The commit 18a3195a ('Keep track of INCLUDE_DIRECTORIES as a vector
of structs.', 2012-11-19) moved the handling of includes from
cmGeneratorTarget to cmTarget, but in the process introduced this
bug.
link.exe has problems with very very long lines in rsp files too.
Use $in_newline instead of $in variable for rspcontent which
separates the arguments with a newline instead of a simple space
and was specially made for this purpose.
Since commit d46d8df0 (Re-implemented cmGeneratedFileStream to look like a
real stream and replace the destination file atomically, 2004-11-03) our
RenameFile implementation tries to deal with MoveFile not replacing
read-only files. In order to avoid the Get/SetFileAttributes pair we used
stat to test for existence of the destination file. This has a race in
which the destination could be created between the test for existence and
the MoveFile call.
Remove the stat call and use GetFileAttributes to detect whether the file
exists. This shortens the race but does not eliminate it. Use a loop to
try multiple times in case we lose the race. While at it, drop Win9x
support and always use MoveFileEx.
Generator expressions whose output depends on the configuration
now record that fact. The GetIncludeDirectories method can use
that result to cache the include directories for later calls.
GetIncludeDirectories is called multiple times for a target
for each configuration, so this should restore performance for
multi-config generators.
While porting boost to use these features, the generation step took
too long (several minutes before I stopped it). The reason was that
the boost libraries form a large interdependent mesh. The libraries
list their dependencies in their INTERFACE such as:
$<LINKED:boost::core>;$<LINKED:boost::config>;$<LINKED:boost::mpl>
As boost::core already depends on the boost::config libraries, that
expression has no impact on the end-content, as it is removed after
the generation step. There is no DAG issue though, so the generator
expression evaluation would fully evaluate them. In the case of the
config library, it also depends on the core library, so all depends
are followed through that again, despite the fact that they've just
been evaluated. After this patch, the evaluation skips libraries if
they have already been seen via depends or directly in the content.
This patch keeps track of targets whose INTERFACE has been consumed
already. The INCLUDE_DIRECTORIES and COMPILE_DEFINITIONS properties
are whitelisted because repeated content will be stripped out later
during generation. For other properties now and in the future, that
may not be the case.
This establishes that linking is used to propagate usage-requirements
between targets in CMake code. The use of the target_link_libraries
command as the API for this is chosen because introducing a new command
would introduce confusion due to multiple commands which differ only in
a subtle way.
This is both a short form of using a TARGET_DEFINED expression
together with a TARGET_PROPERTY definition, and a way to strip
non-target content from interface properties when exporting.
Projects set interface requirements upstream, and existing
downstreams use of target_link_libraries will consume those interfaces.
This can create a backward compatibility concern as the result may
be changing the order of include directories of downstreams, or another
side-effect of using the INTERFACE properties.
Provide a way for them to emulate the behavior of a version-based
policy in the config file.
Both commit 8a37ebec (Add the target_include_directories command,
2013-01-01) and commit fc61a7a7 (Add the target_compile_definitions
command, 2013-01-08) added command implementations deriving from the new
cmTargetPropCommandBase class. Fix cmTypeMacro declarations of the
inheritance relationship.
7bf490e Make subclasses responsible for joining content.
f6b16d4 Don't allow targets args in the new target commands.
b3a7e19 Make the Property name protected so that subclasses can use it.
We need to make sure we can export targets which have content such
as $<0:$<TARGET_PROPERTY:not_a_target,INTERFACE_INCLUDE_DIRECTORIES>
That means making not finding a target non-fatal here.
Explain in the documentation for the legacy signature
target_link_libraries(foo bar)
that the other signatures like
target_link_libraries(foo LINK_INTERFACE_LIBRARIES ...)
target_link_libraries(foo LINK_PRIVATE ...)
will set the LINK_INTERFACE_LIBRARIES target property and therefore make
libraries specified only by the legacy signature private.
These interface-related link-libraries properties are used to determine
the value of the other INTERFACE properties, so we were getting infinite
recursion and segfaults otherwise.
6063fef Output include directories as LOG messages, not warnings.
aa66748 Specify the target whose includes are being listed.
d70204a Only output includes once after the start of 'generate-time' when debugging.
0d46e9a Store includes from the same include_directories call together.
830246e Export the COMPATIBLE_INTERFACE_BOOL content properties
bd82bb4 Clear the link information in ClearLinkMaps.
e987991 Make INTERFACE determined properties readable in generator expressions.
d9afacc Exit early if we find an inconsistent property.
1800f70 Populate the link information cache before checking dependent properties.
