Re: [WIP/PATCH v4 6/8] for-each-ref: rename some functions and make them public

On June 1, 2015 8:23:51 PM GMT+05:30, Junio C Hamano [off-list ref] wrote:

Karthik Nayak [off-list ref] writes:

quoted

Could be achieved using a simple wrapper around 'filter_refs()'
something like this perhaps.

int filter_refs_with_pattern(struct ref_array *ref, int
(*for_each_ref_fn)(each_ref_fn, void *), char **patterns)
{
	int i;
	struct ref_filter_cbdata data;
	data.filter.name_patterns = patterns;
	filter_refs(for_each_ref_fn, &data);

I presume that this is

filter_refs(&refs, for_each_ref_fn, &data);

as you would need to have some way to get the result back ;-)

No it's meant to be that way, see as it takes a struct ref_filter_cbdata, the filters are put in the data.filter.* and we obtain the refs in the data.array.items. So no need for another ref_array to hold the array of items.

quoted

	refs->nr = data.array.nr;
	for(i = 0; i < refs->nr; i++) {
		/* copy over the refs */
	}
	return 0;
}

Is this on the lines of what you had in mind? If it is, than I could
just create a new patch which would make ref_filter_handler() private
and introduce filter_refs() as shown.

Yeah.  Even though I suggested

filter_refs(&for_each_ref, ...);

I actually would think the external interface should not mention
for_each_ref() like I did.  The primary reason why I felt that it is
bad for the API to export a generic callback function the caller can
use to call for_each_ref() or for_each_rawref()" in the longer term
is because it forces us to always iterate all refs; for_each_ref()
does not know what the callback filter function wants to do.  The
most common way to filter in the context of your GSoC project is "we
limit only to refs/heads/*, and then we may also filter by other
criteria" (that is "git branch" "-l" or possibly with "--contains",
etc.), and it is very wasteful for that codepath to allow
for_each_ref() to even enumerate and feed all refs outside the
refs/heads/* area to your callback, which would involve reading all
entries in packed-refs (which is a fixed cost so not an overhead)
and then reading everything in .git/refs/* (which is an overhead we
could and should avoid when we know we are only interested in the
branches that live in refs/heads*).

Your first implementation of course can just call for_each_ref() or
for_each_rawref(), and at the end of GSoC, the code may still do so.
But by keeping the external interface free of for_each_ref(), you
could later optimize.

And your above sample only takes for_each_ref_fn without exposing the
internal use of for_each_ref(), which is good.

That seems good, thanks for the guidance.