GNU ld 2.39 added warnings of the form
ld: warning: %pB has a LOAD segment with RWX permissions
They appear when linking test binaries on `hppa` and `sparc`.
Debian-Issue: https://bugs.debian.org/1098956
Debian-Issue: https://bugs.debian.org/1098970
Fix logic from commit b9ee79b8a1 (GenEx: Add support for custom
transitive compile properties, 2024-05-09, v3.30.0-rc1~82^2~1) to more
precisely know when we are computing the link dependency graph.
Issue: #20416
Issue: #26709
Revise commit fc7aa3cd69 (tests: Preserve empty arguments in test
command lines, 2024-09-30, v3.31.0-rc1~33^2) to delay variable expansion
until code evaluation. Otherwise we may parse the working directory
path as CMake language code.
Account for changes to upstream ninja:
* commit `5d93f2da28` (Add exit code to the failed target, 2024-12-03)
* commit `311bf93416` (Propagate jobs' exit codes to the ninja's exit code,
2024-12-03)
In commit 9c0491a3e4 (cmDyndepCollation: write out scanned source
information too, 2024-03-25) via !9708, the `sf_map` gained a new job of
also being used to track non-file set sources which could import
modules. This was implemented by removing processed `FILE_SET TYPE
CXX_MODULES` sources from the map and working with the sources
remaining. When a `FILE_SET TYPE CXX_MODULES` source appeared multiple
times, this would then erroneously complain that it "was not scheduled
for compilation". Use a set of source paths to track sources that have
already been processed. If duplicates are found, trigger an author
warning and skip the duplicate file.
Fixes: #26549
This was added in commit 98d0f918ba (LFortran: Add support for this
compiler, 2024-01-25, v3.31.0-rc1~303^2~2) because it is needed for
cases covered by CMake's Fortran tests. However, it does not work
with Fortran modules and breaks lfortran's own `examples/project1`.
Move the flag to the test cases that need it, just as the original
commit did with `--implicit-interface`.
Fixes: #26597
Co-authored-by: Brad King <brad.king@kitware.com>
Revert commit 4a11772618 (GenEx: Limit TARGET_PROPERTY transitive
closure optimization to subgraphs, 2024-05-31, v3.31.0-rc1~114^2).
The change caused substantial performance regressions in some
existing use cases. Revert it pending further investigation.
Issue: #25728Fixes: #26457
854eba0c53 target_sources: Improve error message for CXX_MODULES on INTERFACE libraries
cd179e7560 Tests/RunCMake/CXXModules: Rename FileSetModules cases to be more specific
Acked-by: Kitware Robot <kwrobot@kitware.com>
Tested-by: buildbot <buildbot@kitware.com>
Merge-request: !10090
We support non-compiled `SOURCES` on `INTERFACE` libraries, and also
support `CXX_MODULES` on *imported* `INTERFACE` libraries (via synthetic
targets that compile module interface units). However, we do not
support `CXX_MODULES` on non-imported `INTERFACE` libraries because
there is no place to hold module interface unit's object files for their
module initializers. Previously this was not explicitly rejected, and
so was diagnosed only by "CMake Internal Error" messages due to
assumption violations in the implementation.
Fixes: #26524
Co-authored-by: Ben Boeckel <ben.boeckel@kitware.com>
In commit 197cb419d1 (add_custom_command: Add CODEGEN support,
2024-05-27, v3.31.0-rc1~394^2) we accidentally left out the global
`codegen` target's dependencies on the per-directory `codegen` targets.
Add them for parity with the `all` target.
Fixes: #26517
Since commit 474eafe28c (clang-cl: Add support for C++23, 2024-09-13,
v3.31.0-rc1~97^2) we use a Clang-specific flag to enable C++23 since
`clang-cl` has no `-std:c++23` flag, and `-std:c++latest` may enable an
even newer version of C++. However, in `.vcxproj` files there is no way
to express a target-wide `-clang:-std=c++23` flag for only C++ sources
when the target also has C sources. Add a special case to map back to
`-std:c++latest` for targets with C++23 and C together.
Fixes: #26508
Since commit e3895f4a8b (Linking: Preserve nested LINKER: prefixes as
written, 2024-09-17, v3.31.0-rc1~60^2) we may increment an iterator past
the end. Revise logic to avoid that.
Fixes: #26499
When `ui_` include directives have different paths than `CMakeLists.txt`,
the dependency graph is not generated correctly. This is the root cause
of #16776. However, when #26135 was fixed by commit 5363bebc1e (Autogen:
Fix compilation of unchanged source files, 2024-07-16, v3.31.0-rc1~328^2),
it made #16776 worse: the build is always dirty. Revert the fix for now.
