Commit 66b290e7 (export(): Process the export() command at generate
time., 2012-10-06 ) refactored export() so that it could evaluate
strings at generate-time. This was intended for evaluating target
properties, but that commit also removed a check for target
existence at configure-time. Restore that check and add a test for
this case.
5838aba Export: Report error on relative include with genex.
7a3e45b Export: Prefix relative items with genexes in INSTALL_INTERFACE.
f088a32 Export: Process INSTALL_INTERFACE in INCLUDES DESTINATION.
9eedc85 Export: Process relative includes after genex evaluation.
80790f3 Export: Test existing behavior of exporting includes with genexes.
38afc82 target_include_directories: Allow relative path with genex
Diagnostics which check the sanity of exported include paths
previously skipped over any path containing a generator expression.
Introduce a policy to issue an error message in such cases.
The export files created in the OLD behavior are not usable, because
they contain relative paths or paths to the source or build location
which are not suitable for use on installation. CMake will report an
error on import.
In code such as
install(TARGETS ...
INCLUDES DESTINATION $<FOO>include
)
the generator expressions are evaluated at generate-time. Delay
determining whether each entry is a relative path until after
the generator expressions are evaluated. Such relative paths
are based relative to the CMAKE_INSTALL_PREFIX.
4cce44b Help: Document the CMAKE_MAKE_PROGRAM variable in more detail
558c74d VS: Switch to internal CMAKE_MAKE_PROGRAM lookup by generators
5229f2d Tests: Do not use an explicit make program for VS generators
72dd738 Tests: Fix MFC test heuristic for empty CMAKE_TEST_MAKEPROGRAM
fd6076d Tests: Pass CMAKE_MAKE_PROGRAM instead of --build-makeprogram
68031ab Tests: Configure SubProject-Stage2 test more robustly
003d10c Tests: Simplify VSExcludeFromDefaultBuild configuration
e47d934 Tests: Simplify VSProjectInSubdir configuration
e965cb1 Tests: Simplify CTest.BuildCommand.ProjectInSubdir configuration
72bf255 Tests: Pass --build-options to every test
4d1d772 ctest: Teach --build-options to allow zero options
96966b5 ctest: Make the --build-makeprogram optional for --build-and-test
91a0211 Simplify some calls to cmGlobalGenerator::Build
123a060 Teach GenerateBuildCommand to find its own make program
5f5c92b VS: Add internal APIs to find MSBuild, devenv/VCExpress, and msdev
4ac75fd Prefer CMAKE_MAKE_PROGRAM over CMAKE_BUILD_TOOL (#14548)
...
Teach add_custom_target to check the policy too. Extend the policy to
disallow reserved target names that we use for builtin targets like
"all".
Extend the RunCMake.CMP0037 test to cover these cases.
Historically these were both added for the Makefile and Visual Studio
generators, respectively. Later the VS generators started using the
CMAKE_MAKE_PROGRAM cache entry to find the IDE build tool, and the
CMAKE_BUILD_TOOL was simply set as an alias.
Fix the documentation to explain that CMAKE_MAKE_PROGRAM is the modern
variable and that CMAKE_BUILD_TOOL is the compatibility alias, not the
other way around. Replace uses of CMAKE_BUILD_TOOL with
CMAKE_MAKE_PROGRAM in CMake-provided modules. Nothing needs to lookup
CMAKE_BUILD_TOOL in the cache, so simply set it as a normal variable.
Re-implement the build_command() command to use "cmake --build" instead
of generating a native build tool invocation directly. This command
will internally invoke the proper native build tool.
This avoids requiring cmGlobalGenerator::GenerateBuildCommand to produce
a string so that it can be later refactored to produce a vector with no
quoting or escaping. It will also allow us to later teach CMake to
delay the decision about which build tool to invoke until after the
project build system is generated to disk. For example, on Visual
Studio 10 and above the preferred command-line tool is MSBuild, but we
need to fall back to devenv if the .sln has Intel Fortran .vfproj files.
bf755c7 set: Add unit tests for set/unset PARENT_SCOPE
bc280f1 set: Fix handling of empty value with PARENT_SCOPE
20afbd5 set: Handle value-less PARENT_SCOPE explicitly
Since commit 7d47c693 (Drop compatibility with CMake < 2.4, 2013-10-08)
we no longer need to use the configure_file IMMEDIATE option to support
compatibility modes less than 2.0.
Allow ExternalData_URL_TEMPLATES to be empty if a value for
ExternalData_OBJECT_STORES is provided. Assume in this use case that
the object stores will already contain all needed objects. Extend the
Module.ExternalData test to cover this case (all objects in stores).
Extend the RunCMake.ExternalData test to cover the non-failure message
case when stores are provided without URL templates.
