[PATCH v4 0/5] maintenance: add an 'is-needed' subcommand

Hello,

I recently raised a patch series [1] to add 'git refs optimize --required' which checks if the reference backend can be optimized, without actually performing the optimization.

Back then, we had decided [2] that it would be a better to broaden the approach and add a 'is-needed' subcommand to 'git-maintenance(1)'. This would allow users to check if maintenance was required for the repository and users could also provide a task via the '--task' to check if maintenance was needed for a particular task.

Ideally the subcommand will be used with the '--auto' flag which can check the same heuristics as that used with 'git maintenance run --auto'. Future patches can also add support for the '--schedule' flag which can be used to check required schedule it met. However that flag isn't added as part of this series.

This series implements that.

Commits 1-3 add the required functionality in the refs subsystem to expose an 'optimize_required' field which can be used to check if backends need to be optimized. Commit 4 utilizes this within the 'git-maintenance(1)' code. Commit 5 adds the 'is-needed' subcommand to 'git-maintenance(1)'.

This is based on top of master a99f379adf (The 27th batch, 2025-10-30) and is dependent on the following series:

    - kn/refs-optim-cleanup
    - ps/ref-peeled-tags

Merges cleanly with `next`. I think those two topics are close to being merged to `next` so hopefully this dependency tree doesn't get too complicated. I'll rebase as needed to resolve conflicts.

[1]: https://lore.kernel.org/git/20251010-562-add-option-to-check-if-reference-backend-needs-repacking-v1-0-c7962be584fa@gmail.com/ [2]: https://lore.kernel.org/git/CAOLa=ZRdxm787nE4FSr2VUHDB+hW06Ggc6yUcKmeTKAb6B7YOA@mail.gmail.com/

---
Changes in v4:
- In `update_segment_if_compaction_required()` change the argument name
  from `use_heuristics` to `use_geometric` since we only have one
  heuristic currently and this is much clearer to understand.
- There were a lot of discussion on how to return a bool variable when
  the function has a return type of int. We discussed both '!!required',
  and 'required != true'. I'm going to punt this discussion keeping it
  simple as 'return required' as in my first version, since even Junio
  expressed his thoughts in favor of it.
- Add a TODO for improvements to the flow when running `git maintenance
  is-needed` without the `--auto` flag.
- Link to v3: https://patch.msgid.link/20251106-562-add-sub-command-to-check-if-maintenance-is-needed-v3-0-d611a2a95cf5@gmail.com

Changes in v3:
- In patch 2/5 extract out code for deciding if compaction is required
  into a static function. This removes duplication of logic for deciding
  if compaction is needed.
- Link to v2: https://patch.msgid.link/20251104-562-add-sub-command-to-check-if-maintenance-is-needed-v2-0-303462a9e4ed@gmail.com

Changes in v2:
- Added more documentation for `reftable_stack_compaction_required()`.
- Fixed some typos and grammar mistakes in commit messages.
- Clarify which tasks will be run when '--task' is not used.
- Move the call to 'usage_with_options()' to be with 'parse_options()'.
- Link to v1: https://patch.msgid.link/20251031-562-add-sub-command-to-check-if-maintenance-is-needed-v1-0-a03d53e28d0e@gmail.com

---
 Documentation/git-maintenance.adoc | 13 ++++++
 builtin/gc.c                       | 93 ++++++++++++++++++++++++++++++++++----
 object.h                           |  1 -
 refs.c                             |  7 +++
 refs.h                             |  7 +++
 refs/debug.c                       | 13 ++++++
 refs/files-backend.c               | 11 +++++
 refs/packed-backend.c              | 13 ++++++
 refs/refs-internal.h               |  6 +++
 refs/reftable-backend.c            | 25 ++++++++++
 reftable/reftable-stack.h          | 11 +++++
 reftable/stack.c                   | 61 +++++++++++++++++++------
 t/t7900-maintenance.sh             | 54 +++++++++++++++-------
 t/unit-tests/u-reftable-stack.c    | 12 ++++-
 14 files changed, 284 insertions(+), 43 deletions(-)

Karthik Nayak (5):
      reftable/stack: return stack segments directly
      reftable/stack: add function to check if optimization is required
      refs: add a `optimize_required` field to `struct ref_storage_be`
      maintenance: add checking logic in `pack_refs_condition()`
      maintenance: add 'is-needed' subcommand

Range-diff versus v3:

1:  48c877bbc4 = 1:  4026aad0e2 reftable/stack: return stack segments directly
2:  b2c0da304b ! 2:  ed2b307572 reftable/stack: add function to check if optimization is required
    @@ reftable/stack.c: static int stack_segments_for_compaction(struct reftable_stack
     -int reftable_stack_auto_compact(struct reftable_stack *st)
     +static int update_segment_if_compaction_required(struct reftable_stack *st,
     +						 struct segment *seg,
    -+						 bool use_heuristics,
    ++						 bool use_geometric,
     +						 bool *required)
      {
     -	struct segment seg;
    @@ reftable/stack.c: static int stack_segments_for_compaction(struct reftable_stack
     +		return 0;
     +	}
     +
    -+	if (!use_heuristics) {
    ++	if (!use_geometric) {
     +		*required = true;
      		return 0;
     +	}
3:  aec4861598 = 3:  4ac6fe6346 refs: add a `optimize_required` field to `struct ref_storage_be`
4:  36d9bfbfe0 ! 4:  e239f9d1d1 maintenance: add checking logic in `pack_refs_condition()`
    @@ builtin/gc.c: static void maintenance_run_opts_release(struct maintenance_run_op
     +	clear_ref_exclusions(&excludes);
     +	string_list_clear(&included_refs, 0);
     +
    -+	return required == true;
    ++	return required;
      }
      
      static int maintenance_task_pack_refs(struct maintenance_run_opts *opts,
5:  2fc7dcb38c ! 5:  2457e5ce0e maintenance: add 'is-needed' subcommand
    @@ Documentation/git-maintenance.adoc: The `unregister` subcommand will report an e
     +    Exits with a 0 status code if maintenance needs to be run, 1 otherwise.
     +    Ideally used with the '--auto' flag.
     ++
    -+If one or more `--task` options	are specified, then those tasks are checked
    ++If one or more `--task` options are specified, then those tasks are checked
     +in that order. Otherwise, the tasks are determined by which
     +`maintenance.<task>.enabled` config options are true. By default, only
     +`maintenance.gc.enabled` is true.
    @@ builtin/gc.c: static int maintenance_stop(int argc, const char **argv, const cha
     +			}
     +		}
     +	} else {
    -+		/* When not using --auto, we should always require maintenance. */
    ++		/*
    ++		 * When not using --auto we always require maintenance right now.
    ++		 *
    ++		 * TODO: this certainly is too eager, as some maintenance tasks may
    ++		 * decide to not do anything because the data structures are already
    ++		 * fully optimized. We may eventually want to extend the auto
    ++		 * condition to also cover non-auto runs so that we can detect such
    ++		 * cases.
    ++		 */
     +		is_needed = true;
     +	}
     +

base-commit: edd2018f5db39d68d55a7a4af42375b1a06b9406 change-id: 20251021-562-add-sub-command-to-check-if-maintenance-is-needed-01cae01b4606

Thanks
- Karthik