4499d50 Mark CustomCommand test perconfig.out as SYMBOLIC
f0cdb60 Introduce "generator expression" syntax to custom commands (#11209)
4749e4c Record set of targets used in cmGeneratorExpression
ef9e9de Optionally suppress errors in cmGeneratorExpression
45e1953 Factor per-config sample targets out of 'Testing' test
4091bca Factor generator expression docs out of add_test
bfb7288 Record backtrace in cmCustomCommand
One of Cygwin's goals is to build projects using the POSIX API with no
Windows awareness. Many CMake-built projects have been written to test
for UNIX and WIN32 but not CYGWIN. The preferred behavior under Cygwin
in such projects is to take the UNIX path but not the WIN32 path.
Unfortunately this change is BACKWARDS INCOMPATIBLE for Cygwin-aware
CMake projects! Some projects that previously built under Cygwin and
are Cygwin-aware when they test for WIN32 may now behave differently.
Eventually these projects will need to be updated, but to help users
build them in the meantime we print a warning about the change in
behavior. Furthermore, one may set CMAKE_LEGACY_CYGWIN_WIN32 to request
old behavior during the transition.
Normally we avoid backwards incompatible changes, but we make an
exception in this case for a few reasons:
(1) This behavior is preferred by Cygwin's design goals.
(2) A warning provides a clear path forward for everyone who may see
incompatible behavior, and CMAKE_LEGACY_CYGWIN_WIN32 provides a
compatibility option. The warning and compatibility option both
disappear when the minimum required version of CMake in a project is
sufficiently new, so this issue will simply go away over time as
projects are updated to account for the change.
(3) The fixes required to update projects are fairly insignificant.
Furthermore, the Cygwin distribution has no releases itself so project
versions that predate said fixes tend to be difficult to build anyway.
(4) This change enables many CMake-built projects that did not
previously build under Cygwin to work out-of-the-box. From bug #10122:
"I have built over 120 different source packages with (my patched)
CMake, including most of KDE4, and have found that NOT defining
WIN32 on Cygwin is much more accurate." -- Yaakov Selkowitz
A fully compatible change would require patches on top of these project
releases for Cygwin even though they otherwise need not be aware of it.
(5) Yaakov has been maintaining a fork of CMake with this change for the
Cygwin Ports distribution. It works well in practice. By accepting the
change in upstream CMake we avoid confusion between the versions.
CMake itself builds without WIN32 defined on Cygwin. Simply disable
CMAKE_LEGACY_CYGWIN_WIN32 explicitly in our own CMakeLists.txt file.
The target_link_libraries command requires its first argument to be a
target in the current directory. Document this and update the error
message to be more specific. While at it, format the error message with
a call stack.
Commit e01cce28 (Allow add_dependencies() on imported targets,
2010-11-19) started using cmMakefile::FindTargetToUse to follow
dependencies, including those of GLOBAL_TARGETs like INSTALL and
PACKAGE. Since global targets exist in every directory, dependencies
between them must be traced within each directory too.
Teach FindTargetToUse to check the current directory before checking
globally. For global targets this will find the local copy. For for
normal targets this will be a no-op because they are globally unique.
Factor out reading of CMAKE_CONFIGURATION_TYPES and CMAKE_BUILD_TYPE
into cmMakefile::GetConfigurations. Read the former only in
multi-config generators.
These were implementation details of the unused methods
cmMakefile::RegisterData
cmMakefile::LookupData
We simply remove the methods, members, and class cmData.
Prepare to switch to the workflow described by "git help workflows". In
this workflow, the "master" branch is always used to integrate topics
ready for release. Brand new work merges into a "next" branch instead.
We need a new versioning scheme to work this way because the version on
"master" must always increase.
We no longer use an even/odd minor number to distinguish releases from
development versions. Since we still support cvs checkout of our source
tree we cannot depend on "git describe" to compute a version number
based on the history graph. We can use the CCYYMMDD nightly date stamp
to get a monotonically increasing version component.
The new version format is "major.minor.patch.(tweak|date)". Releases
use a tweak level in the half-open range [0,20000000), which is smaller
than any current or future date. For tweak=0 we do not show the tweak
component, leaving the format "major.minor.patch" for most releases.
