The "all" target in each directory is supposed to have targets from that
directory even if the directory itself is marked `EXCLUDE_FROM_ALL` in
its parent. This was broken by commit dc6888573d (Pass EXCLUDE_FROM_ALL
from directory to targets, 2019-01-15, v3.14.0-rc1~83^2) which made the
participation of a target in "all" independent of context. Revert much
of the logic change from that commit to restore the old behavior. Then
re-implement the behavior intended by the commit to keep its test
working. Extend the test to cover the old behavior too.
Fixes: #19753
Update parsing logic from commit 5990ecb741 (Compute implicit include
directories from compiler output, 2018-12-07, v3.14.0-rc1~108^2)
to handle the quoting behavior caused by the `-hlist=` flag.
The value of `CMAKE_VS_PLATFORM_NAME` is computed by Visual Studio
generators based on `CMAKE_GENERATOR_PLATFORM` or some default.
Prior to the VS 2019 generator, the default was always `Win32`.
However, for the `Visual Studio 16 2019` generator, the default is
based on the host platform.
Store the default in a new `CMAKE_VS_PLATFORM_NAME_DEFAULT` variable for
use by project code. This is particularly useful in toolchain files
because they are allowed to set `CMAKE_GENERATOR_PLATFORM` and so
`CMAKE_VS_PLATFORM_NAME` is not yet known. Of course the toolchain file
author knows whether it will set `CMAKE_GENERATOR_PLATFORM`, and if not
then `CMAKE_VS_PLATFORM_NAME_DEFAULT` provides the platform name that
will be used.
Fixes: #19177
In some cases GCC reports *relative* implicit include directories. They
are computed adaptively with respect to the current working directory
such that the effective implicit include directory is an unchanging
absolute path. Teach our implicit include directory extraction to
recognize such paths and normalize them.
Fixes: #19133
The implicit include directory extraction added by commit 5990ecb741
(Compute implicit include directories from compiler output, 2018-12-07,
v3.14.0-rc1~108^2) leaves paths like `/usr/lib/../include` unchanged.
Fix the logic to canonicalize such paths (e.g. to `/usr/include`)
as we do for implicit link directories already. This is important
to ensure the set of implicit directories is represented in the same
form as the include directories that will be compared to them.
Issue: #19095
Any `include_directories()` calls in toolchain files are used during our
ABI detection step even though it does not include any system headers.
Since commit 5990ecb741 (Compute implicit include directories from
compiler output, 2018-12-07, v3.14.0-rc1~108^2), that check is also used
to detect implicit include directories. Any `include_directories()` in
a toolchain file are detected as implicit and later excluded from
explicit specification on compiler command lines, thus breaking the
purpose of the calls in the first place.
Fix the implicit include directory detection step to avoid using paths
from `include_directories()` calls in the toolchain file.
Fixes: #19079
Since commit e89ad0f94e (install: Allow installing targets created in
another directory, 2018-06-18, v3.13.0-rc1~407^2) the `install(TARGETS)`
command may find a global-scoped target outside the calling directory.
Ignore an `IMPORTED GLOBAL` target if it is found in this way. Imported
targets cannot be installed, and trying to do so violates internal
invariants.
Fixes: #19022
Since commit 11da882a12 (Apple: Introduce separate system name for iOS,
tvOS, and watchOS, 2018-01-15, v3.14.0-rc1~14^2~1) we support setting
`CMAKE_SYSTEM_NAME` to `iOS`. Existing iOS toolchain files already
set `IOS` as a short-hand variable, so do the same here.
The change in commit a5dd159990 (Tests: Fix RunCMake.try_compile
CxxStandard case on MSVC, 2017-09-22, v3.10.0-rc1~63^2) introduced use
of `if()` conditions not supported until CMake 3.7, so while it was
intended to restore support for CMake versions prior to 3.9, it actually
only did so for 3.7 and 3.8. Backport the logic to work with CMake 3.1.
Fixes: #18987
The original warning pre-dates support for install components.
There are now legitimate scenarios where an install(TARGETS)
command may list a target that is excluded from all, e.g.
hierarchical projects that will never install the component such a
target belongs to.
Fixes: #18938
e8ee8cab97 Xcode: Completely disable code signing for compiler id detection
11da882a12 Apple: Introduce separate system name for iOS, tvOS, and watchOS
36cf44a7a3 Tests: Isolate RunCMake.XcodeProject per-device cases from host arch
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !2392
- Remove code signing requirements for non-macOS
- Do not set deployment target for non-macOS
- Build static library for compiler feature detection for non-macOS
- Use framework to run CompilerId tests for watchOS
- Port tests to new SDK handling
- Add new Apple cross-compiling section to toolchain documentation
Closes: #17870
91d98542d2 Merge branch 'autogen-qt-version-from-dirprops-release' into autogen-qt-version-from-dirprops-master
062d21c36a Autogen: Read the Qt version from directory properties as well
17ac7c4024 Tests: add cases for providing Qt5Core_VERSION manually
2df6d69014 AutoGen: query Qt5 version from directory properties
b598dfb65e Tests: add cases for providing Qt5Core_VERSION manually
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !2883
The `FRAMEWORK` target property affects the way the `install()` command
treats the target and so should be set first. Our implementation
assumed that this was always the case and led to an assertion failure.