With similar reasoning to the parent commit, as downstreams, we can't
determine what $<CONFIG> generator expressions would be appropriate.
Upstream would have populated the INTERFACE_INCLUDE_DIRECTORIES with
config-specific generator expressions, possibly appropriate for
their DEBUG_CONFIGURATIONS. In theory, if we would add include
directories for a DEBUG intent, we would have to match the upstream
configurations for that.
Rather than attempting to discover the appropriate configurations
at this time, simplify the feature instead. The use of IMPORTED targets
with these commands could still be added in the future if targets
would export their DEBUG_CONFIGURATIONS somehow.
This reverts commit 9cfe4f1b76.
It turns out that correctly adding the content to
the IMPORTED_LINK_INTERFACE_LIBARIES_<CONFIG> of an upstream target
from the buildsystem of a downstream project is not simple.
If upstream had added the INTERFACE content, the config-specific
properties would be determined by the DEBUG_CONFIGURATIONS of
upstream.
As downstream, we don't have any information about what
the DEBUG_CONFIGURATIONS of upstream were, so we can't determine
which configuration-specific properties to populate. The best we can do
is add it to all of them or add it to the ones downstream considers to
be DEBUG_CONFIGURATIONS, neither of which is a good solution.
So, removing the porcelain API for that is the best approach. A human
can still determine which properties to populate and use
the set_property API to populate the desired properies.
Another solution to this would be for upstream targets to publish
what they consider DEBUG_CONFIGURATIONS, but that can be added in
a future release.
During configure-time, GetIncludeDirectories may be called too, for example
if using the export() command. As the content can be different, it should
be output each time then.
Otherwise, we get a separate IncludeDirectoriesEntry for each include,
and that causes unnecessary and confusing splitting in the output when
debugging the INCLUDE_DIRECTORIES property.
The cache here needs to be cleared if GetLinkInformation is called
at configure-time, such as during an export().
The next commit does exactly that, and without this patch,
the LinkLanguage test would fail.
It is not necessary and the current state is unintentional. Before
this patch,
target_link_libraries(foo bar)
causes the LINK_LIBRARIES property of foo to contain
$<TARGET_NAME:bar>
instead of just
bar
This new expression allows checking how a policy was set when a target
was created. That information is only recorded for a subset of policies,
so a whitelist is used.
1d74ba2 Test evaluation target via export for generator expressions
522bdac Export the INTERFACE_PIC property.
4ee872c Make the BUILD_INTERFACE of export()ed targets work.
1d47cd9 Add a test for the interfaces in targets exported from the build tree.
6c828f9 Move the exported check for file existence.
cfd4f0a Move the exported check for dependencies of targets
d8fe1fc Only generate one check per missing target.
f623d37 Don't write a comment in the export file without the code.
b279f2b Strip consecutive semicolons when preprocessing genex strings.
Don't add generator expressions to variables which are used
for CMP0003, CMP0004, and the old-style _LIB_DEPENDS content. They
will not be evaluated when read anyway and would probably confuse
the code reading them.
This makes it legitimate to use target_link_libraries with generator
expressions as arguments.
77d2646 Allow generator expressions in LINK_INTERFACE_LIBRARIES.
94aeaf7 Split LINK_INTERFACE_LIBRARIES export handling into dedicated method.
a3aedb8 Split the generator expression before extracting targets.
b6036d1 Extract the AddTargetNamespace method.
cb1afbf Don't pass a position when determining if a target name is a literal.
f99196d Add cmGeneratorExpression::Split() API.
In the implementation of "cmake -E cmake_link_script", skip lines from
the input file that are empty or contain only whitespace. Do not try to
run a child with no command line.
The Config and IMPORTED_ variants may also contain generator
expressions.
If 'the implementation is the interface', then the result of
evaluating the expressions at generate time is used to populate
the IMPORTED_LINK_INTERFACE_LIBRARIES property.
1) In the case of non-static libraries, this is fine because the
user still has the option to populate the LINK_INTERFACE_LIBRARIES
with generator expressions if that is what is wanted.
2) In the case of static libraries, this prevents a footgun,
enforcing that the interface and the implementation are really
the same.
Otherwise, the LINK_LIBRARIES could contain a generator
expression which is evaluated with a different context at build
time, and when used as an imported target. That would mean that the
result of evaluating the INTERFACE_LINK_LIBRARIES property for
a static library would not necessarily be the 'link implementation'.