Split the test cases covering spaces and colons into separate units.
Run the space cases everywhere. Disable the colon cases where they
are known to fail. This approach increases platform coverage for the
test and makes the known-failure logic as local as possible.
No Makefile generator on Windows can generate targets with ':'
in their name because the CMakeFiles/<target>.dir directory cannot
be created. Skip this part of the test on all Windows Make tools.
Commit 239b0c6b (Don't add invalid content to static lib
INTERFACE_LINK_LIBRARIES., 2013-10-20) extended a test which
excercised the logic of the LINK_ONLY generator expression.
Commit ef10b87c (CMP0022: Plain target_link_libraries must populate
link interface, 2013-11-02) removed the instance of LINK_ONLY which
was excercised by that test.
Add a new test which excercises the other instance of LINK_ONLY by
setting the CMP0022 policy to NEW and consuming the contents of the
INTERFACE_LINK_LIBRARIES target property.
Resolve conflict in Source/cmTarget.cxx by integrating the changes to
the internal copy constructor from both sides. Also resolve a logical
conflict by dropping the special case for INTERFACE_LIBRARY targets.
Since cmTarget::SetMakefile already forces CMP0022 to NEW for such
targets we need no special handling.
Resolve conflict in Source/cmTargetLinkLibrariesCommand.h by dropping
the documentation change. We will make the same change in the new
location of the same documentation in a separate commit.
Resolve conflicts in
Tests/RunCMake/CMP0022/CMP0022-WARN-empty-old-stderr.txt
Tests/RunCMake/CMP0022/RunCMakeTest.cmake
by taking the side from the 'policy-CMP0022-fixes' branch.
Since cmTarget::ComputeLinkInterface is called separately for each
"head" target that links a target, the warning we produce when
CMP0022 is not set could be repeated. Add explicit logic to allow
the warning to appear at most once. Multiple copies of the warning
for the same target are almost always identical and therefore
redundant. In the rare case that two copies of the warning are
different, the second can appear in a future run after the first
is fixed.
When cmExportFileGenerator::SetImportLinkInterface exports the old
LINK_INTERFACE_LIBRARIES property values it skips doing so for
non-linkable targets because target->GetLinkInterface returns NULL for
such targets.
Since cmExportFileGenerator::PopulateInterfaceLinkLibrariesProperty
looks at the INTERFACE_LINK_LIBRARIES property directly instead of using
the computed link interface, teach it to skip exporting the property if
target->IsLinkable returns false.
Extend the RunCMake.CMP0022 test with a case covering this. Simply
export an executable target that links to a library that is not
exported.
The CMP0022 NEW behavior is that the INTERFACE_LINK_LIBRARIES property
exactly defines the link interface. The plain target_link_libraries
signature says linking is transitive by default, so it should populate
the property.
Teach the target_link_libraries plain signature to populate the
INTERFACE_LINK_LIBRARIES regardless of the CMP0022 setting. Refactor
the cmTarget::ComputeLinkInterface checks that warn when the policy is
not set to compare the new property to either the explicitly set old
link interface properties or the link implementation fallback for all
linkable target types, not just static libraries.
This fixes a regression in 2.8.12.0 that caused target_link_libraries to
not implement transitive linking in the plain signature once the policy
CMP0022 is set to NEW.
Teach cmGlobalGenerator::CheckTargets to include OBJECT_LIBRARY targets
in the check for source file existence.
Extend the RunCMake.ObjectLibrary test to cover this case.
Teach the cmGeneratorExpressionEvaluator filesystem artifact logic
to reject OBJECT_LIBRARY targets since they have no main artifact.
Without the explicit rejection evaluation falls through to an
internal CMake error message in cmTarget::GetOutputInfo.
Extend the RunCMake.GeneratorExpression test to cover these cases.
In certain scenarios, it is preferable to keep a 'dirty' install prefix
than to clear it, and to expect that content will not be found there.
Add a CMAKE_FIND_NO_INSTALL_PREFIX variable that can be set to disable
searching the install prefix.
Bracket arguments recorded in command invocations inside foreach,
function, and macro blocks should not have any replacements done when
the arguments are replayed later. Teach the RunCMake.Syntax test to
cover these cases.
Add an Escape1 test case covering \-escape cases inside bracket, quoted,
and unquoted arguments. Also cover comments immediately after quoted
and unquoted arguments on lines containing \# escapes.
3e04946 Require CMAKE_<LANG>_COMPILER to be found as a full path
6007f7c CMakeDetermineCompilerId: Always use compiler detected from IDE
332771c CMakeDetermine*Compiler: Remove temporary cache entry
All generators now support detection of the full path to the compiler, so
require it to be so. This will allow CMake<LANG>Information.cmake and
other logic to assume the full path to the compiler tool is available.