Development versions use date=CCYYMMDD for the tweak level. The
major.minor.patch part of development versions on "master" always
matches the most recent release.
For example, a first-parent traversal of "master" might see
v2.8.1 2.8.1.20100422 v2.8.2
| | |
----o----o----o----o----o----o----o----o----
Since the date appears in the tweak component, the next release can
increment the patch level (or any more significant component) to be
greater than any version leading to it. Topic branches not ready for
release are published only on "next" so we know that all versions on
master lead between two releases.
Previously we silently ignored such calls and set nothing. The commit
"Initialize directory scope with closure of parent" inroduced a bad test
for the top scope. This commit fixes the test to avoid dereferencing a
null pointer, and adds a warning when the case is encountered.
We revert commit "Create CMake Policy CMP0015 to fix set(CACHE)" because
the NEW behavior of the policy breaks a valid use case:
# CMakeLists.txt
option(BUILD_SHARED_LIBS "..." ON)
add_library(mylib ...)
set(BUILD_SHARED_LIBS OFF) # we want only mylib to be shared
add_subdirectory(ThirdParty)
# ThirdParty/CMakeLists.txt
option(BUILD_SHARED_LIBS "..." ON)
# uh, oh, with NEW behavior this dir uses shared libs!!!
We'll re-introduce the policy later with a different change in behavior
to resolve the motivating case, which was more subtle but less common.
See issue #9008.
In cmMakefile::AddCacheDefinition we collapse paths specified in PATH or
FILEPATH cache entries originally specified on the command line with
UNINITALIZED type. This commit fixes the logic to avoid collapsing
<var>-NOTFOUND and other false values. The change allows other CMake
code to force a NOTFOUND value on an entry with UNINITALIZED type.
This method is called during ConfigureFinalPass on every target. It
gives each target a chance to do some final processing after it is known
that no more commands will affect it. Currently we just call the old
AnalyzeLibDependencies that used to be called directly.
This commit creates target and directory properties to enable the Intel
interprocedural optimization support on Linux. Enabling it adds the
compiler option '-ipo' and uses 'xiar' to create archives.
See issue #9615.
This creates cmTarget::GetFeature and cmMakefile::GetFeature methods to
query "build feature" properties. These methods handle local-to-global
scope and per-configuration property lookup. Specific build features
will be defined later.
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.
In cmMakefile::AddSubDirectory we were checking for addition of the same
source directory multiple times. However, the check code was incorrect
because it compared pointers instetad of pointed-to strings. Since the
check was written, a better check was added right after it to enforce
unique binary directories (in which case duplicate sources are fine).
This commit simply removes the old-style check code.
The commit "Improve dynamic variable scope implementation" optimized
function scopes using an efficient parent scope pointer. However, the
parent scope used to initialize a new directory might not exist later
(like add_subdirectory called inside a function of the parent scope).
This caused CMake to crash when following the dangling pointer to the
original parent scope.
We fix the problem in this commit by always computing the closure of the
parent scope at directory initialization time so that no parent scope
pointer is needed. See issue #9538.
The set(CACHE) and option() commands should always expose the cache
value. Previously we failed to expose the value when it was already set
if a local variable definition hid it. When set to NEW, this policy
tells the commands to always remove the local variable definition to
expose the cache value. See issue #9008.
The boolean overload of this method was used only to implement option().
We re-implement option() in terms of the main method and removes the
now-unused signature. This removes some duplicate code that had already
fallen behind on changes (it was not removing the local definition
instead of setting it).
The try_compile command builds the cmTryCompileExec executable using the
cmTryCompileExec/fast target with Makefile generators in order to save
time since dependencies are not needed. However, in project mode the
command builds an entire source tree that may have dependencies.
Therefore we can use the /fast target approach only in one-source mode.
In cmMakefile we save all invoked commands so that FinalPass can be
called on them later. Most commands have no final pass, so we should
keep only the few that do.
Previously each new variable scope (subdirectory or function call) in
the CMake language created a complete copy of the key->value definition
map. This avoids the copy using transitive lookups up the scope stack.
Results of queries answered by parents are stored locally to maintain
locality of reference.
