Re: [RFC PATCH] switch: provide configurable detach
- From
- Thibaud CANALE <thican@thican.net>
- Date
- Apr 4, 2026, 21:10 UTC
- Message-ID
- <4gpvxpm335bvnml57ks5zbfyzd5pzp6jheebvytsobmlffpllg@6ddpoiruhgh6>
- In-Reply-To
- <CAN5EUNR6x_HWQKrMMHcHj0hF+D_6QBiw782_QGA8E8SS7Q3pcg@mail.gmail.com>
On Saturday 2026-04-04 17:23:46+0200, Pablo <pabloosabaterr@gmail.com> wrote:
> El sáb, 4 abr 2026 a las 16:36, Thibaud CANALE (<thican@thican.net>) escribió: > > Hi Thibaud,
Hello,
Show 7 quoted lines
> > Its purpose is to provide for git-switch(1) same detach behavior on > > commit it than git-checkout(1) through configuration option > > `checkout.switchDetach`. > > The commit needs some work, following Documentation/SubmittingPatches: > The subject looks fine but the body, you should write imperatively, something > like "Allow git-switch to..." instead of "Its purpose is to...".
It was a (almost) copy of the message I wrote in the documentation, I wanted to avoid giving different and possibly confusing explanations between the two of them. I just noticed the typo on "commit it". By the way, is "commit id" correct for constructs such as "${branch}~x" or "${commit_id}~y", or is there a better word for this?
Show 5 quoted lines
> > same detach behavior on commit it than > > This is confusing. > > Also the body should explain WHY this change would be wanted.
May you tell me what you find confusing? I don’t see what else I can explain without over-explaining.
Maybe I can provide a real case example: Using `git checkout my_branch~2` works and automatically detach on the 2nd parent of the top commit of branch "my_branch", but the same argument with `git switch` does not. The command `git switch -` also does not work if the previous step was a detached commit while `git checkout -` can move back and forth between detached mode and branch.
About why force_detach flag is necessary, read below.
About why this feature, I know some users won’t change to switch/restore commands and keep relying on checkout command because of this “missing” feature, as its “disrupts” their workflow (humans, ’am I right?).
Show 25 quoted lines
> > > > Signed-off-by: Thibaud CANALE <thican@thican.net> > > --- > > Documentation/config/checkout.adoc | 4 ++++ > > builtin/checkout.c | 18 +++++++++++++++--- > > t/t2060-switch.sh | 27 +++++++++++++++++++++++++++ > > 3 files changed, 46 insertions(+), 3 deletions(-) > > > > diff --git a/Documentation/config/checkout.adoc b/Documentation/config/checkout.adoc > > index e35d2129..3a6c1049 100644 > > --- a/Documentation/config/checkout.adoc > > +++ b/Documentation/config/checkout.adoc > > @@ -22,6 +22,10 @@ commands or functionality in the future. > > option in `git checkout` and `git switch`. See > > linkgit:git-switch[1] and linkgit:git-checkout[1]. > > > > +`checkout.switchDetach`:: > > + Provides for linkgit:git-switch[1] the same detach behavior on commit id > > + than linkgit:git-checkout[1] without the explicit --detach option. > > + > > I think it would be better to describe how it would be if set rather > than directly > the what it does. e.g.: > "If set, detaches HEAD when..."
Rewritten with details and example.
Show 17 quoted lines
> > `checkout.workers`::
> > The number of parallel workers to use when updating the working tree.
> > The default is one, i.e. sequential execution. If set to a value less
> > diff --git a/builtin/checkout.c b/builtin/checkout.c
> > index e031e618..1da1062e 100644
> > --- a/builtin/checkout.c
> > +++ b/builtin/checkout.c
> > @@ -53,6 +53,7 @@ struct checkout_opts {
> > int force;
> > int force_detach;
> > int implicit_detach;
> > + int switch_detach; // For checkout.switchDetach configuration
>
> // coments are not used in Git, if you need to add comments make them with
> /* */.
> Nit I don't think a comment here is necessary, it is very self
> explanatory already.I agree, I originally didn’t, I thought we could tell me to describe it here to avoid looking for explanation somewhere else.
Show 16 quoted lines
> > int writeout_stage;
> > int overwrite_ignore;
> > int ignore_skipworktree;
> > @@ -1005,7 +1006,10 @@ static void update_refs_for_switch(const struct checkout_opts *opts,
> > else
> > strbuf_insertstr(&msg, 0, reflog_msg);
> >
> > - if (!strcmp(new_branch_info->name, "HEAD") && !new_branch_info->path && !opts->force_detach) {
> > + if (!strcmp(new_branch_info->name, "HEAD") &&
> > + !new_branch_info->path &&
> > + !opts->force_detach &&
> > + !opts->switch_detach) {
>
> seems that you're checking force_detach and switch_detach together
> every time,they look very similar,
> sure you need them separated?Not every time. There is cases where switch_detach is not used such as using a branch.
