`WIN32` checks whether current build context is windows. This is not correct in case of cross-compiling, because we are interested in finding host's tools, not SDK's tools.
* replace deprecated calls to FetchContent_Populate
The single argument signature for FetchContent_Populate is deprecated with CMake 3.30.
It was used, in order to call add_subdirectory manually with the EXCLUDE_FROM_ALL and SYSTEM flags.
These have been added to FetchContent_Declare with 3.25 and 3.28.
Calling FetchContent_MakeAvailable will internally call add_subdirectory with EXCLUDE_FROM_ALL and SYSTEM.
There is therefore no need to call this manually.
* fix: OPTIONS passed to CPMAddPackage not set
where previously parsed in cpm_add_subdirectory which is not called
on the new code path.
* refactor: remove an unnecessary else branch
* ci: include cmake 3.30 in test matrix
* fix: forward SOURCE_SUBDIR to FetchContent_Declare
For CMake version <3.28 this is done by calling add_subdirectory manually.
For newer version FetchContent_Declare/MakeAvailable handles this for us.
* fix: only set options if download_only is false
this replicates the old behaviour
* fix: DOWNLOAD_ONLY test
* refactor: always use *_Populate to reduce code paths
* Revert "refactor: always use *_Populate to reduce code paths"
This reverts commit 0e8ca2a0e9.
---------
Co-authored-by: Avus <48911667+Avus@users.noreply.github.com>
* Adding PATCHES keyword.
* Formatting fix.
* Move cpm_add_patches() outside if/else scopes.
* cmake-format: add PATCHES to CPMAddPackage.
* Integration tests for PATCHES command.
* Use get_filename_component() in place of cmake_path() for use with all cmake versions 3.14 and above.
* Added an example and improved comment for cpm_add_patches.
* Custom cache directory name
Enable setting a custom directory name for cached packages.
Required if e.g. patch_command uses external variables or
just to set a human readable name.
* fixup! review comment CUSTOM_CACHE_DIR -> CUSTOM_CACHE_KEY
* run cmake-format
---------
Co-authored-by: Olsson Gerhard <gerhard.olsson@volvo.com>
Co-authored-by: Lars Melchior <lars.melchior@gmail.com>
If I have a project, which has:
1. /CMakeLists.txt:
```cmake
cmake_minimum_required(VERSION 3.29)
project(test)
include(./cmake/CPM.cmake)
add_subdirectory(subdir)
cpmfindpackage(
NAME
nlohmann_json
VERSION
3.11.2
URL
"https://github.com/nlohmann/json/releases/download/v3.11.2/json.tar.xz"
EXCLUDE_FROM_ALL
ON)
if(NOT TARGET nlohmann_json::nlohmann_json)
message(FATAL_ERROR "BUG")
endif()
```
2. /subdir/CMakeLists.txt
```cmake
cpmfindpackage(
NAME
nlohmann_json
VERSION
3.11.2
URL
"https://github.com/nlohmann/json/releases/download/v3.11.2/json.tar.xz"
EXCLUDE_FROM_ALL
ON)
```
When nlohmann_json is first founded by find_package,
CPM.cmake register this package and later call to CPMFindPackage will
not actually perform a find_package again, which leads to this buggy
behavior.
As CPM write a FindXXX.cmake file for packages added by CPM to pervert
find_package to get local package, I think directly use cpm_find_package
to check if a package is already added is good enough.
* Treat relative paths as relative to parent project's remote
* replace unit test case with integration test
* run relative URL test only on supported CMake versions (3.27)
* omit cmake 2.25.x test runner as replaced by 2.27.x
---------
Co-authored-by: Chris Wright <chris.wright@mqa.co.uk>
Co-authored-by: Lars Melchior <lars.melchior@gmail.com>
Add [SPDX][1] licensing/copyright tags to the `get_cpm.cmake` script,
so that anybody that copies the `get_cpm.cmake` script into their repo
(following the instructions in the README.md)
will automatically comply with CPM's MIT license.
