[RFC PATCH 00/14] repack: incremental MIDX/bitmap-based repacking

[RFC PATCH 00/14] repack: incremental MIDX/bitmap-based repacking

[Taylor Blau]

[NOTE: this series has not yet been tested thoroughly and is based on
the v3 of 'tb/incremental-midx-part-3.2', which is d54da84bd9d (midx:
enable reachability bitmaps during MIDX compaction, 2026-02-24).

I suspect that there will be non-trivial changes to the approach in this
series based on feedback, hence the RFC.]

Note to the maintainer:

 * This series is based on 'tb/incremental-midx-part-3.2', I suggest
   queueing as 'tb/incremental-midx-part-3.3'.

OVERVIEW
========

This series implements the third and final major component for an
incremental MIDX/bitmap-based repacking strategy. As a refresher, those
are:

 1. Refactoring code out of builtin/repack.c into individual compilation
    units outside of the builtin directory.

 2. Implementing MIDX layer compaction, i.e., the ability to combine a
    contiguous sequence of MIDX layers in an existing chain and replace
    them with a single layer representing the union of objects/packs
    among the compacted layers.

 3. (this series) An incremental repacking strategy that, unlike our
    current '--geometric' approach, does not rely on periodic
    all-into-one repacks.

BACKGROUND
==========

Today, a '--geometric' repack with '--write-midx' rewrites the entire
MIDX (and, when enabled, its reachability bitmap) on every invocation.
For most repositories this is acceptable. In large monorepos, these
costs can add up significantly, especially when those repositories are
repacked frequently.

The incremental MIDX support introduced in the earlier parts of this
effort allows us to append new layers to a MIDX chain without rewriting
anything. Combined with the support for MIDX compaction in 3.2, we now
have all of the building blocks needed to maintain an incrementally
growing and shrinking MIDX chain as part of the repack cycle.

STRATEGY
========

The incremental repacking strategy implemented in this series works
(roughly) as follows:

 1. Repack non-MIDX'd packs using an ordinary '--geometric' repack,
    optionally including packs from the tip MIDX layer if and only if it
    contains more than 'repack.midxNewLayerThreshold' number of packs.

 2. Prepare to write a new MIDX layer containing the freshly written
    pack(s) (and any surviving packs from the tip layer, if it was
    rewritten).

 3. Perform a compaction pass over adjacent MIDX layers to restore a
    geometric progression on object count among layers in the chain
    (determined by 'repack.midxSplitFactor').

 4. Write the new MIDX chain, link it into place, and remove redundant
    layers.

In effect, this approach encourages MIDX chains where:

 * older layers contain fewer, larger packs, and

 * newer layers contain many smaller packs.

As a result of the compaction pass, we prevent the chain itself from
ever growing too long.

This roughly matches the description of this algorithm I gave at Git
Merge last year, which is covered on slides 80-131 of this presentation:

    https://ttaylorr.com/presentations/git-merge-2025.pdf

THIS SERIES
===========

This series is organized roughly as follows:

 * The first three patches perform small-ish quality-of-life refactors
   within the MIDX machinery.

 * The next two patches introduce `--checksum-only` and `--base`
   options for `git multi-pack-index write` and `compact`, which are
   needed by the incremental repacking machinery to assemble MIDX
   chains without prematurely updating the chain file.

 * The next six patches prepare the repack infrastructure and pack
   geometry code for the new repacking strategy.

 * The final three patches introduce the new repacking machinery, expose
   it to users, and then extend it to work with non-geometric repacks in
   that order. The first two are the substantive patches of this series.

WHERE WE'RE AT
==============

This series delivers the final substantive component of this overall
effort to enable the new repacking strategy implemented here. There are
a couple of (comparatively much smaller) items that will be useful as
follow-on items, including:

 * "Reachability-infused" geometric repacking to emit small cruft packs
   to introduce other ways to update the set of cruft objects beyond a
   whole-repository traversal.

 * Richer bitmap configuration to determine which bitmap(s) to carry
   forward between adjacent MIDX layers when doing an incremental
   repack, to prevent an endless accumulation of reachability bitmaps.

