On some Xcode versions, `xcodebuild` may warn:
... xcodebuild[...] Requested but did not find extension point with
identifier ...
Teach RunCMake to drop such incidental lines before matching against
expected output.
Apply the change from commit db35e3cfd6 (VS: Fix support for '/guard:cf'
linker flag for v142, 2019-01-24, v3.14.0-rc1~74^2~2) to the v143 flag
table.
The entry for `LinkControlFlowGuard` in `v143_Link.json` does not work
when used in a `.vcxproj` file. Drop our link flag table entries for
this toolset so that the flag will be passed via `AdditionalOptions`.
Also add a test case.
When `Visual Studio` and `Xcode` generators are used, directory for depfile
is not implicitely created by CMake when OUTFILE_DIR option is used.
Fixes: #22932
Revert commit 798c1c3192 (GNUInstallDirs: Comply with Debian Policy on
LIBEXECDIR, 2020-10-08, v3.19.0-rc1~11^2).
The Debian Policy builds upon FHS 3.0 and permits installing to
`/usr/libexec`. While Policy does grant an additional exception for
applications to use a single subdirectory of `/usr/lib/<triplet>`, this
is not meant to replace `/usr/libexec` as valid target.
Fixes: #22731
On Windows, imported shared libraries which only have an
`IMPORTED_IMPLIB` set fail if they are depended upon by a target using
automoc. Add a test for the behavior of depending upon all imported
target types from an automoc-using target.
Since commit bd844387df (ROCMClang: Add the ROCm toolkit derived clang
compiler to CMake, 2020-08-28, v3.21.0-rc1~66^2~6) and commit ff0d2858e1
(HIP: Extract clang compiler details from hipcc, 2020-10-21,
v3.21.0-rc1~66^2~5), the separate `ROCMClang` compiler id for `hipcc`
has caused a few problems:
* The compiler id changed from behavior of CMake 3.20 and below,
breaking projects that already built with `hipcc` treated as `Clang`.
* The implementation of `target_compile_features` was incomplete for
the `ROCMClang` identity.
* Only `hipcc` was identified as `ROCMClang`, so after it is unwrapped
to the underlying `clang++`, future runs of new CMake versions on
an existing build tree would not repeat this.
* Clang should be usable as a HIP compiler without the `hipcc` wrapper.
Remove the `ROMClang` compiler identity, and revise HIP language support
to work directly with a Clang compiler.
Reject direct `hipcc` usage as a HIP compiler. For now it cannot be
supported because it interferes with flags CMake needs to pass to Clang.
Fixes: #22536, #22460, #22593
Revert commit e5ec0e52f4 (AUTOUIC: Fix generating of dependency rules
for UI header files, 2021-07-22, v3.21.1~8^2) because it caused
regressions. For example, changing one C++ source can now cause many
others to rebuild. Revert the change pending further investigation.
Fixes: #22531
Issue: #16776
In commit d7159f9591, (Ninja: Fix LINK_WHAT_YOU_USE link flag placement,
2021-06-24), we fixed usage of -Wl,--no-as-needed to override
-Wl,--as-needed from LDFLAGS. However, that change also passes
-Wl,--no-as-needed to `ar`, which is incorrect flag for `ar`.
Let's check the target type before passing down the flags.
Fixes: #22487
Since commit 4391913133 (Add INTERFACE libraries to generated
buildsystem if they have SOURCES, 2020-07-20, v3.19.0-rc1~346^2~1), the
VS generator may process INTERFACE libraries. Avoid code paths in the
generator that process include directories because they should not be
used by INTERFACE libraries since they do not compile anything.
Fixes: #22494
Seems the relative paths were wrong basically all around such that only
compiling files in the top-level directory would work. I've modified
CudaOnly.SeparateCompilation to cover this.
Fixes#22482.
44ad3f0b added multi-options to CTest, but didn't reset them,
causing ctest_test() to fail if it was run multiple times with
different label arguments. Reset the multi-options.
Fixes: #22485
06e6981 added support for optional binaryDir and generator, but
use the dependent preset's file version instead of the current
preset's file version. Check presets with their own file version
instead of their dependent's file version.
Fixes: #22428
The fix from commit eafe740ead (FindXCTest: Fix output directory for
test bundle with new build system, 2021-02-09, v3.19.5~5^2) is not
necessary with Xcode 12.5, which seems to have changed/fixed the
behaviour again.
