Xcode 2.1 through 4 supported $(CURRENT_ARCH) in a PBXFileReference
'path' value used in the "Link Binary with Libraries" build phase.
CMake uses this to reference object file locations on link lines to
bring in OBJECT library content. However, Xcode 5 now evaluates the
$(CURRENT_ARCH) reference in this context as "undefined_arch" so the
wrong path is given to the linker. There seems to be no alternative way
to produce an architecture-specific value in a PBXFileReference.
Fortunately Xcode 5 now also handles link dependencies for paths linked
through OTHER_LDFLAGS. For Xcode >= 5, move the OBJECT library object
file references from the link build phase to OTHER_LDFLAGS. We can
still show the object files in the source group listing in either case.
Xcode 5.0 now computes dependencies from files linked through
OTHER_LDFLAGS, so we no longer need the XCODE_DEPEND_HELPER hack to
re-link dependents when targets change.
In buggy code like
add_custom_command(
OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/out.h
MAIN_DEPENDENCY ${CMAKE_CURRENT_SOURCE_DIR}/out.h.in
...)
add_custom_command(
OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/out.h
...)
that has more than one rule to generate the same output CMake has always
used the first rule. However, since commit 2268c41a (Optimize custom
command full-path dependency lookup, 2013-08-06) we update the map from
output to cmSourceFile for every rule generating an output, effectively
keeping the last command instead of the first.
Fix this regression by checking for each map update if the output
already has an entry. If so, keep only the original entry. The VS 8
generator triggers this with a special case for generate.stamp rules
that differ between ZERO_CHECK and normal targets, so do not warn for
now. Leave a TODO comment for warning in the future.
Multiple libraries in a single buildsystem can be exported to multiple
export-sets.
If a library from one export set depends on a library from
another export set, the export logic generates a check in the targets
file to verify that the target in the other export set is found. That
check is executed at downstream-find_package-time.
However, a target may depend privately on a target in another export
set. In that case, the depend used to also be listed in the required
targets in the targets file. Change the export logic to ignore the
private link entries instead.
Resolve conflict in Source/cmTestGenerator.cxx by taking "their" side
(test-property-genex). It already accounts for the lower-case change in
"our" side (generate-modern-style).
The old-style add_test() call does not support generator expressions at
all. This also applies to the properties for the test, but it is not
mentioned at all.
This is useful for cases like:
add_test(NAME mytest COMMAND mydriver $<TARGET_FILE:myexe>)
set_tests_properties(mytest PROPERTIES
REQUIRED_FILES "$<TARGET_FILE:myexe>"
WORKING_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}/$<CONFIGURATION>"
)
In this example we require the actual test executable to exist to
run the test in addition to the test driver at argv[0]. Also the
$<CONFIGURATION> expression improves over \${CTEST_CONFIGURATION_TYPE}
because the latter is not normalized for case-sensitive filesystems.
Move property generation from GenerateScriptConfigs to separate copies
in GenerateOldStyle and GenerateScriptForConfig. This causes the
per-config tests generated for the add_test(NAME) signature to each get
their own test properties. This will allow us to later change the
property values based on the test configuration.
While at it, generate lower-case CMake code (e.g. set_tests_properties).
Inspired-by: Ben Boeckel <mathstuf@gmail.com>
State explicitly that the dependencies will build before the target.
Drop wording that may imply the opposite.
Suggested-by: Rob Stewart <robert.stewart@sig.com>
Teach cmListFileLexerDestroy to call cmListFileLexerSetToken with a NULL
token to free the token string buffer. Without this, if an error occurs
before the token cleanup happens when EOF is reached, then the token
string buffer may leak.
Only one argument is required. Also avoid using the same variable name twice in
the input list and follow the convention used elsewhere for optional arguments.
Similar incomplete generator expressions are already tested
in the GeneratorExpression unit test, but those are executed
with add_custom_target. The generator expressions in the include
directories are run through the preprocessor, whereas the ones
run through add_custom_target are not.