Like I mentioned earlier, I think that both of those are significantly
smaller challenges than this and the previous series. As a general note,
I am going on vacation for two weeks starting this Friday (2026-02-24)
and will be away from the list.

I'll plan on picking up any review from this series as well as the two
other topics listed above as soon as I get back.

Thanks in advance for your review!

Taylor Blau (14):
  midx: use `string_list` for retained MIDX files
  strvec: introduce `strvec_init_alloc()`
  midx: use `strvec` for `keep_hashes`
  midx: introduce `--checksum-only` for incremental MIDX writes
  midx: support custom `--base` for incremental MIDX writes
  repack: track the ODB source via existing_packs
  midx: expose `midx_layer_contains_pack()`
  repack-midx: factor out `repack_prepare_midx_command()`
  repack-midx: extract `repack_fill_midx_stdin_packs()`
  repack-geometry: prepare for incremental MIDX repacking
  builtin/repack.c: convert `--write-midx` to an `OPT_CALLBACK`
  repack: implement incremental MIDX repacking
  repack: introduce `--write-midx=incremental`
  repack: allow `--write-midx=incremental` without `--geometric`

 Documentation/config/repack.adoc        |  18 +
 Documentation/git-multi-pack-index.adoc |  19 +-
 Documentation/git-repack.adoc           |  44 +-
 builtin/multi-pack-index.c              |  48 +-
 builtin/repack.c                        |  90 ++-
 midx-write.c                            | 119 ++--
 midx.c                                  | 104 ++--
 midx.h                                  |  11 +-
 repack-geometry.c                       |  50 +-
 repack-midx.c                           | 704 +++++++++++++++++++++++-
 repack.c                                |  23 +-
 repack.h                                |  25 +-
 strvec.c                                |   7 +
 strvec.h                                |   5 +
 t/meson.build                           |   1 +
 t/t5334-incremental-multi-pack-index.sh |  47 ++
 t/t5335-compact-multi-pack-index.sh     | 105 ++++
 t/t7705-repack-incremental-midx.sh      | 461 ++++++++++++++++
 18 files changed, 1730 insertions(+), 151 deletions(-)
 create mode 100755 t/t7705-repack-incremental-midx.sh


base-commit: ac6221686db11ba802ced8ea6b8b2fa8388b21b2
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 06/14] repack: track the ODB source via existing_packs

[Taylor Blau]

Store the ODB source in the `existing_packs` struct and use that in
place of the raw `repo->objects->sources` access within `cmd_repack()`.

The source used is still assigned from the first source in the list, so
there are no functional changes in this commit. The changes instead
serve two purposes (one immediate, one not):

 - The incremental MIDX-based repacking machinery will need to know what
   source is being used to read the existing MIDX/chain from that source
   (should one exist).

 - In the future, if "git repack" is taught how to operate on other
   object sources, this field will serve as the authoritative value for
   that source.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 builtin/repack.c | 5 ++---
 repack.c         | 2 ++
 repack.h         | 1 +
 3 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/builtin/repack.c b/builtin/repack.c
index f6bb04bef72..44a95b56f23 100644
--- a/builtin/repack.c
+++ b/builtin/repack.c
@@ -402,7 +402,7 @@ int cmd_repack(int argc,
 		 * midx_has_unknown_packs() will make the decision for
 		 * us.
 		 */
-		if (!get_multi_pack_index(repo->objects->sources))
+		if (!get_multi_pack_index(existing.source))
 			midx_must_contain_cruft = 1;
 	}
 
@@ -549,8 +549,7 @@ int cmd_repack(int argc,
 		unsigned flags = 0;
 		if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX_WRITE_INCREMENTAL, 0))
 			flags |= MIDX_WRITE_INCREMENTAL;
-		write_midx_file(repo->objects->sources,
-				NULL, NULL, flags);
+		write_midx_file(existing.source, NULL, NULL, flags);
 	}
 
 cleanup:
diff --git a/repack.c b/repack.c
index 596841027af..2ee6b51420a 100644
--- a/repack.c
+++ b/repack.c
@@ -154,6 +154,8 @@ void existing_packs_collect(struct existing_packs *existing,
 			string_list_append(&existing->non_kept_packs, buf.buf);
 	}
 