The Makefile generators already rejected CMAKE_<LANG>_COMPILER values
that did not name an existing compiler. Extend this error message to
all generators, make it occur as early as possible, and improve the
message with advice about how to set the compiler. If the full path to
the compiler is not known, finish enabling languages with a fatal error
so configuration does not continue.
For now, allow the RC language compiler to not be a full path. Later we
will need to detect the full path to "rc" under the VS IDE.
Add a RunCMake.CompilerNotFound test to cover failure cases.
Fix the RunCMake.CompilerChange test EmptyCompiler case to work
when configuration does not continue past enable_language.
Only valid target names or generator expressions may appear in
the target field of a LINK_ONLY expression.
Other content like link flags should still be added to that property
(wrapped in config-specific generator expressions), but not wrapped
in LINK_ONLY. Otherwise undue warnings would be issued for the
policy CMP0022.
The LINK_ONLY expression only has an effect for actual target
names anyway, so there is no logical deficit.
Set the minimum required version of CMake high enough to avoid the
warning for CMAKE_LEGACY_CYGWIN_WIN32. The warning appears on stderr
and breaks the expected output matching.
It was a subtest of the RunCMake.ObjectLibrary test. However, we need
to test a build with ExternalProject after running CMake, which RunCMake tests
do not do.
Teach the CMake language lexer to treat the \-LF pair terminating a
line ending in an odd number of backslashes inside a quoted argument
as a continuation. Drop the pair from the returned quoted argument
token text. This will allow long lines inside quoted argument
strings to be divided across multiple lines in the source file.
It will also allow quoted argument text to start on the line after
the opening quote. For example, the code:
set(x "\
...")
sets variable "x" to the value "..." with no opening newline.
Previously an odd number of backslashes at the end of a line inside
a quoted argument would put a \-LF pair (or a \-CR pair) literally
in the argument. Then the command-argument evaluator would complain
that the \-escape sequence is invalid. Therefore this syntax is
available to use without changing behavior of valid existing code.
Teach the RunCMake.Syntax test to cover cases of quoted arguments
with lines ending in \, \\, and \\\. Odd counts are continuations.
Drop all behavior activated by setting CMAKE_BACKWARDS_COMPATIBILITY to
a value lower than 2.4, and generate an error when projects or the user
attempt to do so. In the error suggest using a CMake 2.8.x release.
Teach cmake_minimum_required to warn about projects that do not require
at least CMake 2.4. They are not supported by CMake >= 3.0.
Replace the documentation of CMAKE_BACKWARDS_COMPATIBILITY with a
reference to policy CMP0001.
Other warnings for the same policy already have similar output since
commit 81d2793e (Add differing target property content to policy CMP0022
warning, 2013-09-11).
Add cmCommand::Disallowed helper to check the setting of a policy that
disallows the command. Add a RunCMake.DisallowedCommands test
placeholder. Add a Help/policy/DISALLOWED_COMMAND.txt file for
inclusion by each policy document to avoid duplication of the common
text.
Add a string(CONCAT) command to simply concatenate input arguments
together. This will be useful for combining strings from different
quoting syntaxes. Add a RunCMake.string test covering these cases.
Introduce a policy to control the behavior.
The AliasTargets unit test already tests that using a
double-semicolon in the name is not an error. Change the ExportImport
test to use a namespace with a double-semicolon too.
daa0f6f Add Lua-style long brackets and long comments to CMake language
a8c6523 cmListFileLexer: Convert CRLF -> LF newlines explicitly
dbd9333 cmListFileLexer: Allow a leading UTF-8 Byte-Order-Mark (#11137)
5645783 cmListFileLexer: Allow command names with one letter (#14181)
Set the minimum required version of CMake high enough to avoid the
warning for CMAKE_LEGACY_CYGWIN_WIN32. The warning appears on stderr
and breaks the expected output matching.
Teach the CMake language parser to recognize Lua-style "long bracket"
arguments. These start with two '[' separated by zero or more '='
characters e.g. "[[" or "[=[" or "[==[". They end with two ']'
separated by the same number of '=' as the opening bracket. There is no
nesting of brackets of the same level (number of '='). No escapes,
variable expansion, or other processing is performed on the content
between such brackets so they always represent exactly one argument.
Also teach CMake to parse and ignore "long comment" syntax. A long
comment starts with "#" immediately followed by an opening long bracket.
It ends at the matching close long bracket.
Teach the RunCMake.Syntax test to cover long bracket and long comment
cases.