The version of Intel Fortran that actually uses 9.10 as a project format
is very old. Default to the latest format version (11.0) and use the
older format only when known to be necessary.
Suggested-by: Dick Munroe <munroe@csworks.com>
We populate and maintain a vector of structs to keep track of
backtraces already, so no need to populate the Properties container
additionally.
For completeness, it is necessary to remove the condition for
populating the vector for only valid library names and generator
expressions. That condition is now determined when evaluating the
generator expressions.
Since commit ad502502 (cmMakefile: Track configured files so we can
regenerate them, 2013-06-18) cmMakefile::ConfigureFile records the
configured file as an output file generated by CMake. The intention is
that for make and ninja we can re-run CMake when one of the files it
generates goes missing. However, files configured temporarily in
CMakeTmp directories by Check* modules do not live past the CMake
invocation.
We have to also track input files to the configure command. In theory
the input to a configure command could it self be a file that is going
to be deleted later (output from a custom command or configure_file).
The feature needs access to all link libraries. In the future that
will only be possible to calculate at generate-time.
Even when the files were generated at configure time, they were
generated after user code in CMakeLists files were generated. No
policy is needed to handle manipulation of the files from CMake
code, because that was never possible.
Since commit ad502502 (cmMakefile: Track configured files so we can
regenerate them, 2013-06-18) cmMakefile::ConfigureFile records the
configured file as an output file generated by CMake. The intention is
that for make and ninja we can re-run CMake when one of the files it
generates goes missing. However, files configured temporarily in
CMakeTmp directories by Check* modules do not live past the CMake
invocation.
Teach cmMakefile::FinalPass to stop tracking files that don't
exist after we are finished generation.
When CMAKE_SUPPRESS_REGENERATION tells us not to create the ZERO_CHECK
target we should not add dependencies on it from other targets either.
Reviewed-by: Mateusz Loskot <mateusz@loskot.net>
This is necessary because custom commands and targets may create
custom files whose names are determined by generator expressions.
For example, clang should be using $<TARGET_FILE> and $<TARGET_FILE_DIR>
instead of reverse engineering the output file name:
http://thread.gmane.org/gmane.comp.compilers.clang.scm/80523
However, that can only be done when ADDITIONAL_MAKE_CLEAN_FILES
also accepts and evaluates generator expressions.
Similarly, KDE uses the LOCATION property where $<TARGET_FILE>
would also be better in KDE4_HANDLE_RPATH_FOR_EXECUTABLE but
also appends the result to ADDITIONAL_MAKE_CLEAN_FILES.
After this patch, both can be ported to generator expressions.
Commit 10a069b5 (Genex: Fix $<CONFIG> with IMPORTED targets and
multiple locations., 2013-07-15) changed the logic here to include
handling of the MAP_IMPORTED_CONFIG_<CONFIG> target property, but
it was buggy in several ways.
Uppercase the configs in all cases, and compare the mapped configs
with the parameter to the CONFIG genex, instead of with the key of
the mapping.
Ensure CMAKE_DATA_DIR, CMAKE_DOC_DIR, and CMAKE_MAN_DIR are always
relative paths in CMake code, and set defaults accordingly. Use the
install() command instead of install_files() and install_targets().
This is more modern and also avoids stripping of the first character
from user-specified destinations.
While at it, fix the default destinations reported in the bootstrap
help.
The MSBuild version for each Visual Studio generator isn't 4.0. With
Visual Studo 2013 the ToolsVersion moved from being tied to the .NET
framework and now has its own version number.
Exclude Ninja and Xcode from the CMP0021 test
They do not behave the same as the makefile generator with
relative paths.
Don't overwrite the header file for in-source builds.
The commits 9db31162 (Remove CMake-language block-end command
arguments, 2012-08-13) and 77543bde (Convert CMake-language
commands to lower case, 2012-08-13) changed most cmake code
to use lowercase commands and no parameters in termination
commands. However, those changes excluded cmake code generated
in c++ by cmake.