+	existing->source = existing->repo->objects->sources;
+
 	string_list_sort(&existing->kept_packs);
 	string_list_sort(&existing->non_kept_packs);
 	string_list_sort(&existing->cruft_packs);
diff --git a/repack.h b/repack.h
index bc9f2e1a5de..c0e9f0ca647 100644
--- a/repack.h
+++ b/repack.h
@@ -56,6 +56,7 @@ struct packed_git;
 
 struct existing_packs {
 	struct repository *repo;
+	struct odb_source *source;
 	struct string_list kept_packs;
 	struct string_list non_kept_packs;
 	struct string_list cruft_packs;
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 01/14] midx: use `string_list` for retained MIDX files

[Taylor Blau]

Both `clear_midx_files_ext()` and `clear_incremental_midx_files_ext()`
build a list of filenames to keep while pruning stale MIDX files. Today
they hand-roll an array instead of using a `string_list`, thus requiring
us to pass an additional length parameter, and makes lookups linear.

Replace the bare array with a `string_list` which can be passed around
as a single parameter. Though it improves lookup performance, the
difference is likely immeasurable given how small the keep_hashes array
typically is.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 midx.c | 56 ++++++++++++++++++++++----------------------------------
 1 file changed, 22 insertions(+), 34 deletions(-)

diff --git a/midx.c b/midx.c
index c1b9658240d..c5e3553e2bb 100644
--- a/midx.c
+++ b/midx.c
@@ -755,8 +755,7 @@ int midx_checksum_valid(struct multi_pack_index *m)
 }
 
 struct clear_midx_data {
-	char **keep;
-	uint32_t keep_nr;
+	struct string_list keep;
 	const char *ext;
 };
 
@@ -764,15 +763,12 @@ static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUS
 				const char *file_name, void *_data)
 {
 	struct clear_midx_data *data = _data;
-	uint32_t i;
 
 	if (!(starts_with(file_name, "multi-pack-index-") &&
 	      ends_with(file_name, data->ext)))
 		return;
-	for (i = 0; i < data->keep_nr; i++) {
-		if (!strcmp(data->keep[i], file_name))
-			return;
-	}
+	if (string_list_has_string(&data->keep, file_name))
+		return;
 	if (unlink(full_path))
 		die_errno(_("failed to remove %s"), full_path);
 }
@@ -780,48 +776,40 @@ static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUS
 void clear_midx_files_ext(struct odb_source *source, const char *ext,
 			  const char *keep_hash)
 {
-	struct clear_midx_data data;
-	memset(&data, 0, sizeof(struct clear_midx_data));
-
-	if (keep_hash) {
-		ALLOC_ARRAY(data.keep, 1);
-
-		data.keep[0] = xstrfmt("multi-pack-index-%s.%s", keep_hash, ext);
-		data.keep_nr = 1;
-	}
-	data.ext = ext;
-
-	for_each_file_in_pack_dir(source->path,
-				  clear_midx_file_ext,
-				  &data);
+	struct clear_midx_data data = {
+		.keep = STRING_LIST_INIT_NODUP,
+		.ext = ext,
+	};
 
 	if (keep_hash)
-		free(data.keep[0]);
-	free(data.keep);
+		string_list_insert(&data.keep, xstrfmt("multi-pack-index-%s.%s",
+						       keep_hash, ext));
+
+	for_each_file_in_pack_dir(source->path, clear_midx_file_ext, &data);
+
+	string_list_clear(&data.keep, 0);
 }
 
 void clear_incremental_midx_files_ext(struct odb_source *source, const char *ext,
 				      char **keep_hashes,
 				      uint32_t hashes_nr)
 {
-	struct clear_midx_data data;
+	struct clear_midx_data data = {
+		.keep = STRING_LIST_INIT_NODUP,
+		.ext = ext,
+	};
 	uint32_t i;
 
