Re: [PATCH 2/2] builtin/history: implement "drop" subcommand
- From
Patrick Steinhardt <ps@pks.im>
- Date
- Jun 3, 2026, 10:06 UTC
- Message-ID
- <ah_8s1Q1Kg0dgw6a@pks.im>
- In-Reply-To
- <xmqqbjdt25e3.fsf@gitster.g>
On Tue, Jun 02, 2026 at 08:43:48AM +0900, Junio C Hamano wrote:
> Patrick Steinhardt <ps@pks.im> writes:
[snip]
Show 20 quoted lines
> > +static int update_worktree(struct repository *repo,
> > + const struct commit *old_head,
> > + const struct commit *new_head,
> > + bool dry_run)
> > +{
> > +...
> > +
> > +out:
> > + clear_unpack_trees_porcelain(&opts);
> > + rollback_lock_file(&lock);
> > + release_index(&index);
> > + free(desc_buf[0]);
> > + free(desc_buf[1]);
> > + return ret;
> > +}
>
> The function looks very familiar---anybody who wants to perform
> "checkout <other-commit>" needs to do exactly the above. It is a
> bit surprising and disappointing that this topic needs to *invent*
> its own helper function and carry it as a file-scope static.It certainly is. We basically have this whole dance in ~8 different locations by now, and given the verbosity that is required for the whole setup it's a good hint that the interface is not exactly great.
One of the functions that we might be able to reuse is `reset_head()`... goes down the rabbit hole... ugh, this is turning out to be somewhat painful. I'll send a v2 that does the whole exercise, but I'm not a 100% convinced it's the right thing to do. There's various assumptions that we have to break:
- It assumes that the index is always clean.
- We don't have a dry-run mode.
- We need to stop it from updating any refs.
- We need to introduce another field to let the caller decide which
commit we're moving from.So I'm 7 commits deep now adapting the function to our needs. But maybe the end result is ultimately worth it...? We'll see.
Patrick