Add the .NOTPARALLEL target to each local Makefile command-line
interface entry point file so that even with -j we launch only
one "make -f Makefile2" at a time. The actual build rules
in Makefile2 and lower will still run in parallel.
Do not add .NOTPARALLEL for Borland or Watcom make tools because
they do not tolerate it. Other make tools that do not understand
.NOTPARALLEL will not be hurt.
Suggested-by: Robert Luberda <robert-cmake@debian.org>
Commit b04f3b9a (Create make rules for INTERFACE_LIBRARY
targets., 2013-08-21) extended the makefile generator to create
build targets for INTERFACE_LIBRARY targets. No other generators
were extended with this feature.
This conflicts with the feature of whitelisting of target properties
read from INTERFACE_LIBRARY targets. The INTERFACE_* properties
of the INTERFACE_LIBRARY may legitimately contain TARGET_PROPERTY
generator expressions for reading properties from the 'head target'.
The 'head target' would be the INTERFACE_LIBRARY itself when creating
the build rules for it, which means that non-whitelisted properties
would be read.
This was missing from commit c34968a9 (Port some of the generator
API to cmGeneratorTarget., 2012-10-10). The generator targets
stored with the cmMakefile include IMPORTED targets, unlike the
accessor for resgular targets. Before this patch, rules would
be generated for Qt5::Core for example, which result in broken
makefiles.
As an INTERFACE_LIBRARY has no direct link dependencies, we can
short-circuit in cmGeneratorExpressionEvaluator and
in cmGlobalGenerator::CheckLocalGenerators.
As they do not generate any output directly, any generate- or install-
related code acn also be short-circuited. Many of the local generators
already do this.
Because only INTERFACE related properties make sense on INTERFACE_LIBRARY
targets, avoid setting other properties, for example via defaults.
Just enough to reach the BuildMacContentDirectory method and the
NeedRelinkBeforeInstall methods.
In the future, those methods can be moved to cmGeneratorTarget.
Refactor edit_cache tool selection to ask each global generator for its
preference. Teach the Ninja generator to always use cmake-gui because
Ninja by design cannot run interactive terminal dialogs like ccmake.
Teach the Makefile generator to use cmake-gui when also using an "extra"
generator whose IDE has no terminal to run ccmake, and otherwise fall
back to CMAKE_EDIT_COMMAND selection for normal Makefile build systems.
The result is that the depends of the target are created.
So,
add_library(somelib foo.cpp)
add_library(anotherlib EXCLUDE_FROM_ALL foo.cpp)
add_library(extra EXCLUDE_FROM_ALL foo.cpp)
target_link_libraries(anotherlib extra)
add_library(iface INTERFACE)
target_link_libraries(iface INTERFACE anotherlib)
Executing 'make iface' will result in the anotherlib and extra targets
being made.
Adding a regular executable to the INTERFACE of an INTERFACE_LIBRARY
will not result in the executable being built with 'make iface' because
of the logic in cmComputeTargetDepends::AddTargetDepend.
So far, this is implemented only for the Makefile generator. Other
generators will follow if this feature is possible for them.
Make INTERFACE_LIBRARY targets part of the all target by default.
Test this by building the all target and making the expected library
EXCLUDE_FROM_ALL.
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.
Drop the "vsProjectFile" argument from cmTarget::TraceDependencies. It
appears to be the modern equivalent to a hunk added in commit ba68f771
(...added new custom command support, 2003-06-03):
+ name = libName;
+ name += ".dsp.cmake";
+ srcFilesToProcess.push(name);
but was broken by refactoring at some point. The current behavior tries
to trace dependencies on a source file named the same as a target, which
makes no sense. Furthermore, in code of the form
add_executable(foo foo.c)
add_custom_command(OUTPUT "${somewhere}/foo" ... DEPENDS foo)
the "vsProjectFile" value "foo" matches source "${somewhere}/foo.rule"
generated to hold the custom command and causes the command to be added
to the "foo" target incorrectly.
Simply drop the incorrect source file trace and supporting logic.
