Re: [PATCH 2/2 v2] ls-files: add pathspec matching for submodules

From: Junio C Hamano <gitster@pobox.com>
Date: Sep 23, 2016, 18:48 UTC
Message-ID: <xmqqfuoq1mh9.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <1474500486-101640-1-git-send-email-bmwill@google.com>

Brandon Williams <bmwill@google.com> writes:

Show 12 quoted lines

> -	/* Find common prefix for all pathspec's */
> -	max_prefix = common_prefix(&pathspec);
> +	/*
> +	 * Find common prefix for all pathspec's
> +	 * This is used as a performance optimization which unfortunately cannot
> +	 * be done when recursing into submodules
> +	 */
> +	if (recurse_submodules)
> +		max_prefix = NULL;
> +	else
> +		max_prefix = common_prefix(&pathspec);
>  	max_prefix_len = max_prefix ? strlen(max_prefix) : 0;

This is OK for now, but for a future enhancement, I think we could do better than this. In a superproject with a submodule at "sub/", the current implementation of the common_prefix() helper would yield "sub/a/" when given "sub/a/x" and "sub/a/y" (a pathspec with two elements), which we want to avoid.

But somebody should be able to notice, before "sub/a/" is given to max_prefix here, that "sub/" is the leaf level in our repository and reduce the max_prefix to it. dir.c::common_prefix_len() might be a place we could do so, but I didn't think about the ramifications of doing so for other callers of common_prefix() or when we are not recursing into submodules. Doing it in the caller here, i.e.

	max_prefix = common_prefix(&pathspec);
        if (recurse_submodules)
        	max_prefix = chomp_at_submodule_boundary(max_prefix);

is certainly safer.

If the superproject has submodules at "a/b/{sub1,sub2,...}", this matters more. We do want to notice that we won't have to scan outside "a/b/" of the index given "a/b/sub1" and "a/b/sub2" as a pathspec.

The common_prefix_len() function also looks beyond symbolic links, which is another thing that we may want to think about. In a repository with a symbolic link "link" pointing somewhere else, when you give "link/a/x" and "link/a/y" (a pathspec with two elements), we would get "link/a/" as a common prefix, but we won't find anything underneath "link" in our index. In such a case, leaving the common prefix to "link/a/" _might_ allow us to notice that no pathspec elements can ever match, so not noticing that the common prefix points beyond a symbolic link might be a feature. I dunno.