using git without blobs in the object database

Use Case: A large set of large binary files, geographically distributed. Each location has some unique files, some identical files and some slightly modified files.

I want to use git to tell me what changed and when. But I cannot afford to have it store the blobs in the object database, nor do I need to; knowing the signature of previous objects is good enough.

It seems to me that most operations should work without these objects, and that some people do so. For instance, git-update-cache has a --cacheinfo option that facilitates this operation.

But not all commands work so well. For instance, git-write-tree will fail if --cacheinfo was used to add a file. This failure is caused by a call to check_valid_sha1(). It appears that this check is not strictly necessary, just very useful for normal operation.

Would a patch that added a flag "--disable-sha1-check" to git-write-tree be accepted? I hope that's all I need to add, but I haven't completed my evaluation yet.

thanks, Bryan