The API for retrieving per-config COMPILE_DEFINITIONS has long
existed because of the COMPILE_DEFINITIONS_<CONFIG> style
properties. Ensure that the provided configuration being generated
is also used to evaluate the generator expressions
in cmTarget::GetCompileDefinitions.
Both the generic COMPILE_DEFINITIONS and the config-specific
variant need to be evaluated with the requested configuration. This
has the side-effect that the COMPILE_DEFINITIONS does not need to
be additionally evaluated with no configuration, so the callers can
be cleaned up a bit too.
As INTERFACE_COMPILE_DEFINITIONS are now possible, we can have
situations like this:
add_library(foo ...)
add_library(bar ...)
target_link_libraries(foo bar)
target_compile_definitions(bar INTERFACE SOME_DEF)
The INTERFACE_COMPILE_DEFINITIONS of bar determine how foo should be
compiled, and if they change, foo should be rebuilt.
Additionally, as of commit d1446ca7 (Append the COMPILE_DEFINITIONS
from the Makefile to all targets., 2012-09-17), we don't need to
read definitions from the makefile if we read them from the target,
so also de-duplicate the cached info.
The DependInfo for INTERFACE_INCLUDE_DIRECTORIES is already handled
correctly.
This is for consistency throughout cmake. The cmsys version exists
becaues uses of auto_ptr types as return types does not work with
some implementations in ancient compilers.
The code handling IMPLICIT_DEPENDS was only able to track a single file,
the latest file replaced earlier files in the list.
The documentation now mentions that the language has to be prefixed to
every file and the test now uses two implicit dependencies, where only
the second is modified to trigger re-running of the custom command.
Alex
Inspired-by: Michael Wild <themiwi@users.sourceforge.net>
Since commit c8ef6430 (Allow directory names containing '=' and warn if
necessary, 2012-02-06) we allow directories with '=' instead of
rejecting them as was previously done since commit 8704525f (Reject
directory names containing '=', 2011-01-14). However, we did not warn
in all cases that '=' may cause failure, such as when it appears on the
right-hand side of a dependency line.
Both commits above were made assuming that '=' cannot be escaped in Make
syntax, but it can be achieved with a variable:
EQUALS = =
left$(EQUALS)side : right$(EQUALS)side
Use this approach to escape '=' in dependency lines, thus supporting
the character in paths.
All our tests now pass when CMake is built in source and build trees
both containing '=', except for the "OutOfSource" test. It fails in
its coverage of the obscure "OutOfBinary" test case where part of the
build tree is located outside the main build tree of the test. The
reason is that CMake must invoke a command like
$(MAKE) -f /path/with=sign/build.make /path/with=sign/somefile
but the make tool interprets the last argument as a variable assignment.
This is an acceptable limitation, since the case is so obscure, in
exchange for supporting '=' cleanly otherwise.
Add a virtual cmGlobalGenerator::ComputeTargetObjects method invoked
during cmGeneratorTarget construction. Implement it in the Makefile
generator to pre-compute all object file names for each target. Use
the results during generation instead of re-computing it later.
Remove partial implementation added by commit ca0230a3 (check in initial
conv library stuff, 2007-02-16) since it was never finished. It does
not make sense for multi-configuration generators since no specific
build configuration is processed at CMake time.
Eliminate callers of cmMakefile::GetIncludeDirectories.
All callers of GetIncludeDirectories should go through the local generator
object.
Only the local generator calls cmTarget::GetIncludeDirectories directly.
The approach taken by commit 8704525f (Reject directory names containing
'=', 2011-01-14) was perhaps too heavy-handed for avoiding the obscure
cases when '=' in the path fails due to limitations of Make syntax.
Only two CMake tests:
LinkDirectory
OutOfSource
fail when the path contains '=' and they cover obscure cases. Instead
of rejecting such paths outright just warn when the problem may occur.
Now also CMAKE_ASM_INCLUDE_PATH is written into
CMakeDirectoryInformation.cmake, which is necessary to make
the dependency scanning for included files work.