Fixes: #26485
Issue: #26135
Issue: #25436
In commit 5363bebc1e (Autogen: Fix compilation of unchanged source files,
2024-07-16, v3.31.0-rc1~328^2) we relied on Ninja Multi-Config dependency graph
optimizations from commit 7c39dabdbc (Autogen: AUTO*_EXECUTABLE: add support
for per-config values, 2023-10-18, v3.29.0-rc1~105^2~1). However, those graph
optimizations are conditional on versions of Qt that enable
[`AUTOGEN_BETTER_GRAPH_MULTI_CONFIG`](https://codereview.qt-project.org/c/qt/qtbase/+/513648).
`UseBetterGraph` should be checked to add ui files to `timestampByproducts`.
Fixes: #26475
Teach `export()` and `install(EXPORT)` to record this setting on the
imported targets they generate. Make this property's non-presence
authoritative on imported targets since they represent what was built
elsewhere.
There is an inadvertent inconsistency in the notice for the CPS export
experimental feature as compared to other experimental feature notices.
Change it to be consistent.
Since commit 84db8506ff (Deprecate compatibility with CMake versions older
than 3.10, 2024-10-03, v3.31.0-rc1~23^2) the `cmake_minimum_required` and
`cmake_policy` commands warn if the project does not enable policies as
of 3.10 or higher. Clarify the advice given in the warning message about
how to update the version specification, particularly without actually
requiring a newer minimum version of CMake.
There's no need to check if flushing is needed after every command
is added to the variable holding the script content. It is enough to only
check once per test name. This simplifies the flushing logic, removing
the need for a separate flush_script() command. Previously, we were
not clearing the flushed script contents in all cases, but this is now
rigorously enforced at the one location where flushing is performed.
Also simplify the flushing of the list of test names, since that too doesn't
need a separate command. It is simpler and safer to handle that
directly inline where the one call to flush_tests_buffer() was
previously being made.
Fixes: #26431
When a single-value keyword is repeated, and the first instance is
missing a value, it prevents the value from the second instance from
being stored in a variable. This was a regression introduced by
commit ceeea4e511 (cmake_parse_arguments: Set variable if empty string
given after keyword, 2024-08-18). That change also didn't create a
variable if the keyword was given but without a value. The purpose
of the change was to always define a variable if a keyword was given.
Lastly, that change didn't protect the CMP0174 logic to make it only
apply to the PARSE_ARGV form.
The first two of the above problems are fixed here by tracking the
keywords given instead of checking which keywords were missing
values. The third problem is also fixed here, being tightly coupled
to the same logic as the first two problems.
Fixes: #26397
Since commit 7abd3137b7 (Linking: Optionally reorder direct dependencies
from LINK_LIBRARIES, 2024-09-19, v3.31.0-rc1~53^2) the strategy name
`PRESERVE_ORDER` has led users to expect that it strictly preserves
order. While the part of the link line generation logic controlled by
`LINK_LIBRARIES_STRATEGY` does preserve order, it is not the last step.
Toolchain-specific de-duplication can cause the order to change on the
actual link line generated in the build system.
Rename the strategies:
* `PRESERVE_ORDER` => `REORDER_MINIMALLY`
* `REORDER` => `REORDER_FREELY`
The new names make it clear that reordering is always possible, just to
varying degrees. Update the `LINK_LIBRARIES_STRATEGY` documentation to
clarify that the strategies do not directly control the final link line.
Fixes: #26400
Issue: #26271
Follow up the python 3.13 and 3.14 support from:
* commit be958c8f35 (FindPython: Add support for Python 3.14, 2024-10-17)
* commit dfab044c44 (FindPython: add support for Python 3.13, 2023-08-14)
with an update for the FindBoost test.
When using the WindowsKernelModeDriver, do not add the new paths
if the variable was not previously defined. Instead, append them
to the existing value.
Add a new GUID for the Experimental mode gate
Add documentation to clarify that `find_package` searches paths in the
specified order and stops at the first match. Clarify documentation of
`CMAKE_FIND_PACKAGE_SORT_*` to match the actual behavior.
Note that no behavior is actually changed, this merely improves the
documentation to reflect actual behavior rather than seeming to imply
something else.
Also, update the test to verify that what we claim in the updated
documentation is what's actually happening.
The default behavior of bin2c is to create binary arrays that have no
null termination. So to use it as a C array we need to explicitly tell
it to create valid null terminated strings.
This will safely allow us to run PTX tests across all CUDA drivers.
Brad King