Make a similar style change to code generated by cmake.
With this, projects can enable MACOSX_RPATH by default, but still have
a way to install libraries with no install name dirs into locations
such as /usr/local/lib by setting INSTALL_NAME_DIR="".
Convert the incoming value to a path in the same way that the
already-cached value was converted before comparison.
Do the actual comparison only one time, after both values are
normalized.
Since commit 58e52416 (Warn about arguments not separated by whitespace,
2013-02-16) we warn about arguments not separated by spaces. Loosen the
warning to not complain about left parens not separated by spaces from
the preceding token. This is common in code like "if(NOT(X))".
Teach the RunCMake.Syntax test to cover cases of left parens not
separated by spaces and check that no warning appears.
Fixes this warning from g++ 4.8:
CMake/Source/cmDocumentation.cxx:835:23: warning: typedef 'sdoc_t' locally defined but not used [-Wunused-local-typedefs]
SDOC_UNKNOWN} sdoc_t;
^
The extendResult method expects a non-empty parameters vector, as
assured by the normal case. Avoid calling the method when the parser
finds an incomplete generator expression, but has already entered
the state of expecting to find parameters.
b93982f Merge branch 'dev/fix-variable-watch-crash' into cmake-syntax
c50f7ed cmListFileLexer: Modify flex output to avoid Borland warning
bf73264 Warn about unquoted arguments that look like long brackets
58e5241 Warn about arguments not separated by whitespace
e75b69f cmListFileCache: Convert CMake language parser to class
e945949 Add RunCMake.Syntax test cases for command invocation styles
0546484 cmListFileArgument: Generalize 'Quoted' bool to 'Delimeter' enum
28685ad cmListFileLexer: Split normal and legacy unquoted arguments
1eafa3e cmListFileLexer: Fix line number after backslash in string
f3155cd Add RunCMake.Syntax test to cover argument parsing
6aa0c21 variable_watch: Add test for watching a variable multiple times
b86e37c variable_watch: Check newValue for NULL
f9bb20f variable_watch: Don't share memory for callbacks
05dad99 variable_watch: Fix a typo in the error message
00ce12a variable_watch: Prevent making extra entries in the watch map
34b397e variable_watch: Allow specifying the data to match in RemoveWatch
e43e207 variable_watch: Match client_data when finding duplicates
0d6acb1 variable_watch: Add a deleter for the client data
fc7c3b4 variable_watch: Store client data as pointers
The command itself is owned by the cmMakefile class, but the
cmVariableWatch which holds a pointer to the cmVariableWatchCommand via
the client_data for the callback outlives the cmMakefile class in the Qt
GUI. This means that when the cmMakefile is destroyed, the variable
watch is still in effect, but with a stale pointer.
To fix this, each callback is now a separate entity completely and
doesn't rely on the command which spawned it at all.
An example CMakeLists.txt which demonstrates the issue (only displayed
in cmake-gui, so no tests can be written for it):
set(var 0)
variable_watch(var)
When removing a watch on a variable, using the operator [] on the
internal map will create an empty watch if the variable doesn't have any
existing watches. Rather than creating this empty structure in the map,
return if there isn't a watch on the variable already.
If a callback has the same data as another call, we don't want to delete
the old callback. This is because if the client_data is the same, it
might get deleted causing the new client_data to be bogus. Now, AddWatch
will return true if it will use the watch, false otherwise. Callers
should check the return value to know whether client_data was adopted by
the watch or not.
The client data is arbitrary and the callback may be called an
unspecified number of times, so the cmVariableWatch must be the one to
delete the client data in the end (if it is needed at all).
Brad King
Kitware Robot
Stephen Kelly
Eric NOULARD
Ben Boeckel
Rolf Eike Beer
Clinton Stimpson
Robert Maynard
Zack Galbreath
Yury G. Kudryashov