From: Eric Wong <e@80x24.org>
Date: Fri, 30 Aug 2019 19:36:36 GMT
Subject: Re: [PATCH 11/11] hashmap_get_next returns "struct hashmap_entry *"
Message-ID: <20190830193636.xg6pcd5hjcgk3nob@dcvr>
In-Reply-To: <2f22841c-779e-6d6e-19b7-0a119088e343@gmail.com>

Derrick Stolee <stolee@gmail.com> wrote:
> On 8/25/2019 10:43 PM, Eric Wong wrote:
> > --- a/diff.c
> > +++ b/diff.c
> > @@ -1035,8 +1035,10 @@ static void pmb_advance_or_null_multi_match(struct diff_options *o,
> >  {
> >  	int i;
> >  	char *got_match = xcalloc(1, pmb_nr);
> > +	struct hashmap_entry *ent = &match->ent;
> >  
> > -	for (; match; match = hashmap_get_next(hm, &match->ent)) {
> > +	for (; ent; ent = hashmap_get_next(hm, ent)) {
> > +		match = container_of(ent, struct moved_entry, ent);
> 
> Lines like this are very difficult to parse. In this
> container_of() macro, 'ent' is taking both the 'ptr' and
> 'member' values.

Agreed, naming is hard :<

In the Linux kernel list.h implementation, there's actually
list_for_each_entry, list_next_entry and a bunch of other
macros which allow the caller to avoid using container_of.
We only have list_first_entry, so far.

We can draw inspiration from those macros by creating
hashmap_get_next_entry and hashmap_for_each_entry macros
which allow callers specify the type once; and there'd
be no need for callers to specify the hashmap_entry
pointer name at all :)

Unlike the kernel, it looks like we can't rely on __typeof__ in
git, but I think we can let the caller specify the type once...