When linking with cmake and vs_link_* the command line
could be too long for cmd.exe, which needs not to be
called in this case. (was not cached by a test)
Introduce rules which don't use the shell and use this
rule when there are no pre or post step.
For free we get a small speedup, because cmd is then
not called.
Also be more accurate when estimating the
command line length.
All tests are passing as of this morning on the CMake dashboard in
the Nightly Expected section.
Nice job to all the Ninja+CMake contributors. Congrats!
rc.exe doesn't support /showIncludes.
Because .rc files also #include stuff we can
misuse cl.exe to get the included files.
Done one the fly by cmcldeps.
Windows and Apple machines have predominantly used case-insensitive
filesystems so our file(INSTALL) command uses case-insensitive pattern
matching. It is implemented by converting the pattern and file path to
lower case before matching. The FILES_MATCHING option is implemented by
excluding a path that does not match any pattern unless it is a
directory that must be searched recursively. However, the test that an
excluded path is a directory is executed on the lower-case path and
therefore fails on mixed-case input paths on case-sensitive filesystems.
Fix the file(INSTALL) implementation to use the lower-case path only for
pattern matching and preserve the original path for tests against the
real filesystem.
bd34963 Refactor generation of shared library flags
55d7aa4 Add platform variable for flags specific to shared libraries
31d7a0f Add platform variables for position independent code flags
Older C++ compilers do not provide a standard std::stringstream.
Use our compatibility interfaces instead.
Also avoid std::stringstream(openmode) signature. Our approximate
stringstream implementation provided when the standard one is not
available does not support the openmode argument.
CMAKE_SHARED_LIBRARY_<lang>_FLAGS has flags on various platforms for a
variety of purposes that are correlated with shared libraries but not
exclusive to them. Refactor generation of these flags to use new
purpose-specific platform variables
CMAKE_<lang>_COMPILE_OPTIONS_DLL
CMAKE_<lang>_COMPILE_OPTIONS_PIC
CMAKE_<lang>_COMPILE_OPTIONS_PIE
Activate the DLL flags specifically for shared libraries. Add a new
POSITION_INDEPENDENT_CODE target property to activate PIC/PIE flags, and
default to true for shared libraries to preserve default behavior.
Initialize the new property from CMAKE_POSITION_INDEPENDENT_CODE to
allow easy global configuration in projects.
Although the default behavior is unchanged by this refactoring, the new
approach ignores CMAKE_SHARED_LIBRARY_<lang>_FLAGS completely. We must
leave it set in case projects reference the value. Furthermore, if a
project modifies CMAKE_SHARED_LIBRARY_<lang>_FLAGS it expects the new
value to be used. Add policy CMP0018 to handle compatibility with
projects that modify this platform variable.
Add a PositionIndependentCode test on platforms where we can get
meaningful results.
When global property TARGET_SUPPORTS_SHARED_LIBS is FALSE we should
still allow OBJECT libraries. This was an oversight in commit b87d7a60
(Add OBJECT_LIBRARY target type, 2012-03-12). While at it, fix the
warning message to report context.
cmcldeps wraps cl and adds /showInclude before calling cl.
It parses the output of cl for used headers, drops system
headers and writes them to a GCC like dependency file.
cmcldeps uses ATM ninja code for process handling,
but could be ported later to SystemTools.
TODO: Why needs ninja multiple calls in the BuildDepends test?
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.
When reading archive entries from disk strip any "fflags" entry headers
that may have been loaded from the filesystem when libarchive is built
with HAVE_STRUCT_STAT_ST_FLAGS (struct stat has 'st_flags'). The local
filesystem flags are not useful for distribution. Furthermore, GNU tar
does not understand the "SCHILY.fflags" extended header used to store
the flags in the archive. Use the approach from commit e8558efa
(cmArchiveWrite: Clear xattr and acl from entries, 2011-04-07) to remove
the flags and avoid producing the non-portable extended header.
When MinGW is used slashes are used for dependencies
because ar.exe can't read rsp files with backslashes.
Many thx to Claus Klein for starting working on this.
Building CMake with g++ 2.9-aix51-020209 on an AIX 5.3 system gives:
cmsys/hashtable.hxx: In function `const long unsigned int *cmsys::get_stl_prime_list ()':
cmsys/hashtable.hxx:399: warning: sorry: semantics of inline function static data
`const long unsigned int _stl_prime_list[31]' are wrong (you'll wind up with multiple copies)
cmsys/hashtable.hxx:399: warning: you can work around this by removing the initializer
Give get_stl_prime_list internal linkage.
3545645 Exclude the CompileCommandOutput test on WIN32.
fbaddf4 Escape the source file to be compiled if required.
db839be Make the CMAKE_EXPORT_COMPILE_COMMANDS option work with Ninja.
8778357 Add newline to the output.
2c04bc0 Move the EscapeJSON method to a sharable location.