-	memset(&data, 0, sizeof(struct clear_midx_data));
-
-	ALLOC_ARRAY(data.keep, hashes_nr);
 	for (i = 0; i < hashes_nr; i++)
-		data.keep[i] = xstrfmt("multi-pack-index-%s.%s", keep_hashes[i],
-				       ext);
-	data.keep_nr = hashes_nr;
-	data.ext = ext;
+		string_list_append(&data.keep,
+				   xstrfmt("multi-pack-index-%s.%s",
+					   keep_hashes[i], ext));
+	string_list_sort(&data.keep);
 
 	for_each_file_in_pack_subdir(source->path, "multi-pack-index.d",
 				     clear_midx_file_ext, &data);
 
-	for (i = 0; i < hashes_nr; i++)
-		free(data.keep[i]);
-	free(data.keep);
+	string_list_clear(&data.keep, 0);
 }
 
 void clear_midx_file(struct repository *r)
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 02/14] strvec: introduce `strvec_init_alloc()`

[Taylor Blau]

When the caller knows upfront how many elements will be pushed onto a
`strvec`, it is useful to pre-allocate enough space in the array to fit
that many elements (and one additional slot to store NULL, indicating
the end of the list.)

Introduce `strvec_init_alloc()`, which allocates the backing array large
enough to hold `alloc` elements and the termination marker without
further reallocation.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 strvec.c | 7 +++++++
 strvec.h | 5 +++++
 2 files changed, 12 insertions(+)

diff --git a/strvec.c b/strvec.c
index f8de79f5579..f7f32a53b56 100644
--- a/strvec.c
+++ b/strvec.c
@@ -10,6 +10,13 @@ void strvec_init(struct strvec *array)
 	memcpy(array, &blank, sizeof(*array));
 }
 
+void strvec_init_alloc(struct strvec *array, size_t alloc)
+{
+	CALLOC_ARRAY(array->v, st_add(alloc, 1));
+	array->nr = 0;
+	array->alloc = alloc + 1;
+}
+
 void strvec_push_nodup(struct strvec *array, char *value)
 {
 	if (array->v == empty_strvec)
diff --git a/strvec.h b/strvec.h
index f74e061e141..34cb1f939f0 100644
--- a/strvec.h
+++ b/strvec.h
@@ -43,6 +43,11 @@ struct strvec {
  */
 void strvec_init(struct strvec *);
 
+/*
+ * Initializes an array large enough to store `alloc` elements.
+ */
+void strvec_init_alloc(struct strvec *, size_t alloc);
+
 /* Push a copy of a string onto the end of the array. */
 const char *strvec_push(struct strvec *, const char *);
 
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 03/14] midx: use `strvec` for `keep_hashes`

[Taylor Blau]

The `keep_hashes` array in `write_midx_internal()` accumulates the
checksums of MIDX files that should be retained when pruning stale
entries from the MIDX chain. For similar reasons as in a previous
commit, rewrite this using a strvec, requiring us to pass one fewer
parameter.

Unlike the aforementioned previous commit, use a `strvec` instead of a
`string_list`, which provides a more ergonomic interface to adjust the
values at a particular indice. The ordering is important here, as this
value is used to determine the contents of the resulting
`multi-pack-index-chain` file when writing with "--incremental".

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 midx-write.c | 51 ++++++++++++++++++++++++++-------------------------
 midx.c       | 20 ++++++++++----------
 2 files changed, 36 insertions(+), 35 deletions(-)

diff --git a/midx-write.c b/midx-write.c
index 0ff2e45aa7a..7fdc4d31243 100644
--- a/midx-write.c
+++ b/midx-write.c
@@ -29,8 +29,7 @@ extern void clear_midx_files_ext(struct odb_source *source, const char *ext,
 				 const char *keep_hash);
 extern void clear_incremental_midx_files_ext(struct odb_source *source,
 					     const char *ext,
-					     const char **keep_hashes,
-					     uint32_t hashes_nr);
+					     const struct strvec *keep_hashes);
 extern int cmp_idx_or_pack_name(const char *idx_or_pack_name,
 				const char *idx_name);
 
