Re: [PATCH 4/6] Add check_repo_format check for all major operations.

Junio C Hamano <junkio@cox.net> writes:

Let's back up a bit. I'll deal with only low-level commands here.

First, the easiest group. The following commands do not look at the repository (.git directory) at all.

    check-ref-format get-tar-commit-id git index-pack mailinfo
    mailsplit patch-id shell show-index stripspace verify-pack

We do not need to do anything special about them.

The following three use enter_repo() to the given path (either from the end user at the command line or over the network):

    daemon receive-pack upload-pack

The repo format check should be done at the same place as we make sure enter_repo() finds the given directory a satisfactory path. That is, just before putenv(GIT_DIR=.) in enter_repo().

The following use setup_git_directory(), which chdir()s to the toplevel unless GIT_DIR is set.

    cat-file config-set diff-files diff-index diff-stages diff-tree
    ls-files name-rev rev-list rev-parse show-branch symbolic-ref
    update-index update-ref

Maybe we can have a thin wrapper around setup_git_directory() and after it returns check "${GIT_DIR-.git}" for repository format mismatch. What to do when GIT_DIR is set? Then we can just use it to read the config from "$GIT_DIR/config" and check the version.

The following commands implicitly assume that they are either run from the toplevel or GIT_DIR environment tells them where the .git/ directory is:

    apply checkout-index clone-pack commit-tree convert-objects
    fetch-pack fsck-objects hash-object http-fetch http-push init-db
    local-fetch ls-tree merge-base merge-index mktag pack-objects
    pack-redundant peek-remote prune-packed read-tree send-pack
    ssh-fetch ssh-upload tar-tree unpack-file unpack-objects
    update-server-info var write-tree

With some exceptions, they are pretty much repository wide commands, so I think it is OK for them to assume they start at the toplevel (the Porcelain would chdir to the top for them otherwise). The ones that take paths, namely, checkout-index, hash-object and ls-tree, may want to use setup_git_directory() and do the path prefixing.

That means we would need to have some way for the rest of the commands to check if "${GIT_DIR-.git}" is of the right format version, and call that *once* per process invocation, perferrably at the beginning of the main().

We need an access to .git/config file to do the repository format check anyway, which means we need setup_git_directory() if we ever want to run them from subdirectories. And running setup_git_directory() from the toplevel would not hurt, so perhaps if we add setup_git_directory() at the beginning of main() for the "implicity toplevel" class, and rewrite setup_git_directory() like this:

        static const char *setup_git_directory_main(void)
        {
                /* current setup_git_driectory() */
        }

	const char *setup_git_directory(void)
        {
        	const char *retval = setup_git_directory_main();
                check_repository_format_version(); /* dies on mismatch */
		return retval;
	}

it _might_ be good enough.

I said "it _might_" here, because we need to be careful. Right now, if you run the "implicitly toplevel" commands from a subdirectory, they fail. Some Porcelains and scripts may be relying on that and there will be consequences if things suddenly start not to fail but do something unexpected in higher directories.