For example:
add_library(libone STATIC libone.cpp)
add_library(libtwo STATIC libtwo.cpp)
add_library(libthree STATIC libthree.cpp)
target_link_libraries(libtwo
$<$<STREQUAL:$<TARGET_PROPERTY:TYPE>,STATIC_LIBRARY>:libone>)
target_link_libraries(libthree libtwo)
If the LINK_LIBRARIES content was simply copied to the
IMPORTED_LINK_INTERFACE_LIBRARIES, then libthree links to libone, but
executables linking to libthree will not link to libone.
3) As the 'implementation is the interface' concept is to be
deprecated in the future anyway, this should be fine.
Now that we're processing a LINK_INTERFACE_LIBRARIES string, it
can contain targets. Make sure they are extracted for
namespacing purposes.
This needs to be restricted to strings which can actually have
targets named in them. For example, this is not done for
INTERFACE_COMPILE_DEFINITIONS, because even if there is a target
named 'foo', the string 'foo' in that property means that '-Dfoo'
will be set when compiling.
3581b96 Process the INTERFACE_PIC property from linked dependencies
042ecf0 Add API to calculate link-interface-dependent bool properties or error.
bf5ece5 Keep track of properties used to determine linker libraries.
0380f36 FindOpenGL: add Haiku paths
8e9630c FindGLUT: BeOS does not have libXi and libXmu
50bfedf FindLua51: do not try to link libm on BeOS
3d2e6a0 check for Haiku only with __HAIKU__
7a1b961 Haiku no longer defines __BEOS__
ed96d9a bootstrap: use better defaults for Haiku
This allows a dependee to inform a target that it should have its
POSITION_INDEPENDENT_CODE property set to ON, or OFF. The value of
the POSITION_INDEPENDENT_CODE property, if set, must be consistent
with any INTERFACE_POSITION_INDEPENDENT_CODE properties on dependees.
Add a test covering the consistency checks on platforms where they run.
This new method checks that the property FOO on a target is consistent
with the INTERFACE_FOO properties of its dependees. If they are not the
consistent, an error is reported. 'Consistent' means that iff the
property is set, it must have the same boolean value as all other
related properties.
This makes
set(CMAKE_BUILD_INTERFACE_INCLUDES ON)
add the equivalent of
set_property(TARGET tgt APPEND PROPERTY
INTERFACE_INCLUDE_DIRECTORIES
$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR};${CMAKE_CURRENT_BINARY_DIR}>
)
to every target.
If the headers are in CMAKE_CURRENT_SOURCE_DIR, and the generated headers
are in CMAKE_CURRENT_BINARY_DIR, this is a convenient way to build a target
bar, which depends on foo, just by using target_link_libraries() and adding
the INTERFACE_INCLUDE_DIRECTORIES to the INCLUDE_DIRECTORIES of the target
being linked. There will be more-convenient porcelain API to consume the
property in the future.
This makes it possible to use:
target_link_libraries(foo LINK_INTERFACE_LIBRARIES bar)
where foo is an IMPORTED target. Other tll() signatures are not
allowed.
Previously we kept direct link dependencies in OriginalLinkLibraries.
The property exposes the information in the CMake language through the
get/set_property commands. We preserve the OriginalLinkLibraries value
internally to support old APIs like that for CMP0003's OLD behavior, but
the property is now authoritative. This follows up from commit d5cf644a
(Split link information processing into two steps, 2012-11-01).
This will be used later to populate the link interface properties when
exporting targets, and will later allow use of generator expressions
when linking to libraries with target_link_libraries.
Also make targets depend on the (config-specific) union of dependencies.
CMake now allows linking to dependencies or not depending on the config.
However, generated build systems are not all capable of processing
config-specific dependencies, so the targets depend on the union of
dependencies for all configs.
The 'head' is the dependent target to be linked with the current target.
It will be used to evaluate generator expressions with proper handling
of mapped configurations and is used as the source target of properties.
This requires that memoization is done with a key of a pair of target
and config, instead of just config, because now the result also depends
on the target. Removing the memoization entirely is not an option
because it slows cmake down considerably.
cc676c3 OS X: Detect implicit linker framework search paths
2dd67c7 OS X: Detect implicit link directories on modern toolchains
ba58d0c OS X: Link with all framework search paths, not just the last
894f52f Handle INTERFACE properties transitively for includes and defines.
f5b1980 Populate the ExportedTargets member early in GenerateMainFile
c67b812 Make cycles in target properties ignored, not an error.
d0f950f Use mapped config properties to evaluate $<CONFIG>
26def17 Make all relevant targets available in the genex context.
0c657dc Add API to populate INTERFACE properties in exported targets.
e04f737 Add API to extract target names from a genex string.
b0c8f73 Add the TARGET_NAME generator expression.