@@ -1109,8 +1108,7 @@ static int link_midx_to_chain(struct multi_pack_index *m)
 }
 
 static void clear_midx_files(struct odb_source *source,
-			     const char **hashes, uint32_t hashes_nr,
-			     unsigned incremental)
+			     const struct strvec *hashes, unsigned incremental)
 {
 	/*
 	 * if incremental:
@@ -1124,13 +1122,15 @@ static void clear_midx_files(struct odb_source *source,
 	 */
 	struct strbuf buf = STRBUF_INIT;
 	const char *exts[] = { MIDX_EXT_BITMAP, MIDX_EXT_REV, MIDX_EXT_MIDX };
-	uint32_t i, j;
+	uint32_t i;
 
 	for (i = 0; i < ARRAY_SIZE(exts); i++) {
-		clear_incremental_midx_files_ext(source, exts[i],
-						 hashes, hashes_nr);
-		for (j = 0; j < hashes_nr; j++)
-			clear_midx_files_ext(source, exts[i], hashes[j]);
+		clear_incremental_midx_files_ext(source, exts[i], hashes);
+		if (hashes) {
+			for (size_t j = 0; j < hashes->nr; j++)
+				clear_midx_files_ext(source, exts[i],
+						     hashes->v[j]);
+		}
 	}
 
 	if (incremental)
@@ -1266,7 +1266,7 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	int pack_name_concat_len = 0;
 	int dropped_packs = 0;
 	int result = -1;
-	const char **keep_hashes = NULL;
+	struct strvec keep_hashes = STRVEC_INIT;
 	size_t keep_hashes_nr = 0;
 	struct chunkfile *cf;
 
@@ -1721,7 +1721,7 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	} else {
 		keep_hashes_nr = ctx.num_multi_pack_indexes_before + 1;
 	}
-	CALLOC_ARRAY(keep_hashes, keep_hashes_nr);
+	strvec_init_alloc(&keep_hashes, keep_hashes_nr);
 
 	if (ctx.incremental) {
 		FILE *chainf = fdopen_lock_file(&lk, "w");
@@ -1758,39 +1758,45 @@ static int write_midx_internal(struct write_midx_opts *opts)
 			for (i = 0; i < num_layers_before_from; i++) {
 				uint32_t j = num_layers_before_from - i - 1;
 
-				keep_hashes[j] = xstrdup(midx_get_checksum_hex(m));
+				keep_hashes.v[j] = xstrdup(midx_get_checksum_hex(m));
+				keep_hashes.nr++;
 				m = m->base_midx;
 			}
 
-			keep_hashes[i] = xstrdup(hash_to_hex_algop(midx_hash,
+			keep_hashes.v[i] = xstrdup(hash_to_hex_algop(midx_hash,
 								   r->hash_algo));
+			keep_hashes.nr++;
 
 			i = 0;
 			for (m = ctx.m;
 			     m && midx_hashcmp(m, ctx.compact_to, r->hash_algo);
 			     m = m->base_midx) {
-				keep_hashes[keep_hashes_nr - i - 1] =
+				keep_hashes.v[keep_hashes_nr - i - 1] =
 					xstrdup(midx_get_checksum_hex(m));
+				keep_hashes.nr++;
 				i++;
 			}
 		} else {
-			keep_hashes[ctx.num_multi_pack_indexes_before] =
+			keep_hashes.v[ctx.num_multi_pack_indexes_before] =
 				xstrdup(hash_to_hex_algop(midx_hash,
 							  r->hash_algo));
+			keep_hashes.nr++;
 
 			for (uint32_t i = 0; i < ctx.num_multi_pack_indexes_before; i++) {
 				uint32_t j = ctx.num_multi_pack_indexes_before - i - 1;
 
