Before, find_package in Config mode always set Foo_FOUND to true if the
Config file has been found and could be executed.
If the Config file itself detected some problem, like a missing dependency,
it did not have a way to signal to the outside that the package is not working.
With this patch, if a Config file sets Foo_FOUND to FALSE, this is taken into
account and not overridden.
Alex
55b2aa8 Use generalized RunCMake test infrastrucure for build_command test
eb33000 Use generalized RunCMake test infrastrucure for find_package test
42a81e7 Add stronger infrastructure for CMake-only tests
CMake does not automatically pass the Fortran module output directory
location to the module search path. Add a note to the documentation
that the location must also be specified by INCLUDE_DIRECTORIES.
The CMakeCommands.build_command test performs output/error checking
so move it over to RunCMake to re-use the generalized infrastrucure.
This is the only test left using Tests/CMakeCommands/CMakeLists.txt
so remove it.
The CMakeOnly directory added by commit 9a20abf0 (Add infrastructure for
CMake-only tests, 2012-01-11) was sufficient only for tests that always
run CMake to successfully configure a project. Later commit eeaaffcb
(find_package: Test error and warning messages in failure cases,
2012-02-28) added a sample test that covers failure cases.
Generalize the above to create new "RunCMake" test infrastructure that
can run CMake multiple times for a single project with different
variations and check for expected result/stdout/stderr. Allow for both
successful and failing CMake project configuration cases. This will be
useful to test error messages and failure behavior.
Our TEST* macro calls pass arguments such as "int64_t" with the
expectation that the preprocessing token will be used literally.
Some platforms #define int64_t as "long long" which is not a valid
preprocessing token. Perform preprocessor symbol concatenation
on the type names at the first level of macro evaluation to avoid
expanding the names.
The long-standing sporadic failures of CPack tests on the Mac dashboards
are caused by an occasional problem running hdiutil. To compensate for
this, a retry loop was added in the code in a previous commit: a9fa71a4
... but the logic for breaking out of the retry loop was flawed, breaking
out of the loop (and not retrying) when the hdiutil command returns an
error instead of when it returns success.
This commit fixes the flawed logic, bumps up the number of retries from
4 to 10, and adds a half-second delay in between retries.
The delay is specifically added in case a virus checker or spotlight indexer
is temporarily causing the hdiutil failure by hanging onto a newly created
file longer than hdiutil expects it to.
As with all sporadically occurring issues, we'll never know if this is
really fixed all the way. But I'll be happy even if we can only get it to
happen just a bit less often.
b28e7fa VS6: Avoid SBCS test on VS6 (#12189)
df19b9c VS6: Avoid _MBCS define when _SBCS is defined (#12189)
ba89e92 Visual Studio: Allow setting Single Byte Character Set (#12189)
d662dff Fix shadowed variable warning on dashboard results
f66e735 Fix compiler warning reported on older Borland dashboard.
d90eed4 Fix compiler error reported on older Borland dashboard.
8233636 Update the documentation regarding INCLUDE_DIRECTORIES.
d899eb7 Call ExpandVariablesInString for each target's INCLUDE_DIRECTORIES
c21db87 Make search paths ordered and unique
22021f0 Remove cmMakefile::GetIncludeDirectories
9106b56 Extract and use the INCLUDE_DIRECTORIES target properties.
840509b Keep the INCLUDE_DIRECTORIES target property up to date.
a4d5f7b Add API to get the ordered includes for a target.
8adaee2 CMake: Eliminate cmMakefile::IncludeDirectories
7620932 Remove include flags memoization.
97a5faa Make it safe to call this method without creating duplicates.
edd5303 Refactor GetIncludeFlags to take includes instead of fetching them
The function configure_package_config_file() may be used instead of
configure_file() for generating the Config.cmake files for installation,
they help to make those files relocatable.
Alex
The previous commit, df19b9ca, assumed that the corresponding
_UNICODE functionality in the VS6 generator actually worked.
That turns out not to be the case. Unicode definition does not
actually suppress _MBCS definition, so neither does this new
code...
Don't test it here, since the patch submitter for _SBCS does
not need VS6 support.
If somebody needs _UNICODE and _SBCS support to suppress the
definition of _MBCS in the VS6 generator, it is work yet to do.
Patches welcome.
Alex Neundorf
Brad King
KWSys Robot
David Cole
Alexandru Ciobanu
Clinton Stimpson
Droscy
Eric NOULARD