The class cmDefinitions replaces cmMakefile::DefinitionsMap, and is
aware of its enclosing scope. Each scope stores only the definitions
set (or unset!) inside it relative to the enclosing scope.
In CMake 2.6.3 and below we silently accepted duplicate build
directories whose build files would then conflict. At first this was
considured purely a bug that confused beginners but would not be used in
a real project. In CMake 2.6.4 we explicitly made it an error.
However, some real projects took advantage of this as a "feature" and
got lucky that the subtle build errors it can cause did not occur.
Therefore we need a policy to deal with the case more gracefully.
See issue #9173.
Previously we rejected all preprocessor definition values containing
spaces for the VS6 IDE generator. In fact VS6 does support spaces but
not in combination with '"', '$', or ';', and only if we use the sytnax
'-DNAME="value with spaces"' instead of '-D"NAME=value with spaces"'.
Now we support all definition values that do not have one of these
invalid pairs. See issue #8779.
The add_definitions() command and COMPILE_DEFINITIONS dir/tgt/src
properties support preprocessor definitions with values. Previously
values were not supported in the VS6 generator even though the native
tool supports them. It is only values with spaces that VS6 does not
support. This enables support and instead complains only for values
with spaces. See issue #8779.
This moves code which generates ADD_TEST and SET_TESTS_PROPERTIES calls
into CTestTestfile.cmake files out of cmLocalGenerator and into a
cmTestGenerator class. This will allow more advanced generation without
cluttering cmLocalGenerator. The cmTestGenerator class derives from
cmScriptGenerator to get support for per-configuration script
generation (not yet enabled).
The second argument of add_subdirectory must name a unique binary
directory or the build files will clobber each other. This enforces
uniqueness with an error message.
This defines global, directory, and target properties
RULE_LAUNCH_COMPILE, RULE_LAUNCH_LINK, and RULE_LAUNCH_CUSTOM. Their
values specify 'launcher' command lines which are prefixed to compile,
link, and custom build rules by Makefile generators.
This teaches cmMakefile::GetProperty and cmake::GetProperty methods to
return NULL when the property name is NULL, making them more robust and
consistent with the behavior of cmTarget::GetProperty.
Isolation of policy changes inside scripts is important for protecting
the including context. This teaches include() and find_package() to
imply a cmake_policy(PUSH) and cmake_policy(POP) around the scripts they
load, with a NO_POLICY_SCOPE option to disable the behavior. This also
creates CMake Policy CMP0011 to provide compatibility. See issue #8192.
This teaches functions and macros to use policies recorded at creation
time when they are invoked. It restores the policies as a weak policy
stack entry so that any policies set by a function escape to its caller
as before.
A 'weak' poilcy stack entry responds normally to queries. However,
setting a policy in a weak entry will recursively set the policy in the
next entry too. This also gives the internal interface to create a weak
entry the option to provide an initial PolicyMap for it.
This creates cmMakefile::PolicyPushPop to push and pop policy scope
automatically. It also enforces balanced push/pop pairs inside the
scope it handles.
This defines PolicyMap as a public member of cmPolicies. Its previous
role as a policy stack entry is now called PolicyStackEntry and
represented as a class to which more information can be added later.
If a logical block terminates with mismatching arguments we previously
failed to remove the function blocker but replayed the commands anyway,
which led to cases in which we failed to report the mismatch (return
shortly after the ending command). The recent refactoring of function
blocker deletion changed this behavior to produce an error on the ending
line by not blocking the command. Furthermore, the function blocker
would stay in place and complain at the end of every equal-level block
of the same type.
This teaches CMake to treat the begin/end commands (if/endif, etc.) as
correct and just warns when the arguments mismatch. The change allows
cases in which CMake 2.6.2 silently ignored a mismatch to run as before
but with a warning.
This centralizes construction of the error message for an unclosed
logical block (if, foreach, etc.). We record the line at which each
block is opened so it can be reported in the error message.
This uses a stack of 'barriers' to efficiently divide function blockers
into groups corresponding to each input file. It simplifies detection
of missing block close commands and factors it out of ReadListFile.
Brad King
David Cole
Bill Hoffman
Alex Neundorf