On Fri, May 15, 2020 at 02:17:54PM -0400, Jeff King wrote: > On Fri, May 15, 2020 at 12:04:46PM +0200, Christian Couder wrote: > > > As we cleanup 'upload-pack.c' by using 'struct upload_pack_data' > > more thoroughly, let's pass 'struct upload_pack_data' to > > get_common_commits(), so that this function and the functions > > it calls can use all the fields of that struct in followup > > commits. > > This one is optional, I think. We're not removing any globals nor moving > towards removing any (I don't think), because we're really just > replacing two object_array parameters with the big data struct: > > > -static int get_common_commits(struct packet_reader *reader, > > - struct object_array *have_obj, > > - struct object_array *want_obj) > > +static int get_common_commits(struct upload_pack_data *data, > > + struct packet_reader *reader) > > I say "I don't think" because it's possible that get_common_commits() > could use one of the other fields (or pass it to another function) in a > later patch. But I don't see it. > > So I think one _could_ argue that it should continue to take the minimum > set of arguments it needs. I doubt it would ever be useful outside of the > context of upload_pack, so I think in practice it doesn't matter much > (beyond perhaps documenting which parts of the global data it needs). So > I don't feel strongly either way. OK, I take it back. I finally go to the spot later in the series where you use data->stateless_rpc, etc. So I think this is a positive movement. I think it might have been easier to review if the changes were made in the opposite order: globals removed and passed down the call stack, and then the big struct used to simplify passing the data around. But that would have been a lot more tedious to write (and perhaps to review, as I'd have made the opposite complaint ;) ). And definitely not worth doing now. -Peff