Show 26 quoted lines
> > /* Nothing to do. */
> > } else if (opts->force_detach || !new_branch_info->path) { /* No longer on any branch. */
> > refs_update_ref(get_main_ref_store(the_repository), msg.buf,
> > @@ -1014,7 +1018,9 @@ static void update_refs_for_switch(const struct checkout_opts *opts,
> > REF_NO_DEREF, UPDATE_REFS_DIE_ON_ERR);
> > if (!opts->quiet) {
> > if (old_branch_info->path &&
> > - advice_enabled(ADVICE_DETACHED_HEAD) && !opts->force_detach)
> > + advice_enabled(ADVICE_DETACHED_HEAD) &&
> > + !opts->force_detach &&
> > + !opts->switch_detach)
> > detach_advice(new_branch_info->name);
> > describe_detached_head(_("HEAD is now at"), new_branch_info->commit);
> > }
> > @@ -1541,8 +1547,11 @@ static void die_expecting_a_branch(const struct branch_info *branch_info)
> > */
> > code = die_message(_("a branch is expected, got '%s'"), branch_info->name);
> >
> > - if (advice_enabled(ADVICE_SUGGEST_DETACHING_HEAD))
> > + if (advice_enabled(ADVICE_SUGGEST_DETACHING_HEAD)) {
> > advise(_("If you want to detach HEAD at the commit, try again with the --detach option."));
> > + advise(_("Also you can detach by default by setting the config variable "
> > + "checkout.switchDetach to true."));
> > + }
>
> Nit "Also, you can..." or "You can also..."As you prefer.
Show 36 quoted lines
> > > > exit(code); > > } > > @@ -1660,6 +1669,7 @@ static int checkout_branch(struct checkout_opts *opts, > > > > if (!opts->implicit_detach && > > !opts->force_detach && > > + !opts->switch_detach && > > Again force_detach + switch_detach > > > !opts->new_branch && > > !opts->new_branch_force && > > new_branch_info->name && > > @@ -2119,6 +2129,8 @@ int cmd_switch(int argc, > > options = add_common_options(&opts, options); > > options = add_common_switch_branch_options(&opts, options); > > > > + repo_config_get_bool(the_repository, "checkout.switchDetach", &opts.switch_detach); > > + > > cb_option = 'c'; > > > > return checkout_main(argc, argv, prefix, &opts, options, > > diff --git a/t/t2060-switch.sh b/t/t2060-switch.sh > > index c91c4db9..3435ae98 100755 > > --- a/t/t2060-switch.sh > > +++ b/t/t2060-switch.sh > > @@ -177,4 +177,31 @@ test_expect_success 'switch back when temporarily detached and checked out elsew > > git -C wt2 switch --ignore-other-worktrees shared > > ' > > > > +test_expect_success 'switch with configured detach behavior from configuration ' ' > > Trailing space here at the end, also the title is kinda confusing with the > configured + configuration, what about: > "switch detaches on commit when switchDetach is set"
For the trailing space I was mislead by the previous test, line 149; it has its trailing space, I thought it was mandatory somehow. Thank you for the rewording.
Show 24 quoted lines
> > + test_when_finished " > > + test_config checkout.switchDetach false; > > + git switch main > > + " && > > + test_config checkout.switchDetach true && > > + git switch main && > > + git symbolic-ref HEAD && > > + git switch main~ && > > + test_must_fail git symbolic-ref HEAD && > > + git switch - && > > + git symbolic-ref HEAD && > > + git switch - && > > + test_must_fail git symbolic-ref HEAD > > +' > > + > > +test_expect_success 'switch without configured detach behavior from configuration ' ' > > Trailing space. > > > + test_when_finished " > > + test_config checkout.switchDetach false; > > test_config already has test_when_finished > I would git config --unset.
ACK.
Show 5 quoted lines
> > + git switch main > > + " && > > + test_config checkout.switchDetach false && > > This is already false.
I prefer to be explicit, to test the case where the user disables this feature on purpose.
Show 22 quoted lines
> > + git switch main && > > + git symbolic-ref HEAD && > > + test_must_fail git switch main~ > > +' > > + > > test_done > > > > base-commit: 2855562ca6a9c6b0e7bc780b050c1e83c9fcfbd0 > > -- > > Thibaud CANALE > > thican [at] thican [dot] net > > https://thican.net/ > > GPG: rsa4096 2013-10-14 485EF628CB85CDD4CB7CFF0D52F5127650733A18 > > Overall seems reasonable, it might be a design choice having to > explicitly use --detach, force_detach it's pretty clear about it but > I dunno. > > For now: > I would rewrite the commit. > Make sure that you need a new flag, force_detach looks very similar. > Fix the style issue.
It was already sure a new flag was necessary, using force detach flag is uncorrect as it will not anymore allow to use back a branch for example. We can see this new flag is not everywhere force_detach is. For testing, I tried only force_detach, the tests are failing.
V2 will follow shortly.
Best regards,
-- Thibaud CANALE thican [at] thican [dot] net https://thican.net/ GPG: rsa4096 2013-10-14 485EF628CB85CDD4CB7CFF0D52F5127650733A18