Refactoring in commit 77982de955 (CMakePackageConfigHelpers: only emit
arch check if needed, 2023-01-29, v3.26.0-rc1~16^2) did not correctly
preserve the check for empty `CMAKE_SIZEOF_VOID_P`. Fix it.
The difference between the version files generated before and after is:
-if(CMAKE_SIZEOF_VOID_P STREQUAL "" OR "8" STREQUAL "")
+if("${CMAKE_SIZEOF_VOID_P}" STREQUAL "" OR "8" STREQUAL "")
This restores the check generated before the above-mentioned commit.
Fixes: #24608
Revert commit 5b45a3d0ce (CheckCompilerFlag: Match the Clang "argument
unused" output for all languages, 2023-01-23, v3.26.0-rc1~38^2). It
broke existing projects that were silently tolerating unrelated unused
arguments in their checks for C and CXX. For example, using
`CFLAGS=-nostdinc` or `CXXFLAGS=-nostdinc++` causes those flags to be
used when driving the linker as well, and Clang warns they are unused in
that case.
Add a test case covering the now-restored behavior.
Fixes: #24591
Fix a regression from commit 073dd1bd81 (GoogleTest: Change format for
typed tests, 2022-02-07, v3.23.0-rc1~4^2) in the suite name detection.
Co-authored-by: Evgeniy Shcherbina <ixsci@pm.me>
Fixes: #24563
Revert commit 66b5d51f38 (ExternalProject: Install CMake projects
using 'cmake --install', 2022-09-08, v3.25.0-rc1~150^2).
It changed the ExternalProject install step command from:
cmake --build <dir> --target install --config <cfg>
to:
cmake --install <dir> --config <cfg>
The latter command no longer runs the external project build system
during the install step. We could consider using the commands:
cmake --build <dir> --target all --config <cfg>
cmake --install <dir> --config <cfg>
as the install step, but if `CMAKE_SKIP_INSTALL_ALL_DEPENDENCY` is
used in the external project, that can change semantics too.
Revert the original change pending further investigation on other ways
to support its motivating use case. Add a test covering the
previously-regressed use case.
Fixes: #24567
Issue: #23946
Because -pthread doesn't begin with -l, it doesn't show up in the
pkgconfig LIBRARIES variable, causing FindOpenSSL to not add the
Threads::Threads dependency. Explicitly search LDFLAGS_EXTRA
for -pthread and add the dependency if it's found.
Fixes: #24532
Move the permissions modification added by commit 88fed668b1 (Make
bundle items writable before fixup, 2010-09-07, v2.8.3~129^2) inside the
condition added by commit 45ed314bff (BundleUtilities: do not run
install_name_tool on scripts, 2020-07-27, v3.19.0-rc1~404^2).
There is no reason to add write permissions to a file that we are
not going to modify.
Fixes: #24424
Since commit 1b7804edd0 (FindOpenSSL: use extra dependencies from
pkg-config as well, 2022-12-05, v3.26.0-rc1~227^2) we conditionally find
Threads but unconditionally depend on it. Make the conditions
consistent.
Fixes: #24505
LLVM Flang's preprocessor output is always valid fixed form Fortran with
a fixed line length of 72. If an application tries to use fixed form
with a different line length, this will fail in the
post-preprocessing compilation step as they will pass a flag asking for
a different fixed line length.
To fix this we can pass -ffixed-line-length-72 to the post-preprocess
compilation stage. This will be ignored when the input is free form
Fortran, and force to a 72 line length when the input is fixed.
This variable has been provided since commit 746906242d (Android: Detect
NDK version number, 2021-02-26, v3.20.0-rc3~1^2~3) when using CMake's
NDK support or the modern NDK toolchain file. Since commit 005e2cdfb0
(Android: Do not use gold for ndk >= r22, 2021-02-26, v3.20.0-rc3~1^2)
we need the value in our compiler/platform information files, so provide
it when using the NDK legacy toolchain file too.
Revert commit 1c86e397fe (Android/Clang: Tolerate undefined
CMAKE_ANDROID_NDK_VERSION, 2022-09-16, v3.25.0-rc1~118^2) since the
variable should now always be defined.