Read input files in binary mode instead of text mode and convert CRLF
newlines to LF newlines explicitly in our own buffer. This is necessary
to read CMake source files with CRLF newlines on platforms whose C
runtime libraries do not transform newlines in text mode. For example,
a Cygwin or Linux binary may not transform CRLF -> LF in files read from
a Windows filesystem. Perform the conversion ourselves to ensure that
multi-line string literals in CMake source files have LF newlines
everywhere.
Teach the lexer to read a UTF-8, UTF-16 BE/LE, or UTF-32 BE/LE
Byte-Order-Mark from the start of a file if any is present. Report an
error on files using UTF-16 or UTF-32 and accept a UTF-8 or missing BOM.
Teach the lexer to treat a single letter as an identifier instead of an
unquoted argument. Outside of a command invocation, the parser treats
an identifier as a command name and an unquoted argument as an error.
Inside of a command invocation, the parser treats an identifier as an
unquoted argument. Therefore this change to the lexer will make what
was previously an error case work with no other behavioral change.
The final location and name of a build-target is not determined
until generate-time. However, reading the LOCATION property from
a target is currently allowed at configure time. Apart from creating
possibly-erroneous results, this has an impact on the implementation
of cmake itself, and prevents some major cleanups from being made.
Disallow reading LOCATION from build-targets with a policy. Port some
existing uses of it in CMake itself to use the TARGET_FILE generator
expression.
Make the API for adding targets string based so that it can easily
use cmGeneratorTarget.
Teach the cmIncludeCommand to generate the exported file at
configure-time instead if it is to be include()d.
The RunCMake.ExportWithoutLanguage test now needs a dummy header.h
file as expected error from export() is now reported after the
missing file error.
f973737 GenerateExportHeader: Port to use message(DEPRECATION)
f69606d Qt4Macros: Port to use message(DEPRECATION)
509c142 message: Add a DEPRECATION mode
1763c31 Set policy CMP0025 to NEW while building CMake itself
aa53ee5 Add policy CMP0025 for Apple Clang compiler id compatibility
ab65862 Clang: Add separate "AppleClang" compiler id
By default, the message is not issued. If CMAKE_ERROR_DEPRECATED
is on, the message is fatal. If CMAKE_WARN_DEPRECATED is on, the
message is a warning.
Previously if headers required to check if a struct has a member can be
compiled with C++ compiler only, the check would fail because the C
compiler fails. As a consequence, the result variable would be set to
false, even if the struct has that particular member.
Teach CHECK_STRUCT_HAS_MEMBER to accept a new optional argument LANGUAGE
that allows one to explicitly set the compiler to use. The new
signature is therefore:
CHECK_STRUCT_HAS_MEMBER (<struct> <member> <header> <variable>
[LANGUAGE <language>])
CMake is aware of the policy's NEW behavior and the AppleClang compiler
id. Set the policy to NEW explicitly to avoid the warning and get the
NEW behavior.
Also teach the RunCMake test infrastructure to build tests with
-DCMAKE_POLICY_DEFAULT_CMP0025=NEW to avoid the policy warning
in test output that must match specific regular expressions.
Currently, export() is executed at configure-time.
One problem with this is that certain exported properties like
the link interface may not be complete at the point the export() is
encountered leading to an incorrect or incomplete exported
representation. Additionally, the generated IMPORTED_LOCATION
property may even be incorrect if commands following the export()
have an effect on it.
Another problem is that it requires the C++ implementation of cmake
to be capable of computing the exported information at configure time.
This is a limitation on the cleanup and maintenance of the code. At
some point in the future, this limitation will be dropped and more
implementation will be moved from cmTarget to cmGeneratorTarget.
This target type only contains INTERFACE_* properties, so it can be
used as a structural node. The target-specific commands enforce
that they may only be used with the INTERFACE keyword when used
with INTERFACE_LIBRARY targets. The old-style target properties
matching LINK_INTERFACE_LIBRARIES_<CONFIG> are always ignored for
this target type.
The name of the INTERFACE_LIBRARY must match a validity generator
expression. The validity is similar to that of an ALIAS target,
but with the additional restriction that it may not contain
double colons. Double colons will carry the meaning of IMPORTED
or ALIAS targets in CMake 2.8.13.
An ALIAS target may be created for an INTERFACE library.
At this point it can not be exported and does not appear in the
buildsystem and project files are not created for them. That may
be added as a feature in a later commit.
The generators need some changes to handle the INTERFACE_LIBRARY
targets returned by cmComputeLinkInterface::GetItems. The Ninja
generator does not use that API, so it doesn't require changes
related to that.
When RunCMake tests run during dynamic analysis, valgrind may add lines
of the form "==[0-9]+==..." to the output. Remove such lines from the
actual output before matching it against the expected output.
Stephen Kelly
Brad King
Nils Gladitz
Daniele E. Domenichelli