Prior to CMake 3.12 this was visible only when using an explicit
`LIBRARY ... NAMELINK_ONLY` option, but commit 0212d7c762 (install: add
NAMELINK_COMPONENT argument, 2018-04-18, v3.12.0-rc1~139^2~3) made
it possible with a simple `LIBRARY DESTINATION`.
Fully supporting out-of-order specification will require non-trivial
refactoring to defer install generator creation to generate time.
For now simply restore the old behavior of installing the framework
to the library destination and warn about the case.
Fixes: #18848
Add implicit include parser for PGI compiler ID. PGI verbose output
for CXX differs from C and Fortran, so CXX is broken out into its own
case. The C and Fortran compilers take "-YI,path" and "-Mnostdinc" to
change or disable the builtin include path. The last arg on the command
line appears to override previous args (e.g. "-YI,path1 -YI,path2" will
set the path to "path2" ... the previous "-YI,path1" gets undone).
The CXX compiler verbose output reports with "-I" rather than "-stdinc"
for the built in path. In addition with CXX "-Mnostdinc" does not
completely zero the include path (e.g. "#include <stdio.h>" still works
with "-Mnostdinc"... "-I/usr/include" still shows up in the verbose output).
Minor adjustments to get the SunPro parser to handle Fortran as well.
Fixes for Cray compiler support (Modules/Compiler/Cray-{C,CXX}.cmake):
The *_COMPILE_OPTION flags contain options like "-h c99,gnu" ...
these options need to be in double quotes (they are currently not).
Otherwise, cmake treats them as lists and tries to run the compiler
with "-h;c99,gnu" and fails when it is "Detecting C compile features"...
Also, the Cray-CXX.cmake contains "__compiler_cray(C)" instead of
"__compiler_cray(CXX)" -- this error prevents the correct VERBOSE
flags for CXX from being defined which prevents the implicit include
parser from running.
Add additional test cases for PGI and SunPro Fortran to the
Tests/RunCMake/ParseImplicitIncludeInfo area.
Currently cmake is only setting CMAKE_Fortran_IMPLICIT_INCLUDE_DIRECTORIES
the first time it runs (the implicit includes parser is only run when
the cache variables are initially populated). It is left unset
during additional runs. This behavior is both confusing and
inconsistent.
Modify CMakeFortranCompiler.cmake.in to preserve the value of
CMAKE_Fortran_IMPLICIT_INCLUDE_DIRECTORIES so that cmake's Fortran
behavior matches C and CXX (i.e. CMAKE_Fortran_IMPLICIT_INCLUDE_DIRECTORIES
retains its value across runs). The implicit includes variable may
not be as useful for Fortran as it is for C and CXX, but we do want
cmake to be consistent in how it is handled.
Adjust CMakeParseImplicitIncludeInfo to parse verbose Fortran output
for Cray and XL compiler IDs (GNU and clang were already ok since
they use the same verbose output format as they do with C/CXX).
Allow Tests/RunCMake/ParseImplicitIncludeInfo/data/CMakeLists.txt
to generate test inputs for Fortran too, and add some sample Fortran
input files for the Tests/RunCMake/ParseImplicitIncludeInfo test.
Add parser code for compiler IDs SunPro and XL. For SunPro, /usr/include
appears to be hardwired in and it does not have a -nostdinc-type flag.
For XL, the "xlc" and "xlC" commands determine if you are C or C++
based on input filename. So compiling x.c with "xlC" reverts to C
and compiling x.cc with "xlc" upgrades to CXX. System include paths
are specified by:
-qc_stdinc=[path] and -qgcc_c_stdinc=[path] (for C)
-qcpp_stdinc=[path] and -qgcc_cpp_stdinc=[path] (for CXX)
If you specify "-qnostdinc" then the above flags are ignored.
Specifying an empty path reverts the value to the default
path (e.g. "-qgcc_c_stdinc=" reverts that to the default).
Adjust the code to handle parsing the case where the user specifies
additional compiler include flags via CMAKE_C_FLAGS/CMAKE_CXX_FLAGS.
For example: "-DCMAKE_C_FLAGS=-nostdinc" ... note that gcc and clang
output slightly differ when -nostdinc is specified (clang skips
printing 'search starts here:' with the angle brackets, gcc always
prints it).
Enable sunos-{C,CXX}-SunPro test cases and add XL test cases.
Also add a few -nostdinc style tests.
