AutoMoc uses the moc-emitted dependency file of Qt 5.15 to track
dependencies. Such a dependency may well live outside the project and
can vanish, for example when installing a new compiler version.
This situation was detected before, but merely a warning was issued.
Now, we're considering a generated file as out of date if a dependency
is missing and re-generate it.
We also have to remove the missing dependency from the ParseCache.
Otherwise the AUTOMOC target for all generators other than Ninja will
always be out of date.
The ParseCacheChanged flag had to be made atomic, because we're
potentially accessing it from multiple threads. The dependencies vector
itself is not vulnerable in this regard, because there's one vector per
file, and we're accessing exactly one ParseCacheT::FileHandleT per thread.
Fixes: #21136
For Qt >= 5.15.0 and Ninja generators AutoMoc creates a depfile to let
Ninja decide when to run AutoMoc. This was introduced by commit aebfbcaa46
(AutoGen: Use depfiles for the XXX_autogen ninja targets, 2020-01-14,
v3.17.0-rc1~58^2).
However, AutoMoc was not triggered after adding a new moc-able file to
the project. This patch adds the project file (and potentially included
files) to the dependencies in the depfile.
Now, a re-run of AutoMoc is triggered if the project file changes.
Fixes: #21127
Previously, cm::static_reference_cast used invoke_result_t and took the
address of O::get. This is not in complete conformance with standard.
This MR changes the implementation to use std::declval<O>.get() which is
always well-defined.
When building Qt itself, the moc and uic executables are spcecified
via a generator expression of the form $<TARGET_FILE:Qt6::moc>,
which ends populating Moc's and Uic's 'Executable' field but not the
ExecutableTarget and ExecutableTargetName fields.
In such a scenario, the code in
cmQtAutoGenInitializer::InitAutogenTarget fails to add a dependency
on moc (or uic), because ExecutableTarget is null. First try to add
a dependency on the ExecutableTarget if it's not empty, otherwise try
to add a dependency on the path specified in the 'Executable' field.
Issue: #21118
Fix a regression with MPI and CUDA<10.2 that did let `-pthread` flags
slip to nvcc again. In commit b725a19072 (FindMPI: Deny -fexceptions
from NVCC, 2020-07-02, v3.18.0-rc4~12^2) we accidentally forgot to use
the variable containing the replacement result.
Fixes: #21108
Since commit f7347f28c7 (MSVC: Record support for C11 and c_restrict,
2020-08-09) we know about MSVC C language standards. 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.
Since commit c5dd2ca538 (DetermineCompiler: Relax
_CMAKE_TOOLCHAIN_PREFIX detection, 2020-03-25, v3.18.0-rc1~430^2),
`_CMAKE_TOOLCHAIN_PREFIX` may be set even when not cross-compiling.
In this case we may still need to use binutils without any prefix.
Fixes: #21103
When updates are disconnected, don't depend on skip-update because that
target is always considered out of date. Depend directly on the patch target
instead because it already depends on the appropriate target regardless of
whether updates are disconnected or not. This in turn means nothing depends
on the skip-update target, so it has also been removed.
Relates: #21086
The skip-update target is always considered out-of-date. The change in
7249ba9677 (ExternalProject: Enforce that patch depends on update, 2020-04-03)
made the patch target depend on skip-update, which in turn made it
always out of date too. The patch command should only be re-run if the download
needs to be performed again where updates are disconnected.
Fixes: #21086
Since commit e672db628b (FindRuby: Rename variables to match case of
module name, 2020-03-11, v3.18.0-rc1~546^2), the upper-case-prefixed
variable names are for compatibility only but still exist. Put them
back in the documentation.
Issue: #21064
Joerg Bornemann