4986d52 Use CPACK_xxx and CMAKE_xxx in a consistent way.
f90223c Fix KWStyle warning
47f0dbd CPack add necessary check to detect/warns/error on ABSOLUTE DESTINATION
6ba055b CPack add easy possibility to warn about CPACK_SET_DESTDIR
Commit "KWSys: Fix SystemTools environment memory handling" (2012-04-26)
added a _WIN32 case inside !KWSYS_CXX_HAS_ENVIRON_IN_STDLIB_H to dllimport
the "environ" global. Howver, KWSYS_CXX_HAS_ENVIRON_IN_STDLIB_H is true
on every Windows toolchain we support so the case is never reached.
Furthermore, even if it were reached the use of dllimport is incorrect
because the toolchain might not be compiling with a dynamic runtime
library. Remove the unused incorrect line and supporting conditionals.
Fortran sources that pass through the C preprocessor may use
#include "header"
syntax or
#include <header>
syntax. CMake already follows the former. Teach it to follow the
latter.
More generally add the check for possible generator "activation" at
runtime depending on a generator specific check.
The dynamic behavior is currently implemented only for MacOS
and should be fully backward compatible for other system.
Inspired-By Tom Hughes <tomtheengineer@gmail.com>
CMAKE_xxx vars are now used in the CMake-generated cmake_install.cmake
script while CPACK_xxx equivalent vars are used from within CPack.
CPack is responsible for getting/forwarding definitions of
CPACK_xxxx var corresponding to CMAKE_xxxx when invoking
CMake-generated install scripts.
As a consequence:
CMAKE_ABSOLUTE_DESTINATION_FILES
CMAKE_WARN_ON_ABSOLUTE_INSTALL_DESTINATION
CMAKE_ERROR_ON_ABSOLUTE_INSTALL_DESTINATION
may be used from outside CPack as well.
e.g.
cmake -DCMAKE_ERROR_ON_ABSOLUTE_INSTALL_DESTINATION=1 -P cmake_install.cmake
works as expected.
c806b23 CDash now supports lots of files in coverage. So, show all files.
761d931 Do not try to run bullseye coverage if COVFILE env is empty.
5b69ce4 Update test data to match new coverage format.
1b418f1 Change GT.M Coverage Parser global
b0c07a1 Disable bullseye coverage for mumps coverage test.
0a169e6 Remove uncovered files from cache coverage data.
a7abf5e Add ability to specify more than one package directory or coverage directory.
220afca Use <TARGET_FILE> expression to run ctest so it works with Xcode and VS IDE.
62f6bce Use a script to run the test because WORKING_DIRECTORY is not in 2.8.2.
f5c5db0 Fix some warnings and a bug where it went past the length of a vector.
7955e99 Add support for Cache coverage.
a86cd33 Add virutal destructor to silence warning.
319eeb0 Add test for mumps coverage. Also refactor code to prepare for cache coverage.
72210c2 Fix line length.
dd07161 Fix warning about char* instead of const char*.
e6412e0 Add support to ctest for GTM mumps coverage.
In trace mode ('--trace'), any 'elseif' or 'else' commands that are
evaluated as part of a conditional block will be printed. Previously,
only the opening 'if' command of a conditional block was printed.
The [usually] wrong usage of absolute DESTINATION in INSTALL rules
keeps popping-up on the ML. We shall have some way to:
1) easily detect it.
2) forbids this for some CPack generator like NSIS
In fact it should certainly be forbidden for *any* generators
when used on Windows but we may implements that on top of the current
patch.
The patch ask the task to the generated cmake_install.cmake scripts.
Those scripts are a little bit more complicated with that but
iff there are absolute DESTINATION. This cost nothing if relative
DESTINATION are used.
Two new vars are introduced (and documented to handle that):
CPACK_WARN_ON_ABSOLUTE_INSTALL_DESTINATION
and
CPACK_ERROR_ON_ABSOLUTE_INSTALL_DESTINATION
Until now an unnamed component was always named "Unspecified".
Now this name is taken from the new cmake variable CMAKE_INSTALL_DEFAULT_COMPONENT_NAME,
which is initialized to "Unspecified". But it can now be set to something
project-specific, per directory
Alex
The coverage global should be in the local namespace. This means the global
will be ^ZZCOVERAGE instead of ^COVERAGE. Change the parser to look for ^ZZCOVERAGE
instead of the old ^COVERAGE
Header files listed in a target's PUBLIC_HEADER or similar properties
are marked as OS X Framework content. Refactoring performed by
commit 11d9b211 (Add cmGeneratorTarget to represent a target during generation, 2012-03-07)
commit 45c2f932 (Simplify cmMakefileTargetGenerator using cmGeneratorTarget, 2012-03-07)
commit 328c0f65 (Simplify cmVisualStudio10TargetGenerator source classification, 2012-03-19)
and related commits accidentally removed such files from treatment as
normal header files by the VS generator (generators other than Makefiles
and Xcode). Move handling of such files out of cmGeneratorTarget and
back to cmMakefileTargetGenerator. The central cmGeneratorTarget
classification will always treat them as header or extra sources.
