Re: [PATCH 7/9] list-objects-filter-options: implement auto filter resolution
- From
Christian Couder <christian.couder@gmail.com>
- Date
- Feb 4, 2026, 10:29 UTC
- Message-ID
- <CAP8UFD0iBxn6cPFKLAkSW7O3To1ago60MWYwV7YxjxOVxni1Kw@mail.gmail.com>
- In-Reply-To
- <aV4v8HCe6CLqXJ-1@pks.im>
On Wed, Jan 7, 2026 at 11:05 AM Patrick Steinhardt <ps@pks.im> wrote:
Show 12 quoted lines
> > On Tue, Dec 23, 2025 at 12:11:11PM +0100, Christian Couder wrote: > > In a following commit, we will need to aggregate filters from multiple > > accepted promisor remotes into a single filter. > > Ah, interesting. I was always operating under the assumption that when > the server advertises multiple promisors, the client will pick only one > of them. And that made me wonder how the client knows which one to pick > in the first place. > > But of course it's possible to just pick _all_ of them by combining the > filter.
Yeah, that's the idea.
Show 28 quoted lines
> > diff --git a/list-objects-filter-options.c b/list-objects-filter-options.c
> > index f13ae5caeb..4a9c1991c1 100644
> > --- a/list-objects-filter-options.c
> > +++ b/list-objects-filter-options.c
> > @@ -230,6 +230,41 @@ static void filter_spec_append_urlencode(
> > filter->filter_spec.buf + orig_len);
> > }
> >
> > +char *list_objects_filter_combine(const struct string_list *specs)
> > +{
> > + struct strbuf buf = STRBUF_INIT;
> > +
> > + if (!specs->nr)
> > + return NULL;
> > +
> > + if (specs->nr == 1)
> > + return xstrdup(specs->items[0].string);
> > +
> > + strbuf_addstr(&buf, "combine:");
> > +
> > + for (size_t i = 0; i < specs->nr; i++) {
> > + const char *spec = specs->items[i].string;
> > + if (i > 0)
> > + strbuf_addch(&buf, '+');
> > +
> > + strbuf_addstr_urlencode(&buf, spec, allow_unencoded);
>
> Shouldn't we use `filter_spec_append_urlencode()` to do this?Yeah, probably, see below.
Show 6 quoted lines
> > + } > > + > > + return strbuf_detach(&buf, NULL); > > +} > > I'm surprised we didn't have such a function yet.
I have refactored the code so that we use a temporary `struct list_objects_filter_options` and `gently_parse_list_objects_filter()` to construct a combined filter in the next commit instead of this function.
This also takes care of your comment above about `strbuf_addstr_urlencode()` vs `filter_spec_append_urlencode()`.
Show 11 quoted lines
> > +void list_objects_filter_resolve_auto(struct list_objects_filter_options *filter_options,
> > + char *new_filter, struct strbuf *errbuf)
> > +{
> > + if (filter_options->choice != LOFC_AUTO)
> > + return;
>
> I wonder whether we should rather `BUG()` in case the filter is not an
> "auto" filter. Otherwise it's easy to get the callsite wrong, as the
> user may expect that the filter gets resolved tdo the new filter, but
> it's actually not because the original filter wasn't an "auto" filter in
> the first place.Actually the list_objects_filter_resolve_auto() function is not very useful, so I have just removed it too in the v2 I will send.
This way this whole patch is not necessary and has been removed in v2.
Show 11 quoted lines
> > + list_objects_filter_release(filter_options); > > + > > + if (new_filter) > > + gently_parse_list_objects_filter(filter_options, new_filter, errbuf); > > +} > > So as menitoned in a preceding commit `list_objects_filter_release()`, > will retain the `allow_auto` option. But when resolving "auto" filters > I'd expect us to not accept "auto" in the resolved filter anymore. > Otherwise, if `new_filter` was "auto", we'd still end up with an auto > filter, wouldn't we? I'd rather expect us to abort in that case.
Right, I have fixed this by adding the following in the next commit:
/* The result of resolving an 'auto' filter must not be 'auto' */
args->filter_options.allow_auto_filter = 0;Thanks!