On modern macOS, compiler wrappers like `/usr/bin/cc` automatically
choose a SDK to pass via `-isysroot` to an underlying compiler from
Xcode or the CommandLineTools. Other toolchains like Homebrew's `gcc-*`
come with a default SDK too. Therefore, when targeting macOS, we no
longer need to choose any SDK or pass an `-isysroot` flag by default.
Update initialization of `CMAKE_OSX_SYSROOT` to be empty by default when
targeting macOS.
Fixes: #19180
This field was added by commit b3e9fb67bb (file-api: support exporting
file set information, 2022-11-03, v3.26.0-rc1~389^2) but the relative
path convention used elsewhere was accidentally left out.
Fixes: #25422
Since commit dc5fc898f6 (Xcode: Set object file locations using
TARGET_TEMP_DIR, 2022-09-29, v3.25.0-rc1~64^2~1), `xcodebuild clean`
does not remove the object files in our explicit `TARGET_TEMP_DIR`
because it is not under the `SYMROOT`. Put it there.
Fixes: #24096
The backtraces for some of the commands incorrectly come from the
target's directory instead of the caller's directory. Expect the
incorrect backtraces in test results for now.
Increase the minimum required version of CMake in the test case
to enable policy CMP0079 to support cross-directory calls.
Issue: #23873
This needs to be known so that C++ module properties for the install can
be staged beside the other files.
Always perform the MD5 transformation (in non-bootstrap builds) so that
the path can be computed prior to generation (where it used the longest
configuration name to detect too-long paths). Update tests to expect the
always-present MD5 value. Note that this improves robustness of the test
suite as testing in a too-long path may have triggered the MD5
conversion anyways.
In a per-component installation the generated installation scripts
are invoked once for each component.
Per default custom installation script code added by install(CODE|SCRIPT)
only runs for one specific component in this context.
The new ALL_COMPONENTS option allows custom script code to be run once
for each component being installed.
Previously we would synthesize <TARGET_NAME>-NOTFOUND as the location. This
would then end up on the link line and cause build failures.
Policy CMP0110 is added to control this behaviour.
Fixes#19080, #19943.
INTERFACE libraries with SOURCES now appear in the generated
buildsystem, so include them in the codemodel output too.
We do not need to bump the `codemodel-v2` object kind minor
version because that was already done in post-3.18 development
by commit 7d6861f367 (fileapi: Extend codemodel targets with
language standard, 2020-06-18).
Fixes: #18608
When an object library is used via `target_link_libraries`, any targets
listed in the object library's `INTERFACE_LINK_LIBRARIES` closure should
become direct dependencies of the consuming target. However, these were
accidentally left out by `cmComputeTargetDepends::CollectTargetDepends`
because object libraries are encountered through external object sources
first and then added to the `emitted` set which blocks them from being
processed as link dependencies.
This was not noticed by the test case in commit bab24e782c
(target_link_libraries: Propagate dependencies of object libraries,
2018-12-10, v3.14.0-rc1~260^2) because the relevant dependency appears
transitively through the object library target itself.
Re-order the logic to process link dependencies first, and then external
object sources. That way object libraries used via
`target_link_libraries` will be treated as such by dependency analysis.
This also adds missing backtrace information for object libraries used
via `target_link_libraries`. The missing information was mentioned in a
FIXME comment in the RunCMake.FileAPI test added by commit ea0a060168
(fileapi: Add test for codemodel v2, 2018-11-09, v3.14.0-rc1~257^2~7).
That comment itself was dropped by commit a0de350e2f (FileAPI test:
Break gen_check_targets() into JSON files, 2020-02-07), but we can now
update the corresponding location in the `.json` files to have the
now-expected backtrace information.
Fixes: #20421
