Re: [PATCH 3/3] replay: allow to specify a ref with option --ref
- From
Toon Claes <toon@iotcl.com>
- Date
- Mar 25, 2026, 12:50 UTC
- Message-ID
- <87h5q413b5.fsf@iotcl.com>
- In-Reply-To
- <b0d49276-a9d3-435d-bb39-8cf7a08fbd11@malon.dev>
Tian Yuchen <cat@malon.dev> writes:
Show 17 quoted lines
> Hi Toon, > > On 3/24/26 00:09, Toon Claes wrote: > > >> + die_for_incompatible_opt2(!!opts.ref, "--ref", >> + !!opts.contained, "--contained"); > >> + /* >> + * Reference to update with the result of the replay. This stops replay >> + * from taking refs from `onto`, `advance`, and `revert`. Ignores >> + * `contained`. >> + */ >> + const char *ref; >> + > > Do the comment and the code implementation match up here?
The comment is a bit misleading. A user cannot set the field `contained` in the struct when it passed `--ref`. But if a bug is made somewhere, but both fields can be set, that's what the comment is saying.
I can add a check to verify in replay.c that not both are set, and update the comment they are incompatible?
Show 23 quoted lines
>> @@ -427,13 +446,9 @@ int replay_revisions(struct rev_info *revs,
>> goto out;
>> }
>>
>> - /* In --advance or --revert mode, update the target ref */
>> - if (advance || revert) {
>> - const char *ref = advance ? advance : revert;
>> - replay_result_queue_update(out, ref,
>> - &onto->object.oid,
>> + if (ref)
>> + replay_result_queue_update(out, ref, &old_oid,
>> &last_commit->object.oid);
>> - }
>>
>> ret = 0;
>
> It doesn't seem like there's anything here to prevent users from
> inputting multiple branches? I think this will result in at least one
> branch becoming orphaned, without any error being reported.
>
> Consider this command:
>
> git replay --onto main --ref refs/heads/target topic1 topic2That's interesting, I didn't think of that.
Show 21 quoted lines
>
> In replay.c we have:
>
>> if (onto_name) {
>> *onto = peel_committish(repo, onto_name, "--onto");
>> if (rinfo.positive_refexprs <
>> strset_get_size(&rinfo.positive_refs))
>> die(_("all positive revisions given must be references"));
>> *update_refs = xcalloc(1, sizeof(**update_refs));
>> **update_refs = rinfo.positive_refs;
>> memset(&rinfo.positive_refs, 0, sizeof(**update_refs));
>
>
> After topic1 and topic2 are put into update_refs, only one ref
> (refs/heads/target) is updated, since what is written in the patch is:
>
> > + if (ref)
> > + replay_result_queue_update(out, ref, &old_oid,
> > &last_commit->object.oid);
>
> I believe this is not what we expect :PLet me look into that how we can deal with that.
-- Cheers, Toon