Fixes: #22462
In the same spirit of commit ce1cadd35a, (Tests/ConfigSources: fix for
non main stream CMAKE_BUILD_TYPE, 2021-04-01).
Allow distributions' check to pass seemlessly.
Issue: #22470
cb777dd81e Ninja Multi-Config: Restore TARGET_OBJECTS support in cross-configs
83c8272280 cmGeneratorExpressionNode: Factor out local variable for global generator
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !6371
We could not rely on .ui files when generating the ninja rules
for the generated UI header files. .ui files might be added to the
target sources but never processed by AUTOUIC afterward, since UI
header files are never included in a source code. Instead of adding
dependency rules based on the .ui files, this approach scans
non-generated source files for includes of the UI header files,
as AUTOUIC does. This gives the consistent set of UI header files
at configure time, that could be used to generate byproducts rules
for the AUTOUIC. Also, the path to the generated UI header file depends
not on the .ui file location but on the include line is used in source
files.
Fixes: #16776
Since commit 2ae72ef74b (Xcode: Enable multi-arch TARGET_OBJECTS genex
in [INTERFACE_]LINK_LIBRARIES, 2021-05-26, v3.21.0-rc1~126^2) the
TARGET_OBJECTS genex, when referenced for linking, is now evaluated with
EvaluateForBuildsystem enabled. This causes the object file paths to be
computed with a buildsystem-specific placeholder for the configuration.
This is normally fine because the placeholder will be evaluated by the
native buildsystem tool using the proper configuration. However, the
Ninja Multi-Config generator's `${CONFIGURATION}` placeholder may not
have the correct value for cross-config object files. Switch back to
using the per-config location of each object file for this generator.
Fixes: #22436
Extend the `SwiftOnly` test to cover the fix in commit 3ddd7f3576
(enable_language: Fix test for working compiler with CMP0126 NEW
behavior, 2021-07-15), as that commit did for the `CSharpOnly` test.
Fixes: #22451
Since commit 0a0a0f8a74 (cmMessenger: Color messages to terminal by
type, 2021-05-18, v3.21.0-rc1~146^2) the message output no longer goes
through our custom streambuf on Windows that converts output encoding.
This can cause messages to be printed with the wrong encoding in a
Windows Console. It also causes messages to have a mix of LF and CRLF
newlines because `stderr` converts LF to CRLF but our custom streambuf
does not.
Revert to using just `cerr` for messages on Windows. Another approach
will be needed to achieve color output on Windows later.
Fixes: #22444
Revert commit 0665d9092e (CMakeDependentOption: Allow parentheses in the
depends string, 2021-06-13, v3.21.0-rc1~32^2). It broke existing use
cases with parentheses in regular expressions. Also add a test for this.
Fixes: #22447
When resolving binaryDir into a full path from a relative path, we
forgot to check if binaryDir is altogether empty, causing empty
binaryDir's to resolve to the source directory. Fix this.
Fixes: #22434
Update the logic that converts a `try_compile` result from a cache
entry to a normal variable to tolerate an existing normal variable
under CMP0126 NEW behavior. Otherwise the `try_compile` result
is ignored because CMake uses the false value of the normal variable,
and CMake incorrectly reports that the compiler does not work.
This went unnoticed for some languages (e.g. C and CXX) because the
check for a working compiler is skipped if ABI detection works.
It does affect other languages (e.g. CSharp).
Fixes: #22423
Documentation added by
* commit 4f4f2028b8 (Help: Add documentation for buildPresets and
testPresets, 2021-01-13, v3.20.0-rc1~51^2~7)
* commit 676ecf0d37 (cmake-presets: Add build and test presets,
2020-12-14, v3.20.0-rc1~51^2~6)
used square brackets in the `cmake --build` signature to indicate
non-optional alternatives, which is not a typical convention.
A common convention is to use parentheses instead, but in this
case it is probably clearer to list the two signatures separately.
Fixes: #22413
Since commit 84036d30d4 (IntelLLVM: Fix C/C++ standard level flags on
Windows, 2021-07-07, v3.21.0-rc3~8^2~1) we activate C/C++ standard level
logic for IntelLLVM when targeting the MSVC ABI. Update the
`RunCMake.try_compile` test to be aware of this even when CMake is
itself configured by an older CMake that does not know this.