[1]: https://spdx.dev/
* add system property for cpm_add_subdirectory
* add test case for system property
* lock CMake version in test workflow
* refactor to make SYSTEM an extra config option and update tests
* remove old comment change
* use consistent CMake version and extension for all workflows
* make warning more specific and try to trigger on windows
* another attempt to trigger warning on MSVC
* update readme
* simplify test case and use git tag
* add SYSTEM option to .cmake-format
* forward system arg for source overrides
* enable system implicitly for the single argument syntax
* Use SYSTEM option for FetchContent and add_subdirectory (#441)
* Use SYSTEM option for FetchContent and add_subdirectory
* Add SYSTEM option to syntax and doku
* Update CPM.cmake
* Update .cmake-format
---------
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
---------
Co-authored-by: Claus Klein <claus.klein@arcormail.de>
* Optimized multi-threading.
* Remove old multi-threading implementation.
* Release files instead of removing.
* Lock before choosing to check cache or download.
* Unlock directly after verifying directory exists.
* Add file locking to support parallel runs.
* Fixed formatting.
* Prevent double locking file.
* Fix SegFault from test 2.
* Remove unnecessary debugging messages.
* Lock the package directory rather than the cache directory.
Only synchronize if CPM_SOURCE_CACHE is defined.
* Lock the version specific cache entry rather than the package specific entry.
* Remove unnecessary arguments in conditional statements.
* Change back to locking entire cache directory.
* Only check CPM_HAS_CACHE_LOCK.
* Lock on a per-package basis rather than the entire cache.
* Clean up the locked file.
* Unlock then remove to fix Windows.
* Specify use of cmake.lock as the lock file.
* - Changed CPM_HAS_CACHE_LOCK to ${CPM_ARGS_NAME}_CPM_HAS_CACHE_LOCK.
- Removed redundant variable initialization.
* Add unit test.
* Actually test if resulting git cache is clean in unit test.
* - Added comments
- Fixed formatting
- Removed unnecessary imports
* convert parallelism test to integration test
* remove comment
* - Removed now unnecessary variable.
- Only delete file instead of unlocking it then deleting it.
* Forgot to change variable name.
* Add similar changes to the missed section.
* Fixed formatting.
* Unlock the file, but do not delete it.
* Only unlock the file if it exists.
* Changed cache.cmake test to ignore non-directory entries.
* Integration test lib make_project:
* keyword args
* 'name' arg to allow multiple projects from the same test
* - Moved checks to function.
- Fixed small grammatical errors.
* - Fix formatting
* Switch to snake case.
---------
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
Co-authored-by: Lars Melchior <lars.melchior@gmail.com>
Co-authored-by: Borislav Stanimirov <b.stanimirov@abv.bg>
* get_cpm: Recover from failed download
If the download fails during the cmake config step
an empty `CPM_<version>.cmake` file is created.
So far the script only checks the files existence,
so there is no second try until the empty file is deleted manually.
This adds a check if the file is empty
and resumes the download on next config if it is.
* More descriptive function name
* Small helper function for logging :
This PR add an helper function only when `message(STATUS ...)` as the other are `backtracing` and adding the helper function add on stack.
It should fix#328#342 with a few line of code.
* Formatting
* Expand absolute path later in get_cpm.cmake
Supports expanding tildes in the environment.
* Apply cmake-format to get_cpm.cmake
* Use .cmake-format file
* Timestamps for URL downloads match the download time
By enabling CMake policy 135 we ensure that extracted files
timestamp match that of the download time, instead of when the
archive is created. This makes sure that if the URL changes to
an older version we still rebuild everything as the timestamp
stays newer.
* Introduce CPM_SET_RECOMMENDED_CMAKE_POLICIES
Enabling CPM_SET_RECOMMENDED_CMAKE_POLICIES will establish defaults
for all CMake policies so that both CPM and added projects
operate inline with CPM recommended best practices.