-				keep_hashes[j] = xstrdup(midx_get_checksum_hex(m));
+				keep_hashes.v[j] = xstrdup(midx_get_checksum_hex(m));
+				keep_hashes.nr++;
 				m = m->base_midx;
 			}
 		}
 
 		for (uint32_t i = 0; i < keep_hashes_nr; i++)
-			fprintf(get_lock_file_fp(&lk), "%s\n", keep_hashes[i]);
+			fprintf(get_lock_file_fp(&lk), "%s\n", keep_hashes.v[i]);
 	} else {
-		keep_hashes[ctx.num_multi_pack_indexes_before] =
+		keep_hashes.v[ctx.num_multi_pack_indexes_before] =
 			xstrdup(hash_to_hex_algop(midx_hash, r->hash_algo));
+		keep_hashes.nr++;
 	}
 
 	if (ctx.m || ctx.base_midx)
@@ -1799,8 +1805,7 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	if (commit_lock_file(&lk) < 0)
 		die_errno(_("could not write multi-pack-index"));
 
-	clear_midx_files(opts->source, keep_hashes, keep_hashes_nr,
-			 ctx.incremental);
+	clear_midx_files(opts->source, &keep_hashes, ctx.incremental);
 	result = 0;
 
 cleanup:
@@ -1816,11 +1821,7 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	free(ctx.entries);
 	free(ctx.pack_perm);
 	free(ctx.pack_order);
-	if (keep_hashes) {
-		for (uint32_t i = 0; i < keep_hashes_nr; i++)
-			free((char *)keep_hashes[i]);
-		free(keep_hashes);
-	}
+	strvec_clear(&keep_hashes);
 	strbuf_release(&midx_name);
 	close_midx(midx_to_free);
 
diff --git a/midx.c b/midx.c
index c5e3553e2bb..20732c256a6 100644
--- a/midx.c
+++ b/midx.c
@@ -12,6 +12,7 @@
 #include "chunk-format.h"
 #include "pack-bitmap.h"
 #include "pack-revindex.h"
+#include "strvec.h"
 
 #define MIDX_PACK_ERROR ((void *)(intptr_t)-1)
 
@@ -19,8 +20,7 @@ int midx_checksum_valid(struct multi_pack_index *m);
 void clear_midx_files_ext(struct odb_source *source, const char *ext,
 			  const char *keep_hash);
 void clear_incremental_midx_files_ext(struct odb_source *source, const char *ext,
-				      char **keep_hashes,
-				      uint32_t hashes_nr);
+				      const struct strvec *keep_hashes);
 int cmp_idx_or_pack_name(const char *idx_or_pack_name,
 			 const char *idx_name);
 
@@ -791,20 +791,20 @@ void clear_midx_files_ext(struct odb_source *source, const char *ext,
 }
 
 void clear_incremental_midx_files_ext(struct odb_source *source, const char *ext,
-				      char **keep_hashes,
-				      uint32_t hashes_nr)
+				      const struct strvec *keep_hashes)
 {
 	struct clear_midx_data data = {
 		.keep = STRING_LIST_INIT_NODUP,
 		.ext = ext,
 	};
-	uint32_t i;
 
-	for (i = 0; i < hashes_nr; i++)
-		string_list_append(&data.keep,
-				   xstrfmt("multi-pack-index-%s.%s",
-					   keep_hashes[i], ext));
-	string_list_sort(&data.keep);
+	if (keep_hashes) {
+		for (size_t i = 0; i < keep_hashes->nr; i++)
+			string_list_append(&data.keep,
+					   xstrfmt("multi-pack-index-%s.%s",
+						   keep_hashes->v[i], ext));
+		string_list_sort(&data.keep);
+	}
 
 	for_each_file_in_pack_subdir(source->path, "multi-pack-index.d",
 				     clear_midx_file_ext, &data);
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 04/14] midx: introduce `--checksum-only` for incremental MIDX writes

[Taylor Blau]

When writing an incremental MIDX layer, the MIDX machinery writes the
new layer into the multi-pack-index.d directory and then updates the
multi-pack-index-chain file to include the freshly written layer.

