Since commit v3.1.0-rc1~110^2 (Tolerate symlinks during RPATH ordering,
2014-09-09) we call realpath() for every directory ordering constraint
check. On some platforms/filesystems this is slow, so memoize the
result of the call for each directory.
We enforce policy push/pop balance around any scope that pushes/pops a
snapshot. Therefore a snapshot may never reference entries of
PolicyStack that were created in nested scopes. Free storage of
short-lived policy stack entries when they are popped.
Add a method to increment an iterator (follow the "up" pointer) to the
previous level in the stack of scopes and free storage of the top of the
stack if possible. This will allow short-lived scopes to be created and
destroyed by matching Push/Pop pairs without accumulating storage.
Logically the method pushes a nested scope on top of a given scope
because the "up" pointer sequence forms a stack independent of any
other branches of the tree.
Everywhere we use cmMakefile::ScopePushPop to manage variable scopes
also expects policy scopes to be balanced. There is no place that we
use cmMakefile::PolicyPushPop without also using ScopePushPop. Relieve
PolicyPushPop of responsibility for policy scope balance checks by
moving it to ScopePushPop.
The PolicyPushPop constructor arguments and Quiet method were used to
pass non-default arguments to PushPolicy and PopSnapshot, but no clients
use them anymore.
The `PopPolicyBarrier` method is actually responsible for closing any
scope opened by creating a snapshot. Rename it to `PopSnapshot` and add
a comment explaining the purpose of the poilcy-scope-specific part of
the method.
Since commit v3.4.0-rc1~179^2~1 (cmState: Add a VariableScope snapshot
type, 2015-08-23) the snapshot stack may have a VariableScopeType entry.
Skip over these when constructing the call stack, just as we do for
policy scopes. Otherwise we report the command causing the variable
scope to be entered twice (e.g. find_package while loading a package
version file).
In commit v3.4.0-rc1~18^2 (Disable shared library support when compiler
links statically, 2015-09-30) we tried to detect when the compiler is
not capable of linking shared libraries (possibly due to flags in use).
However, the approach is not robust against flags like `-nostdlib`.
Revert it for now pending another solution to the original problem.
This fixes a bug where 64 bit builds with /bigobj incorrectly determined
that the object files were not 64 bit. This manifested itself with
printf type functions showing up as undefined because the leading
underscore was being removed and should not be removed.
Prior to commit v3.4.0-rc1~71^2 (Project: Determine default language
dialect for the compiler, 2015-09-15) we always guessed the default
language standard dialect based on the compiler version. This was not
reliable so that commit switched to computing the default language
standard dialect while detecting the compiler id.
When a toolchain file uses CMakeForceCompiler to set the compiler id
then the detection does not occur. Therefore commit v3.4.0-rc1~54^2
(Project: Don't require computed default dialect if compiler was forced,
2015-09-22) made the lack of detection an error only if the compiler was
not forced. However, this means that projects using CMakeForceCompiler
no longer even get the guess that we had before so <LANG>_COMPILER does
not work.
Due to the sophistication of CMake's compiler detection logic projects
should be ported away from using CMakeForceCompiler. In the meantime,
restore a guess of the default language standard dialect when the
compiler is forced.
Our check of CMAKE_PLATFORM_NO_VERSIONED_SONAME, originally added by
commit v3.1.0-rc1~416^2 (Add basic Android platform module, 2014-06-06),
was dropped accidentally by commit v3.4.0-rc1~250^2~21
(cmGeneratorTarget: Move GetLibraryNames from cmTarget, 2015-08-04).
Restore the check in the new location of the GetLibraryNames method.
In commit v3.4.0-rc1~71^2 (Project: Determine default language dialect
for the compiler, 2015-09-15) we added an "INFO:..." string to the
compiler id detection binary. The value can be optimized out of the
compiler id binary unless we force it to be included by making the
program behavior depend on it at runtime. Add references to the value
as we do for the other info strings already.
Gentoo-Issue: https://bugs.gentoo.org/show_bug.cgi?id=565744
Refactoring in commit v3.4.0-rc1~74^2~1 (MSVC: Rewrite manifest file
handling with Makefile and Ninja, 2015-09-15) broke handling of this
option. Fix it and add a test case.
We already recognize the Cray Fortran compiler id. Extract the version
number using the same predefined macros we already use for Cray C and
C++ compilers.
The moved-from location checks for the existence of whether certain
generators are needed for any target in a loop, rather than whether
the current target has an install rule.
This was introduced in commit v3.4.0-rc1~275^2~3 (cmInstallCommand: Move
the SetHaveInstallRule invocation., 2015-07-28). Move the
SetHaveInstallRule call to a more-obvious place where the
cmInstallTargetGenerator is constructed.
Changes in commit v3.4.0-rc1~124^2~1 (cmFindProgramCommand: Re-implement
search using more flexible approach, 2015-09-01) did not preserve the
behavior of looking for the given name with no search path at all.
Fix this and add a test case covering finding an absolute path with
no search directories.
Brad King
Bill Hoffman
Ty Smith
Stephen Kelly
Andrew Shaitorov