Issue: #21772Fixes: #24386
Update the change from commit 2a94c762ed (FindCUDAToolkit: Add support
for CUDA::nvrtc_static, 2023-01-20, v3.26.0-rc1~55^2). The lib is named
`libnvrtc-builtins_static.a`, not `libnvrtc_builtins_static.a`.
8f82e755f3 Ninja: Fix detection of MSVC showIncludes prefix in Italian
d6e7e4d4a1 Tests: Extend RunCMake.Ninja ShowIncludes cases to cover more languages
9596305c0b Tests: Generalize RunCMake.Ninja ShowIncludes test infrastructure
c6dd4fa21d Tests: Extend RunCMake.Ninja ShowIncludes case with sample path
a9d97492fd Ninja: Record showIncludes detection in configure log
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !8129
Before this patch, the write_basic_package_version_file() function of
the CMakePackageConfigHelpers module always emitted an architecture
check, even if the ARCH_INDEPENDENT option was specified. While this is
not an issue when configuring builds, as the check is skipped, this can
create issues when the "arch independent" version files are installed in
the datadir (e.g. /usr/share) in a MultiArch environment like Debian,
where different architecture packages of the same libraries can be
coinstalled; as the amd64 version of a given library contains "8 * 8"
in the file, while the i386 one contains "4 * 8", there's a conflict, as
files in /usr/share are expected to be identical across architectures.
This patch fixes this issue by only emitting the architecture check code
if needed; when ARCH_INDEPENDENT is specified, no code is written at
all.
Here's a diff between the version files generated before and after this
patch:
diff -u old/indep.cmake new/indep.cmake
--- old/indep.cmake 2023-01-29 13:43:04.840671117 +0100
+++ new/indep.cmake 2023-01-29 13:57:28.475191551 +0100
@@ -52,19 +52,3 @@
endif()
-# if the installed project requested no architecture check, don't perform the check
-if("TRUE")
- return()
-endif()
-
-# if the installed or the using project don't have CMAKE_SIZEOF_VOID_P set, ignore it:
-if("${CMAKE_SIZEOF_VOID_P}" STREQUAL "" OR "8" STREQUAL "")
- return()
-endif()
-
-# check that the installed version has the same 32/64bit-ness as the one which is currently searching:
-if(NOT CMAKE_SIZEOF_VOID_P STREQUAL "8")
- math(EXPR installedBits "8 * 8")
- set(PACKAGE_VERSION "${PACKAGE_VERSION} (${installedBits}bit)")
- set(PACKAGE_VERSION_UNSUITABLE TRUE)
-endif()
diff -u old/no-indep.cmake new/no-indep.cmake
--- old/no-indep.cmake 2023-01-29 13:42:05.010710508 +0100
+++ new/no-indep.cmake 2023-01-29 13:57:40.914237219 +0100
@@ -52,13 +52,8 @@
endif()
-# if the installed project requested no architecture check, don't perform the check
-if("FALSE")
- return()
-endif()
-
# if the installed or the using project don't have CMAKE_SIZEOF_VOID_P set, ignore it:
-if("${CMAKE_SIZEOF_VOID_P}" STREQUAL "" OR "8" STREQUAL "")
+if(CMAKE_SIZEOF_VOID_P STREQUAL "" OR "8" STREQUAL "")
return()
endif()
Fixes: #24375
- With this change we can use e.g. ImageMagick::Magick++ directly
in targt_link_libraries.
- This change also adds CFLAGS which was missing before.
- Also adds example on how to use the targets.
This add correct Open Watcom support for 16-bit Windows 3.x.
It replace existing strange mixture with WIN32 stuff which implement 16-bit Windows target partially as part of WIN32 stuff.
Now pre-defined OS ID Windows3x is used instead of confusing WIN32.
It support properly 16-bit and 32-bit application for 16-bit Windows host.
32-bit applications are build with OW WIN386 extender.
