On WinCE we must link both shared libraries and executables with the
/subsystem:windowsce flag. Teach Platform/Windows-MSVC to pass it
to the linker in both cases.
Suggested-by: Gunnar Roth <gunnar.roth@gmx.de>
The check for this flag added by commit v3.0.0-rc1~284^2 (UseSWIG: Name
python module according to swig flags, 2013-11-26) can get false
positives on flags like "-noproxydel". Improve the check to match only
"-noproxy".
Suggested-by: Garth Wells <gnw20@cam.ac.uk>
Since commit v3.0.0-rc1~374^2 (Refactor internal resource location APIs
and initialization, 2013-11-07) a bootstrap-built "cmake" tries to
reference "ctest" and "cpack" executables next to itself, which never
exist. Teach cmSystemTools::FindCMakeResources, when bootstrap-built,
to refer to the "ctest" and "cpack" executables in the location where
they will be built after "make".
From the GCC 4.9 release notes for Fortran:
https://gcc.gnu.org/gcc-4.9/changes.html
"Module files: The version of the module files (.mod) has been
incremented; additionally, module files are now compressed."
Teach cmDependsFortran::ModulesDiffer to look for the gzip magic numbers
at the beginning of the module file. If found, assume the module was
produced by gfortran >= 4.9. The modules do not appear to contain the
date as earlier versions did so we can compare the content directly
and do not actually need to decompress.
It may happen that CMake include is an explicit dependency for some command,
while all CMake includes are set phony in WriteTargetRebuildManifest, this may
lead to duplicate phony generate rules which causes ninja warnings.
We need to remove implicit CMake includes in WriteUnknownExplicitDependencies.
This fixes FindCUDA ninja warnings.
Since commit v2.8.12~300^2~1 (CLI: Suppress the unused warning if the
key value pair is cached, 2013-05-16), cmake::SetCacheArgs saves a
cachedValue pointer and may cause the memory to be freed (by setting the
cache entry) before reading it again. Fix this by saving the old value
in a separate string.
Newer HP-UX versions (e.g. 11.31) also offers F_SETTIMES, but with a
different struct for the timing information.
Suggested-by: Eric Berge <ericmberge@gmail.com>
If we fail to create the generator named by CMAKE_GENERATOR, exit
with an error message instead of crashing. While at it, fix the
wording of the error message when CMAKE_GENERATOR is not set.
Extend the RunCMake.CommandLine test with cases covering the
"cmake --build" option when the named directory does not provide
a CMakeCache.txt with a valid CMAKE_GENERATOR.
Explain how to use $<BUILD_INTERFACE> and $<INSTALL_INTERFACE> directly
in the documentation of the target_include_directories command and
INTERFACE_INCLUDE_DIRECTORIES target property. Otherwise readers need
to notice the link to the cmake-buildsystem(7) manual and find the
example in that to understand the need for these expressions.
Also fix the explanation in cmake-buildsystem(7) to not claim that
relative paths may be used inside a BUILD_INTERFACE expression.
Since commit bef93dc5 (Couple of changes: cache variables now have a map
of properties, 2002-09-11) the cmCacheManager::AddCacheDefinition method
accesses its map entry by reference. However, the commit left the
original entry assignment at the end of the method. With Apple Clang
5.1 and libc++ this self-assignment destroys the cache entry property
map.
Drop the self assignment. Also drop the condition around the call to
UnwatchUnusedCli since it was a self-comparison that must always have
been true.
Teach cmake::HandleDeleteCacheVariables to tolerate a missing HELPSTRING
(NULL pointer) when saving cache entries. In the absence of other bugs
this should not be possible, but avoid the crash just in case.
Add to the Platform/Windows-GNU module list of VS registry entries those
for VS 2013. Also add the name "vcvars64.bat" used by VS 10 and above
for 64-bit tools.
When we kill a child we send SIGSTOP first so that we can traverse its
child processes recursively. On unwinding the recursion we then send
SIGKILL. Current Cygwin has trouble when both signals are sent in quick
succession by the parent process. Add a usleep(1) after sending the
first signal to give up our time slice and give Cygwin a chance to
process the first signal before sending the second.
Previously when linking the intermediate link file for separable compilation
the CUDA_NVCC_FLAGS* were not used. This caused tremendous confusion when
using this feature, and I consider it to be a bug. This change should fix
this.
The languages used in compiling STATIC libraries need to be propagated
to dependents regardless of the settings of INTERFACE_LINK_LIBRARIES or
CMP0022. They are independent of the libraries in the link interface.
Prior to commit v2.8.12~192^2~2 (Introduce the INTERFACE_LINK_LIBRARIES
property, 2013-06-04) the cmTarget::ComputeLinkInterface code path for
"explicitLibraries" could never be taken for STATIC libraries, so the
logic to propagate languages existed only in the non-explicitLibraries
code path. After that commit, INTERFACE_LINK_LIBRARIES could be set for
STATIC libraries to cause the "explicitLibraries" code path to be taken.
The commit also left the old non-explicitLibraries code path conditional
on CMP0022 not being set to NEW. Thus link language propagation was
left missing from two cases by that commit.
The explicitLibraries code path was fixed to propagate languages by
commit v2.8.12~149^2~1 (cmTarget: Fix iface libraries and languages for
static libraries, 2013-07-26). However, the non-explicitLibraries case
was never taught to propagate languages when CMP0022 is set to NEW. Fix
that now. Factor the logic to propagate link languages out of the link
interface libraries conditions so that it always occurs. Update
Tests/Fortran to set CMP0022 to NEW to test this case (because the test
passes only if link language propagation works).
Add ERROR_SHARING_VIOLATION to the set of errors (previously including
only ERROR_ACCESS_DENIED) that cause a rename (MoveFile) on Windows to
retry. The condition was observed when two renames to the same target
file name were happening simultaneously.
Since commit v3.0.0-rc1~305^2 (Ninja: deptype msvc for Intel's compiler
on Windows, 2013-11-25) we used "deps = msvc" for the Intel compiler.
This is correct only on Windows. On Linux we still want "deps = gcc".
Fix the logic to use "deps = msvc" when the compiler id or the "simulate
id" is "MSVC". This will preserve the behavior on Intel for Windows and
fix the behavior on Intel for Linux. In the future this should be
converted to a platform information module variable.
Brad King
Tim Blechmann
Adam Strzelecki
Rolf Eike Beer
Nils Gladitz
Stephen Kelly
James Bigler
Eric Berge