Allowing CPM to (further) facilitate vendoring of libraries

[studoot]

As the title says, I'm looking to use CPM to vender 3rd party libraries. Placing CPM_SOURCE_CACHE within my application's repository gets me 95% of the way there, but there's one little speed hump that I'd like to eliminate if possible.

Currently, when CPM clones a Git repository into CPM_SOURCE_CACHE, the repository (i.e. the .git directory) is retained - that's just what the underlying CMake functions do. This is an issue when adding the library to my application's repository, as shown by the error message when git adding it:

hint: You've added another git repository inside your current repository.
hint: Clones of the outer repository will not contain the contents of
hint: the embedded repository and will not know how to obtain it.
hint: If you meant to add a submodule, use:
hint:
hint:   git submodule add <url> externals/cpm/catch2/5b0288370c9df775ca7b6073320e23697b91be84
hint:
hint: If you added this path by mistake, you can remove it from the
hint: index with:
hint:
hint:   git rm --cached externals/cpm/catch2/5b0288370c9df775ca7b6073320e23697b91be84
hint:
hint: See "git help submodule" for more information.

Obviously, I can manually remove the .git directory, but I've been thinking about how achieving this might be incorporated into CPM. I've thought of a few mechanisms that could be used to achieve this:

1. Remove the .git directory after FetchContent_Populate has been used to clone the repository

   This is just what I do manually and is pretty simple to accomplish - I've got a PoC of this approach, which removes git repo directories when a variable (CPM_GIT_REMOVE) is set. It just feels... inelegant.

2. Use git subtree or submodule to incorporate the library's repository into my application's repository.

   This has the advantage of retaining where the library came from. However, it modifies the application's repository by running CMake, which doesn't feel like the right thing to do.

3. Clone the repository to a temp. directory and use git archive to copy out the desired commit's files

   The canonical way to do this would be to do something like:

   git -C repo-dir archive --format=tar some-tag | tar -x -C tree-dir
   

   However, given Windows, you'd probably want to archive to a file and then use CMake to extract the archive to the desired directory. This is starting to feel a bit like too many moving parts to me...

4. update parser version #3, but use git checkout-index instead of git archive

   Forget about update parser version #3 - checkout-index does what's needed directly.

I guess the question is - is this something that could go in CPM? I'm happy enough to come up with a PR to implement, but it's more about whether this is something you'd be happy having in CPM.

[TheLartians]

If I understand the use-case correctly, you essentially want to create a directory containing all dependencies without their according version control data and commit that into a monorepo?

In general I'm not sure if this fits into the scope of CPM.cmake, especially as we are generally interested in keeping the version control information, for example to implement #274. Adding such options would increase the complexity of similar additional features.

Imo I think a better approach would be to write a short script to remove any version control directories inside the local CPM_SOURCE_CACHE that runs after the main CMake routines are invoked.

[studoot]

Given

we are generally interested in keeping the version control information

, I'd have to concur.