Reply file names embed the configuration name verbatim. Reconfiguring
a build tree with a build type that differs from the previous one only
in case (e.g. Debug -> debug) makes CMake write a reply whose name also
differs only in case from the existing file. On a case-insensitive
filesystem the write is skipped because the name already exists, but
RemoveOldReplyFiles compared on-disk names to the just-written names
textually and deleted the surviving file, leaving the reply index
citing a target or directory reply that no longer exists on disk.
Prune reply files by file identity via cmSystemTools::GetFileId instead
of by name, so an on-disk entry that aliases a reply we just wrote is
kept. An entry whose identity cannot be obtained is retained rather
than deleted.
Fixes: #28022
As the number of areas in CMake for which we've provided JSON schemas
has expanded, testing infrastructure has mostly been copy-pasted across
the relevant test suites. Unify it to a single reusable validator
script and CMake helper.
Add a regression test for the File API codemodel graph produced for an
import std target. The test verifies that emitted target references,
including those to the CMake-provided stdlib target, resolve to targets
present in the codemodel.
Closes: #27890
Previously, if a single source file was added to multiple file sets via
multiple `target_sources` calls, only the first call originally adding
it to the target was stored in the backtrace graph and referenced by
index in the singular "backtrace" field.
Add a "backtraces" field expanding this to cover all command
invocations which associate the given source file with that target.
Preserve the original "backtrace" field for compatibility. Update
spellings in the test suite to check against both fields.
Issue: #27035
The `fileSetIndex` field of target sources replies is incomplete, as a
single source file can be added to arbitrarily many file sets. Add a
`fileSetIndexes` field to report all file sets to which a source
belongs, and preserve `fileSetIndex` for compatibility as the index of
the most recent addition.
Tweak the test suite to use the plural spelling only in expected JSON
snippets and verify that the singular `fileSetIndex` is always the last
element of the new list. (Note that remaining occurrences of the
singular spelling correspond to "installers" objects, which is entirely
different.)
Fixes: #27035
`LINK_LIBRARIES` entries guarded by `$<COMPILE_ONLY:...>` are not
linked, but must participate in target ordering so files generated by
compile-only dependencies, like headers, are available before
compilation.
Fixes: #27792
CMake 4.1 and earlier did not issue an error if an imported shared library target
was missing an IMPORTED_IMPLIB property and nothing used that imported
library. There was no code path checking for the CMP0111 NEW behavior. Since
b626843d71 (fileAPI: Output all INTERFACE and IMPORTED targets, 2025-09-13),
we now include all imported targets in the file API replies, and that does trigger
that check. We need to tolerate such imported targets to preserve backward
compatibility, and to avoid issuing errors for problems in targets likely to be
coming from outside the project and beyond the developer's control.
Fixes: #27496
71a4e34d97 fileAPI: Expose CMAKE_<LANG>_COMPILER_ARG1
1e02926c9a fileAPI: Refactor toolchains schema to prepare for new version
3caa572c05 fileAPI: Output more info for test failures
Acked-by: Kitware Robot <kwrobot@kitware.com>
Merge-request: !11419
Compiler arguments coming from CC environment variables or multi-element
CMAKE_<LANG>_COMPILER variables set by toolchain files were previously not
exposed in the file API. Among other possible problems, this caused clients
to determine wrong system include paths and built-in preprocessor macros by
calling the compiler without these important arguments.
This is fixed by adding an optional "commandFragment" attribute to the
compiler description in the `toolchains` object, containing these arguments
as a command line fragment. This is already the form in which they are
internally stored in the CMAKE_<LANG>_COMPILER_ARG1 variable, so all that is
required is adding this variable to the set of exported variables, besides
some logic to omit it if empty.
Issue: #22568
Remove literal text "subcommand" from error messages. This addresses the
other use of this and improves consistency (see preceding commit). Also,
omit some unnecessary articles, which also improves consistency.
This adds the following new arrays, which together capture all direct
dependencies and interface dependencies of a target:
- linkLibraries
- interfaceLinkLibraries
- compileDependencies
- interfaceCompileDependencies
- objectDependencies
- orderDependencies
Fixes: #21995, #25213
Commit 62a1d3e7f1 (FileAPI: Add symbolic property to targets, 2025-09-30)
contained an error in added test code where a wrong library was linked.
The naming of the targets involved make clear the intended linking
relationship, but nothing in the test was actually testing or relying on
that relationship so it went undetected.
When targeting the MSVC ABI, define `_MBCS` by default if the project
does not define `_SBCS` or `_UNICODE`. Visual Studio has long defined
one of the three character set macros automatically. For consistency,
define it when compiling for the MSVC ABI with other generators.
Add policy CMP0204 for compatibility.
Fixes: #27275
Visual Studio defines this automatically for `.dll` targets.
For consistency, define it when compiling for the MSVC ABI
with other generators. Add policy CMP0203 for compatibility.
Fixes: #27253
This also removes the JsonCpp_VERSION_STRING result variable, which
doesn't need to be deprecated and further provided by the module as this
is internal module.
Issue: #27088
Since it can be set in the environment when running the test suite,
guard tests which are sensitive to intermediate directory strategy
changes with explicit settings.
This will allow JSON schemas for these two types of files to describe the
version-specific content without requiring any outside information.
Fixes: #27031
Record `find_` command events in the configure log, except
`find_package` as it is far more complicated (and will have its own
event kind).
Note that testing only generates the events of interest, there is no
verification. Also note that testing that the "found" to "notfound"
transition causes an event is not testable because a truthy value in the
variable skips any kind of verification or other logic beyond
normalization.
Co-Authored-by: Ryan Krattiger <ryan.krattiger@kitware.com>
See: #24833
On modern macOS, compiler wrappers like `/usr/bin/cc` automatically
choose a SDK to pass via `-isysroot` to an underlying compiler from
Xcode or the CommandLineTools. Other toolchains like Homebrew's `gcc-*`
come with a default SDK too. Therefore, when targeting macOS, we no
longer need to choose any SDK or pass an `-isysroot` flag by default.
Update initialization of `CMAKE_OSX_SYSROOT` to be empty by default when
targeting macOS.
Fixes: #19180