* Fix style issues found by ci
* Update README.md
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
* Update cmake/CPM.cmake
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
* Set policies when CPM.cmake is included
* CPM_SET_RECOMMENDED_CMAKE_POLICIES default is ON
* Correct failing CI tests
* CPM.cmake always sets policies to on
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
* CPM_DIRECTORY always stores absolute non symlink paths
When CPM is brought into a project via downloading it will be placed
in the build directory. When the build directory is a symlink the
CPM_DIRECTORY value will hold the symlink value. This means that
trying to run CMake via the real build directory will cause CPM
to fail since it considers the current CMAKE_CURRENT_LIST_DIR
to be different and will early exit.
This resolves the issue be always getting CMake to resolve symlinks
before comparing paths for CPM_DIRECTORY.
* Refactor logic based on PR review
* Fixed: Deleted SOURCE_DIR directory would abort just after git stash save --quiet;--include-untracked
* Fixed: Review comments
* Added: Integration test for deleted SOURCE_DIR with FetchContent
* Fixed: Review comments
* Fixed: Review comments
* Override FetchContent paramers, fixes#281
* add check to verify that dependency is cached
* update test comment
* rename test file and class
* update test indentation
* Add per project CPM_DOWNLOAD controls
Introduces support for `CPM_<PACKAGE>_DOWNLOAD` variable ( and env )
which allows finer grained control.
* Address PR feedback on API names
By enabling CMP0126 we tell CMake that when constructing a CACHE variable
with the same name as a local variable to not unset the local variable.
By enabling CMPO126 it makes sure that CPM behavior around `OPTION` is
consistent if the option is created with `set(CACHE)` or with `option`.
* Added directory inside cache to make it shown more friendly in CLion
* Changed hash generation to omit collision with old cache scheme. Removed moving old directory for compatibility reason.
* Added CPM_USE_NAMED_CACHE_DIRECTORIES option
* Fixed formatting
* Added description of CPM_USE_NAMED_CACHE_DIRECTORIES into README.md
Co-authored-by: Paweł Gorgoń <pgorgon@hem-e.com>
* Check that the cache working directory is clean
* cpm_check_working_dir_is_clean for non git repo will not terminate cmake
* sileneced check for git repo, in case of error
* style(CPM.cmake): fixed indentation for ERROR_QUIET
* refactor(CPM.cmake): added logic to handle a cache folder inside a (not correlated) git repo
this is accomplished by checking if we are in the directory where .git lives
* style(CPM.cmake): stray tab to spaces
* test(dirty-cache-check): added unit test to check cpm_check_working_dir_is_clean
the test creates a file in a folder and then uses git to create a repo in that folder, to check various conditions
* Update test/unit/dirty-cache-check.cmake
added user.name and user.email to comply with test machine where git is not completely configured
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
* feat(cpm_check_working_dir_is_clean): now it takes a git tag (works with hashes too) to check for uncommitted and commited changes
* refactor(cpm_check_working_dir_is_clean): early return for the most common case of uncommited changes in a repo
* refactor(cpm_check_git_working_dir_is_clean): removed edgecase detection (a non-git folder in an unrelated git repo)
now the contract is that com_check_git_working_dir_is_clean is given the base folder of a git repo
Co-authored-by: Andrew Gribble <ag131012@renishaw.com>
Co-authored-by: Lars Melchior <TheLartians@users.noreply.github.com>
Add fetchcontent_dependency unit test:
* This test should highlight the fact that cpm_add_subdirectory is always called, even when
cpm_fetch_package isn't populating the dependency
* NO_CACHE YES highlight a bug introduced in 32b063eba5 where
cpm_fetch_package was checking undefined ${lower_case_name}_POPULATED variable
https://github.com/cpm-cmake/CPM.cmake/issues/287
