In AUTORCC with verbose output write the rcc re-generation reason and
the rcc command as on single string to avoid message chopping in concurrent
builds.
41d262bd3d CTestCoverageCollectGCOV: run gcov with -x
911b97867b CTestCoverageCollectGCOV: run gcov only once
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !3258
New features
------------
CMake's `AUTOMOC` and `AUTOUIC` now cache information extracted when parsing
source files in `CMakeFiles/<ORIGIN>_autogen.dir/ParseCache.txt`.
This leads to faster `<ORIGIN>_autogen` target rebuilds, because source files
will be parsed again only if they're newer than the `ParseCache.txt` file.
The parse cache will be recomputed if it is older than the CMake executable.
`AUTOMOC` and `AUTOUIC` now check if `moc` or `uic` output files are older
than the `moc` or `uic` executable. If an output file is older than the
compiler, it will be regenerated. Therefore if a new `moc` or `uic` version
is installed, all output files will be regenerated.
`AUTOMOC` and `AUTOUIC` error and warning messages are more detailed.
Internal changes
----------------
`moc` and `uic` output file names are not computed in the `_autogen`
target anymore but in `cmQtAutoGenInitializer`. This makes the available at
the configuration stage for improved dependency computations (to be done).
In `AutogenInfo.cmake`, equally sized lists for "source file names",
"source file flags" and "compiler output file names" are passed to the
`_autogen` target. This replaces the separate file lists for
`AUTOMOC` and `AUTOUIC`.
Files times are read from the file system only once by using `cmFileTime`
instances instead of `cmQtAutoGenerator::FileSystem::FileIsOlderThan` calls.
All calls to not thread safe file system functions are moved to non concurrent
fence jobs (see `cmWorkerPool::JobT::IsFence()`). This renders the
`cmQtAutoGenerator::FileSystem` wrapper class obsolete and it is removed.
Instead of composing a single large settings string that is fed to the
`cmCryptoHash`, now all setting sub strings are fed one by one to the
`cmCryptoHash` and the finalized result is stored.
The `std::mutex` in `cmQtAutoGenerator::Logger` is tagged `mutable` and most
`cmQtAutoGenerator::Logger` methods become `const`.
Outlook
-------
This patch provides the framework required to
- extract dependencies from `.ui` files in `AUTOUIC`.
These will help to address issue
#15420 "AUTOUIC: Track uic external inputs".
- generate adaptive `make` and `ninja` files in the `_autogen` target.
These will help to address issue
#16776 "AUTOUIC: Ninja needs two passes to correctly build Qt project".
- generate (possibly empty) `moc` and `uic` files for all headers instead of a
`mocs_compilation.cpp` file.
This will help to address issue
#17277 "AUTOMOC: Provide a option to allow AUTOMOC to compile individual "
"moc_x.cxx instead of including all in mocs_compilation.cxx"
When building for iOS, the compiled target is placed into a bundle.
If a single-configuration generator is used, like Makefiles or Ninja,
the try_compile FILE_COPY behavior fails to find the bundle, because
it only looks for the bundle inside a Debug subfolder (presumably
to support a multi-configuration generator like Xcode).
Consider looking for the bundle in the root try_compile folder, as
well as in the location specified by CMAKE_TRY_COMPILE_CONFIGURATION.
Closes: #19211
On macOS ranlib truncates the fractional part of the static achive
file modification time. If the archive and at least one contained
object file were created within the same second this will make look
the archive older than the object file. On subsequent ninja runs
this leads to re-achiving and updating dependent targets.
As a work-around we touch the archive after ranlib.
Closes: #19222
According to Intel MKL Link Line Advisor, there is no GNU Fortran
interface library provided for OS X variant of Intel MKL. Because of
this missing library, FindBLAS was failing on OS X, looking for
nonexistent library libmkl_gf_[i]lp64.
To prevent this, FindBLAS will now always use Intel Fortran interface
for MKL on OS X (libmkl_intel_[i]lp64), even with GNU Fortran.
Using "-x" we avoid overwriting .gcov files from source files in
different paths which happen to have the same name. It's similar to
"-p", but it produces shorter file names, reducing the risk of reaching
the file system limit.
Running gcov once per .gcda file is not only inefficient, it also
generates wrong data since .gcov files can get overwritten and in
general gcov works with less information.
fakegcov.cmake needs to be able to handle multiple .gcda files for the
test to be meaningful.