It is used similar as for other platforms by set CMAKE_SYSTEM_NAME=Windows3x and CMAKE_SYSTEM_PROCESSOR=I86 for 16-bit application or CMAKE_SYSTEM_PROCESSOR=x86 for 32-bit WIN386 extender application running on 16-bit Windows 3.x.
CMAKE_SYSTEM_NAME=Windows is used only for WIN32 applications.
Improve detection of missing compiler flags: move "argument unused
during compilation: .*" pattern from language-specific branches into
the common list.
Add setup of system include directories to language related macro to remove extra lines for C and CXX.
System include directories are always same for both languages (they are defined per platform).
The example shown in the module documentation for
CMakePackageConfigHelpers was showing its age, by:
1. Hardcoding directory prefixes (`etc/`, `lib/`, etc.) instead of
using GNUInstallDirs
2. Handwaving `set()` commands (incorrectly!) as:
`set(VAR dir/ ... CACHE )`
which should, at the very least, be:
`set(VAR dir/ CACHE ... )`
3. Installing CMake configuration files to `lib/Foo/cmake/`, when
current practice favors `${CMAKE_INSTALL_LIBDIR}/cmake/Foo/`
This updated example addresses all of those issues.
If `OpenMP_<lang>_INCLUDE_DIR` is defined, add it to the list of include
directories before checking flags. Previously, this variable was
ignored for all compilers but AppleClang, despite the documentation
mentioning it as one of the possible inputs.
Fixes: #24260
before this change, pkg_check_modules(.. IMPORTED_TARGET GLOBAL)
is used for creating an imported target from which another imported
interface library named OpenSP::OpenSP is created. but pkg-config
does not account for all of CMake's other search behavior controls,
such as CMAKE_FIND_ROOT_PATH. neither does it export the full path
with OpenSP_LIBRARY.
after this change, the paths found by pkg-config are only used
as hints for the find_*() commands. and some cleanup are included:
* be QUIET when calling find_package(PkgConfig ..) and
pkg_check_modules(..) as they are distracting from user's point of
view. what matters is the output of find_package_handle_standard_args()
* parse the version and check for the existance of symbol as long as
header path is found. because they only use header files.
* define OpenSP_LIBRARY as long as it exists. this just follows
the convention. as OpenSP_FOUND implies a valid OpenSP_LIBRARY.
* wrap and intent multi-line command calls for better readability
* check OpenSP_FOUND before adding OpenSP::OpenSP, it's more
idiomatic.
Fixes: #24313
Signed-off-by: Kefu Chai <tchaikov@gmail.com>
This patch adds support for tracking the swiftmodules for executables
exporting symbols.
This fixes a bug in the earlier implementation around emitting the
swiftmodule. Previously, the code would use
`CMAKE_EXE_EXPORTS_Swift_FLAG` to inject the `-emit-module`, and module
path information into the `CMAKE_Swift_LINK_EXECUTABLE` rule. Because
Swift skips the build step and only runs during the link phase, these
flags were injected in `cmNinjaNormalTargetGenerator::ComputeLinkCmd`
instead of `cmLocalGenerator::GetTargetFlags` where it is done normally.
Unfortunately, injecting in `ComputeLinkCmd` didn't do anything because
we have a `linkCmd` so `ComputeLinkCmd` exits early, before the
EXE_EXPORT flags get added to the link command.
Instead of playing with that flag, CMake checks
`CMAKE_Swift_LINK_EXECUTABLE_WITH_EXPORTS` and uses that as the link
rule if it exists and falls back on `CMAKE_Swift_LINK_EXECUTABLE`. I've
defined that variable in terms of `CMAKE_Swift_LINK_EXECUTABLE` with the
necessary additional flags for emitting the swift module instead. This
has the same end effect as the desired behavior, but simplifies things a
bit.
Since we're generating the swiftmodule for executables with exports,
I've also updated the dependency graph to include the swiftmodule as an
output in the build dependency graph if the executable has exports.
Tests updated:
- RunCMake/NoWorkToDo:
Ensure that the build graph does not result in unnecessary rebuilds
with exporting executables.
- SwiftOnly:
Ensure that we can consume functions defined in the executable by a
library getting linked into said executable.
Brad King