Alex
The Watcom WMake tool has trouble running commands in paths that have
parentheses. We already convert most commands to a shortpath for Watcom
if the path contains a space, but the use of $(CMAKE_COMMAND) hides the
true path from that conversion. Factor the shortpath conversion code
out into a new ConvertShellCommand method. Teach it to convert paths
that contain parentheses as well as spaces. Use the new method to
convert the value of $(CMAKE_COMMAND) and other helper variables.
Files for the ASM language are those assembler files which are processed
by the C/CXX compiler, and they may contain preprocessor directives, so
run the C dependency scanner also on them.
Alex
If you had a + in the name of a target with nmake, it created a variable
in the makefile that used + in its name, which is not allowed by nmake.
To make the implementation easier, + is now not allowed for any make
generators as part of a variable name.
The workaround added by commit 7e92f0b4 (Hack to make echo command work
properly in mingw32-make, 2006-10-05) and updated by commit 69356d8a
(Juse use cmake -E echo instead of the native echo, 2006-10-13) no
longer seems necessary with modern mingw32-make. Furthermore it slows
performance due to the time spent loading a cmake process instead of
plain echo.
The parent commit assumed that "cd /d" would work in all Windows shells.
While all modern versions of windows have shells that support it, the
shells used by NMake and Borland make do not. Borland make does not
seem to even support changing drive letters with "d:". Just revert the
feature for all make tools except MinGW where the shell is known to
support this feature.
Teach cmLocalUnixMakefileGenerator3::CreateCDCommand to change working
directories for make tools using a Windows shell using "cd /d" instead
of just "cd". This tells the shell to change the current drive letter
as well as the working directory on that drive.
Commit abaa0267 (When the working directory for a custom command is on
another drive..., 2007-12-17) fixed the same problem for VS IDE
generators as reported by issue #6150.
Create platform option CMAKE_<lang>_USE_RESPONSE_FILE_FOR_INCLUDES to
enable use of response files for passing the list of include directories
to compiler command lines.
The Makefile, VS, and Xcode generators previously duplicated some custom
command line generation code. Factor this out into a separate class
cmCustomCommandGenerator shared by all generators.
Instead of enforcing verbose makefile, now the generated build command
includes "VERBOSE=1" so the output will be verbose when building in
C::B.
Also removed the now unused setForceVerboseMakefiles().
Alex
The solution seems hackish, but it works: for
NMake only, prepend a no-op command before each
real command that begins with ".
This is really a work-around for an NMake problem.
When a command begins with ", nmake truncates the
first argument to the command after the first :
in that arg. It has a parsing problem.
Workaround..., hackish..., but it should solve
the issue for #9963 and its related friends.
Also, modify the CustomCommand test to replicate
the problem reported in issue #9963. Before the
NMake specific code change, the test failed.
Now, it passes. Ahhhhhh.
If there is a .bat or .cmd file used as a custom command
then the Borland Makefiles generator (specifically) requires
using the "call " syntax before the name of the .bat or .cmd
file. This fix applies to all Makefile based generators where
WindowsShell is true.
This converts the CMake license to a pure 3-clause OSI-approved BSD
License. We drop the previous license clause requiring modified
versions to be plainly marked. We also update the CMake copyright to
cover the full development time range.
Now only the dependencies for the file where the dependencies actually may
have changed are rescanned, before that this was done for all source files
even if only one source file had changed.
This reduces e.g. on my machine the time for scanning the dependencies
of kdelibs/khtml/ when only one file (khtml_global.cpp) has changed from
around 7.5 seconds to 1.2 seconds.
The tests succeed, it does what I expected it to do on kdelibs, and Brad
also reviewed the patch, so I think it should be ok.
Alex
If CMakeDirectoryInformation.cmake is newer than depend.internal the include
directories may have changed, so dependencies need to be scanned again.
Ok by Brad.
Alex
Clinton Stimpson
Brad King
Stephen Kelly
Andreas Mohr
Alex Neundorf
David Cole
Bill Hoffman