77475fe Allow generator expressions to require literals.
b2f1700 GenEx: Add expressions to specify build- or install-only values
Since commit 9a6ff950 (Fix for bug where VS2010 did not use .obj files as
part of the build, 2011-04-01) and commit b291d9e7 (VS10: Fix external
objects generated outside target, 2012-03-19) we try to detect whether an
external object file is also a custom command output in the same target.
This is because VS10 includes .obj custom command outputs on the link line
by default.
VS 11 supports a "<LinkObjects ...>false</LinkObjects>" setting in custom
command rules to tell VS not to link the outputs. From the VS help:
Specify whether the Inputs and output files with specific extensions (.obj,
.lib, .res, .rsc) are passed to the linker.
Treat all external object files the same and add "<Object>" settings for them.
The member variable is checked to determine whether to call
HandleMissingTarget(). As that could be called during the loop in
the comming commits, ensure that it contains all targets being
exported.
Constructs such as these are an error as they are direct self-references:
set_property(TARGET foo APPEND PROPERTY
INCLUDE_DIRECTORIES $<TARGET_PROPERTY:foo,INCLUDE_DIRECTORIES>)
set_property(TARGET foo APPEND PROPERTY
INCLUDE_DIRECTORIES $<TARGET_PROPERTY:INCLUDE_DIRECTORIES>)
However, this is an indirect self-reference in a cycle, and not an error:
set_property(TARGET foo APPEND PROPERTY
INCLUDE_DIRECTORIES $<TARGET_PROPERTY:bar,INCLUDE_DIRECTORIES>)
set_property(TARGET bar APPEND PROPERTY
INCLUDE_DIRECTORIES $<TARGET_PROPERTY:foo,INCLUDE_DIRECTORIES>)
The TARGET_NAME expression, which requires a literal, provides
target names. $<TARGET_PROPERTY:tgt,prop> also provides target
names in the cases where tgt is a literal, so that TARGET_NAME is
not needed then in addition.
This is for specifying INCLUDE_DIRECTORIES relevant to the build-location
or the install location for example:
set_property(TARGET foo PROPERTY
INTERFACE_INCLUDE_DIRECTORIES
"$<BUILD_INTERFACE:${CMAKE_CURRENT_BINARY_DIR};${CMAKE_CURRENT_SOURCE_DIR}>"
"$<INSTALL_INTERFACE:${CMAKE_INSTALL_PREFIX}/include>"
)
A 'bar' target can then use:
set_property(TARGET bar PROPERTY
INCLUDE_DIRECTORIES
"$<TARGET_PROPERTY:foo,INTERFACE_INCLUDE_DIRECTORIES>"
)
and it will work whether foo is in the same project, or an imported target
from an installation location, or an imported target from a build location
generated by the export() command.
Because the generator expressions are only evaluated at build-time, these
new expressions are equivalent to the ZeroNode and OneNode.
The GeneratorExpression test is split into parts. Some shells can't run
the custom command as it is getting too long.
Xcode and VS <= 7.1 will not build targets that have no source files
besides $<TARGET_OBJECTS:...>. Suggest in the documentation that
projects always add at least one real source file.
0941d62 Add a way to print the origins of used include directories.
18a3195 Keep track of INCLUDE_DIRECTORIES as a vector of structs.
76ea420 Use cmsys::auto_ptr to manage cmCompiledGeneratorExpressions
Note it is still possible for CTest to start more than the number of
processes specified by PARALLEL_LEVEL, but this prevents the number of
tests to start from being unbounded because of overflow.
Previously we hard-coded a list of implicit framework directories but
did not account for CMAKE_OSX_SYSROOT or for changes to the list across
OS X versions. Instead we should automatically detect the framework
directories for the active toolchain.
The parent commit added the "-Wl,-v" option to ask "ld" to print its
implicit directories. It displays a block such as:
Framework search paths:
/...
Parse this block to extract the list of framework directories.
Detection may fail on toolchains that do not list their framework
directories, such as older OS X linkers. Always treat the paths
<sdk>/Library/Frameworks
<sdk>/System/Library/Frameworks
<sdk>/Network/Library/Frameworks # Older OS X only
/System/Library/Frameworks
as implicit. Note that /System/Library/Frameworks should always be
considered implicit so that frameworks CMake finds there will not
override the SDK copies.
Refactoring in commit 8d674e78 (Ninja: move -LIBPATH behind -link
option, 2012-09-26) accidentally added code that overwrites the
framework search path flags on each iteration instead of appending.
Change '=' to '+=' to fix it. This affects Makefile and Ninja
generators.
Extend documentation for CMAKE_<LANG>_IMPLICIT_LINK_DIRECTORIES to
explain how it is used by CMake and how it can be influenced by
environment variables.
