This test was broken by commit b2f308c8 (Add support for windows
resources with mingw/msys, 2010-12-22) because the test does not set a
resource compiler which is now required on MinGW for the 'RC' language.
Use windres as the resource compiler for the test.
Problem with CMake 2.8.4-rc1: when you launch the NSIS exe installer
on Windows, the default install path shown to the end user is, at first,
"\CMake 2.8".
This problem started occurring when configuring CMake itself with an
older CMake, after adding CPACK_NSIS_INSTALL_ROOT to fix issue 9148.
So... it's a regression from 2.8.3.
I forgot (again) that when you add a new CPack variable, you must
add it to CMake's CMakeCPack.cmake file or else it is empty when
configured with an older CMake. And on Windows, without a bootstrap
build available, the releases are always configured with an older
version of CMake. This may be the last time this has bitten me,
though, because it is now burned into my brain that problems with
CMake's installer itself are inevitably associated with adding new
CPack variables.
In addition to adding a definition for CPACK_NSIS_INSTALL_ROOT,
I've gone ahead and made it differ for the 32- and 64-bit builds
of CMake to give the end user the expected default value for the
Program Files folder for each one.
And, since I was adding a new 32/64 differentiator anyhow, I made
the "NSIS package name" and "installer registry key base" different
for 64-bit builds, too, by appending " (Win64)" to each one.
These address the concerns mentioned in 9148's related issue:
http://public.kitware.com/Bug/view.php?id=9094 (at least as far
as CMake's installer is concerned). 9094 could still use a good
general fix for all projects, though, and remains open for now.
The parent commit added a warning message whenever a required file
does not exist.
As it turns out, the "required" files never exist when built with
Visual Studio Express editions. Add a variable to suppress these
warning messages because only packagers or naive includers of
this file will care to see such warning messages.
We want to warn about this condition by default so that people who
are using InstallRequiredSystemLibraries without understanding it
fully will have a chance of understanding why it's not working in
the event of missing required files.
But we also want to give projects the ability to suppress this warning
(by "project's choice default") so that they can encourage users who
are restricted to using an Express edition to build their project.
Packagers should explicitly use...
-DCMAKE_INSTALL_SYSTEM_RUNTIME_LIBS_NO_WARNINGS=OFF
...when building releases. That way, their release build process will warn
them about any missing files, but only if their project CMakeLists files
use a construct similar to CMake's:
IF(NOT DEFINED CMAKE_INSTALL_SYSTEM_RUNTIME_LIBS_NO_WARNINGS)
SET(CMAKE_INSTALL_SYSTEM_RUNTIME_LIBS_NO_WARNINGS ON)
ENDIF()
Thanks to "J Decker" on the CMake mailing list for pointing out
that one of the MSVC10_CRT_DIR settings was using "VC90" instead
of "VC100".
After fixing that, I added the code to generate a CMake warning
if one of the files we think is "required" does not exist.
Then, with VS10, there were several other problems that the
warning revealed:
- MSVC10_REDIST_DIR needed more PATHS to be found correctly
- the 64-bit directory is named "x64" now, not "amd64" as in
previous VS versions
- manifest files no longer exist as separate files in the
redist subdirectories (they must be built-in as resources
to the dlls...?)
Previously the Xcode generator would rerun CMake only if input file
dependencies in the top-level directory changed. Teach it to depend on
input files from all directories. Other generators already do this.
Reported-by: Johan Björk <phb@spotify.com>
The LSB header files define major() as a macro but if it is ever called
the macro references symbols not available at link time. Improve the
test for major() to actually call the macro and try to link. This
approach is based on upstream libarchive SVN commit 2866 which fixed
libarchive issue 125, submitted in response to CMake issue #11648.
Inspired-by: Tim Kientzle <kientzle@freebsd.org>
The check works for macros, functions, and variables, but not for types
or enumeration values. Clearly describe the behavior of the check with
respect to each symbol type.
Explicitly state up front that the source-file form of the command links
an executable and expects a 'main' to be defined. While at it, update
the command signature documentation to use a syntax more consistent with
other commands. Also tweak some wording.
Teach the ArgumentExpansion test to expect flattened lists as has always
been the case in the CMake language. Now that the test should pass
enable the failure regex even when CMAKE_STRICT is not on. Replace the
reference to the old ArgumentExpansion test behavior in the workaround
comment in cmMakefile::TryCompile with a full inline explanation.
Some older dashboard machines do not have Rez in /usr/bin
like us modern folk. But most of them will have it in
/Developer/Tools... So make sure we look there.
ce28737 Remove usage of CMAKE_CURRENT_LIST_DIR now that we have CMP0017
7db8db5 Improve documentation and messages for the new CMP0017
db44848 Prefer files from CMAKE_ROOT when including from CMAKE_ROOT
1e69c6f Merge branch 'user-policy-defaults' into policy-CMP0017
65a0a2a Merge branch 'include-command-whitespace' into policy-CMP0017
e33cbda VSMidl Test: Use correct include_directories with VS6 (#11461)
262da91 Prohibit space in HOME value for VSMidl test.
13caaa3 VS10: Finish Midl support (#11461)
Control the root directory of the default directory presented to
the end user of an NSIS installer by a CPack variable.
Previously, the value used in the NSIS script was $PROGRAMFILES,
which is equivalent to the "ProgramFiles" environment variable.
That default value is still the same, but now a project may
override the value by setting this new variable.
Previously, it was hard-coded at 1000 seconds. Now it uses
the value of CMAKE_LONG_TEST_TIMEOUT, which has a minimum
value of 1500 seconds, but might have a longer value depending
on the CTEST_TEST_TIMEOUT value of the driving dashboard.
This should help avoid test failures on overloaded dashboard
machines where it really does take that long to finish the
test successfully.
GNU Make 3.82 incorrectly parses make dependencies involving parentheses
in path names. See related upstream issue:
http://savannah.gnu.org/bugs/?30612
Skip testing the ()-named subdirectory.
Brad King
KWSys Robot
Ben Boeckel
David Cole
Richard Bateman