The hash_fun.hxx header is configured whether FundamentalType is enabled
or not and so cannot depend on it. Run the relevant platform tests
whether or not FundamentalType is on and configure the result directly
into hash_fun. While at it, remove the dependence of SystemInformation
on FundamentalType too since it needs only information that we now
always compute.
d931ce9 VS10: Generate relative source paths when possible (#12570)
b2e7c7a VS11: Do not use source path conversion workaround specific to VS 10
4248132 VS10: Convert paths normally unless forced to relative
c2ba6ba VS10: Simplify vcxproj.filter file generation
4f2d9d2 VS10: Refactor custom commands to use WriteSource
Add a boolean target property NO_SONAME which may be used to disable
soname for the specified shared library or module even if the platform
supports it. This property should be useful for private shared
libraries or various plugins which live in private directories and have
not been designed to be found or loaded globally.
Replace references to <CMAKE_SHARED_LIBRARY_SONAME_${LANG}_FLAG> and
hard-coded -install_name flags with a conditional <SONAME_FLAG> which is
expanded to the value of the CMAKE_SHARED_LIBRARY_SONAME_${LANG}_FLAG
definition as long as soname supports is enabled for the target in
question. Keep expanding CMAKE_SHARED_LIBRARY_SONAME_${LANG}_FLAG in
rules in case third party projects still use it. Such projects would
not yet use NO_SONAME so the adjacent <TARGET_SONAME> will always be
expanded. Make <TARGET_INSTALLNAME_DIR> NO_SONAME aware as well. Since
-install_name is soname on OS X, this should not be a problem if this
variable is expanded only if soname is enabled.
The Ninja generator performs rule variable substitution only once
globally per rule to put its own placeholders. Final substitution is
performed by ninja at build time. Therefore we cannot conditionally
replace the soname placeholders on a per-target basis. Rather than
omitting $SONAME from rules.ninja, simply do not write its contents for
targets which have NO_SONAME. Since 3 variables are affected by
NO_SONAME ($SONAME, $SONAME_FLAG, $INSTALLNAME_DIR), set them only if
soname is enabled.
Since commit ed0075bd (Use relative paths for custom command inputs,
2011-06-22) CMake generates full paths to source files in VS 10 project
files to avoid trouble with deep source/build tree paths. However, the VS
10 IDE will not populate the source file property dialog for a file
referenced by full path. Instead use a relative path when possible. When
not possible produce a detailed warning explaining the problem and
suggesting use of shorter directory paths.
CMake <= 2.8.4 generated VS 10 project files with a relative path to
source files. Then commit ed0075bd (Use relative paths for custom command
inputs, 2011-06-22) switched to using relative paths only for source files
holding custom commands and full paths for other sources. This behavior
was inhereted by the VS 11 generator but is not needed so use the
workaround only for exactly VS 10. Explain the behavior in comments.
Most CMake generators produce relative paths within the build tree and
full paths to files outside the build tree. Make VS 10 and VS 11
project files consistent with this approach except for paths forced to
be relative to work around a VS 10 bug.
Remove the duplicate source file classification logic used to generate
the filter files. Instead record during the main vcxproj file
generation the source files generated for each tool. Also record
whether or not each source file is converted to a relative path. Use
the recorded result during filter generation to ensure consistency
between the project file and filter file.
The SystemTools::PutEnv function tries to provide the "putenv" API
without leaking memory. However, the kwsysDeletingCharVector singleton
frees memory that may still be referenced by the environment table,
having been placed there by putenv. If any static destruction or
processing by an external tool happens after the singleton is destroyed
and accesses the environment it will read invalid memory.
Replace use of putenv with setenv/unsetenv when available. The latter
manage internal copies of the values passed instead of referencing the
original memory. When setenv/unsetenv are not available use putenv with
a singleton that removes its values from the environment before freeing
their memory. This requires an "unputenv" implementation. On at least
some platforms it must be written in terms of "putenv" because other
APIs are not available and direct modification of the "environ" global
is not safe (e.g. on Windows there is interaction with "wenviron").
Fortunately either putenv("A=") or putenv("A") will remove "A" from the
environment on these platforms. On other platforms fall back to direct
manipulation of "environ".
Also add UnPutEnv to the API and add a test for the behavior of both.
Kitware Robot
Peter Kuemmel
David Cole
Alex Neundorf
Brad King
Stephen Kelly
Zaheer Chothia
Daniel R. Gomez
Fraser Hutchison
Eric NOULARD
Peter Collingbourne
Brian Helba
Jan Schaffmeister
Bill Hoffman
Nicolas Despres
Minmin Gong
Joseph Snyder
Modestas Vainius
Clinton Stimpson