/usr/lib/libprotobuf.so;debug;/usr/lib/libprotobuf.so`.
Some projects may want to use the detailed mangling information
directly instead of using the FortranCInterface_HEADER function.
We already provide variables encoding the mangling information,
so just document them as available.
This commit introduces two new options to CTestCoverageCollectGCOV.
When GLOB is set we recursively search in the source & binary
directories for .gcda files. Otherwise the default behavior is to
parse TargetDirectories.txt for a list of locations to search.
When DELETE is set we remove any .gcda file found after it has
been used to generate the corresponding .gcov file. The .gcov
file is also removed after the result tarball has been created.
Together these two new features help support the use case of
computing coverage across subprojects.
81ecc726 FindCUDA: Added some additional comments about non-existent dependency files.
4b561b4c FindCUDA: Fix for when a non-existent dependency file is found.
Due to #4662 -isystem support was disabled for all GNU Compilers
on Apple platforms. But the change was probably a just work around
for a broken compiler on Tiger (see 10837#c27206). So we tighten
the condition to only kick in for GCC versions earlier than 4.
That should ensure sane behavior for Xcode 3.2 and later.
Previously if a non-existent dependency file is found we set the file to "" and
then do if(NOT IS_DIRECTORY "${file}"). Later we call get_filename_component
on the empty file which returns basically the current build directory.
Having a dependency on the current build directory is really annoying, because
anything that compiles into that directory will change the file stamp and
cause your files to rebuild every time you call make. :(
Since commit v3.5.0-rc1~32^2~1 (ExternalProject: Simplify `cmake
--build` configuration passing, 2016-01-19) we use the `$<CONFIG>`
generator expression to generate the `cmake --build . --config <config>`
value for the default BUILD_COMMAND instead of the CMAKE_CFG_INTDIR
placeholder value provided by multi-config generators. However, some
projects have been abusing the old implementation detail by setting
CMAKE_CFG_INTDIR themselves to get a specific configuration. Those
projects should be updated to set their own BUILD_COMMAND to get
non-default behavior. Meanwhile we can be compatible with their
existing releases by detecting when CMAKE_CFG_INTDIR is not a
generator-provided placeholder and using its value instead.
The dependency flags require recent versions of `iccarm` and `iccavr`.
The multi-rule dependency generated with `--dependencies=m` does not
work well with Ninja, so use `--dependencies=ns` instead.
Imported targets are now the recommended way of dealing with external
library dependencies. Add one for FindPNG and update documentation
accordingly. Also add a test case activated by CMake_TEST_FindPNG.