Inspired-by: Alex Neundorf <neundorf@kde.org>
f0d9385 Makefile: Use modern link information for framework search paths
2bc22bd Xcode: Add frameworks search paths from link dependeny closure (#13397)
Use cmComputeLinkInformation::GetFrameworkPaths to get the list of
framework paths needed by the linker. Drop the now unused framework
information from the old-style cmTarget link dependency analysis.
This should finally fix#13667 and #13762.
Instead of adding special handling to guess whether implicit include dirs
may have been removed, simply make it possible to query the include dirs
without removing the implicit ones.
Alex
The Xcode generator produces FRAMEWORK_SEARCH_PATHS from:
(1) Include directories of the form /path/to/Foo.framework become
-F/path/to so '#include <Foo/H>' can find H in the framework.
(2) Linked frameworks of the form /path/to/Foo.framework become
-F/path/to -framework Foo so the linker can find the framework.
Originally commit 82bb6fae (add framework support to FIND_FILE,
2005-12-27) added these and used the (then current) old-style link
dependency analysis results to get the frameworks. Later a second
setting was added by commit 2ed6191f (add initial xcode framework stuff,
2007-05-08) to transform -F/path/to linker options produced by the old
link line generation into entries appended to FRAMEWORK_SEARCH_PATHS.
Then commit 96fd5909 (Implement linking with paths to library files,
2008-01-22) updated the second setting to directly use the results of
full modern link dependency analysis, but forgot to remove the use of
old-style link results from the original setting location.
The two settings worked together for a while, with the second one
appending to the first. Then commit f33a27ab (Generate native Xcode 3.0
and 3.1 projects, 2009-06-29) changed the internal representation format
produced by the first setting but did not update the second setting to
append to the new representation. As a result, if the first setting
added any paths (usually via the old-style link analysis) then the
second setting containing the modern link analysis results would not be
applied at all.
Fix this by removing use of the old-style link analysis results.
Replace it using the modern link dependencies and remove the second
setting altogether. Now all values for FRAMEWORK_SEARCH_PATHS are
collected in one place so no special append logic is needed.
Historically CMake has always expanded ${} variable references in the
values given to include_directories(), link_directories(), and
link_libraries(). This has been unnecessary since general ${}
evaluation syntax was added to the language a LONG time ago, but has
remained for compatibility with VERY early CMake versions.
For a long time the re-expansion was a lightweight operation because it
was only processed once at the directory level and the fast-path of
cmMakefile::ExpandVariablesInString was usually taken because values did
not have any '$' in them. Then commit d899eb71 (Call
ExpandVariablesInString for each target's INCLUDE_DIRECTORIES,
2012-02-22) made the operation a bit heavier because the expansion is
now needed on a per-target basis. In the future we will support
generator expressions in INCLUDE_DIRECTORIES with $<> syntax, so the
fast-path in cmMakefile::ExpandVariablesInString will no longer be taken
and re-expansion will be very expensive.
Add policy CMP0019 to skip the re-expansion altogether in NEW behavior.
In OLD behavior perform the expansion but improve the fast-path
heuristic to match ${} but not $<>. If the policy is not set then warn
if expansion actually does anything. We expect this to be encountered
very rarely in practice.
Eliminate the platform difference in calling stat. We call stat normally
in other places in the CMake code base just fine. Works everywhere we
work. Will hopefully also fix the Borland Continuous dashboard failure
that is occurring with respect to correctly measuring the modification
time of a freshly generated file.
...revealed by the Nightly dashboard runs last night.
Use "size_t" instead of "std::size_t" so that it compiles with
Visual Studio 6, too.
Fix warnings about shadowed "tmp" local variable and ordering of
member variables vs. lines of code in the constructor initializer
list.
Solution folders are supported as read-only in the VS11 Express
Edition, so do not prohibit their use just because we detect the
express edition (as we did in the VS10 generator).
Inspired-by: Paris
Use same technique here with UuidToString as already found in
cmGlobalVisualStudio7Generator::CreateGUID to avoid using a type
that is not defined on older systems... (VS 7.0 and Borland)
This new CPack generator produces an *.msi installer file.
Requires having the WiX Toolset installed in order to work
properly.
Download the WiX Toolset installer "WiX36.exe" here:
http://wix.codeplex.com/releases/view/93929
Brad King
Stephen Kelly
Kitware Robot
Patrick Gansterer
Alex Neundorf
Rolf Eike Beer
Yury G. Kudryashov
Robert Maynard
Peter Kümmel
Eric LaFranchi
David Cole
Casey Goodlett
Nils Gladitz