Future callers however may not wish to immediately update the MIDX chain
itself, preferring instead to write out new layer(s) itself before
atomically updating the chain. Concretely, the new incremental
MIDX-based repacking strategy will want to do exactly this (that is,
assemble the new MIDX chain itself before writing a new chain file and
atomically linking it into place).

Introduce a `--checksum-only` flag that:

 * writes the new MIDX layer into the multi-pack-index.d directory

 * prints its checksum

 * does not update the multi-pack-index-chain file.

The MIDX chain file (and thus, the lock protecting it) remain untouched,
allowing callers to assemble the chain themselves. This flag requires
`--incremental`, since the notion of a separate layer only makes sense
for incremental MIDXs.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 Documentation/git-multi-pack-index.adoc |  4 +--
 builtin/multi-pack-index.c              | 28 ++++++++++++++++--
 midx-write.c                            | 38 ++++++++++++++++---------
 midx.h                                  |  1 +
 t/t5334-incremental-multi-pack-index.sh | 17 +++++++++++
 t/t5335-compact-multi-pack-index.sh     | 34 ++++++++++++++++++++++
 6 files changed, 105 insertions(+), 17 deletions(-)

diff --git a/Documentation/git-multi-pack-index.adoc b/Documentation/git-multi-pack-index.adoc
index 61256830141..657e0639f6a 100644
--- a/Documentation/git-multi-pack-index.adoc
+++ b/Documentation/git-multi-pack-index.adoc
@@ -11,9 +11,9 @@ SYNOPSIS
 [verse]
 'git multi-pack-index' [<options>] write [--preferred-pack=<pack>]
 		         [--[no-]bitmap] [--[no-]incremental] [--[no-]stdin-packs]
-		         [--refs-snapshot=<path>]
+		         [--refs-snapshot=<path>] [--[no-]checksum-only]
 'git multi-pack-index' [<options>] compact [--[no-]incremental]
-		         [--[no-]bitmap] <from> <to>
+		         [--[no-]bitmap] [--[no-]checksum-only] <from> <to>
 'git multi-pack-index' [<options>] verify
 'git multi-pack-index' [<options>] expire
 'git multi-pack-index' [<options>] repack [--batch-size=<size>]
diff --git a/builtin/multi-pack-index.c b/builtin/multi-pack-index.c
index 2f24c113c8f..ee1ddf1386f 100644
--- a/builtin/multi-pack-index.c
+++ b/builtin/multi-pack-index.c
@@ -15,11 +15,11 @@
 #define BUILTIN_MIDX_WRITE_USAGE \
 	N_("git multi-pack-index [<options>] write [--preferred-pack=<pack>]\n" \
 	   "  [--[no-]bitmap] [--[no-]incremental] [--[no-]stdin-packs]\n" \
-	   "  [--refs-snapshot=<path>]")
+	   "  [--refs-snapshot=<path>] [--[no-]checksum-only]")
 
 #define BUILTIN_MIDX_COMPACT_USAGE \
 	N_("git multi-pack-index [<options>] compact [--[no-]incremental]\n" \
-	   "  [--[no-]bitmap] <from> <to>")
+	   "  [--[no-]bitmap] [--[no-]checksum-only] <from> <to>")
 
 #define BUILTIN_MIDX_VERIFY_USAGE \
 	N_("git multi-pack-index [<options>] verify")
@@ -152,6 +152,9 @@ static int cmd_multi_pack_index_write(int argc, const char **argv,
 			MIDX_WRITE_BITMAP | MIDX_WRITE_REV_INDEX),
 		OPT_BIT(0, "incremental", &opts.flags,
 			N_("write a new incremental MIDX"), MIDX_WRITE_INCREMENTAL),
+		OPT_BIT(0, "checksum-only", &opts.flags,
+			N_("write a MIDX layer without updating the MIDX chain"),
+			MIDX_WRITE_CHECKSUM_ONLY),
 		OPT_BOOL(0, "stdin-packs", &opts.stdin_packs,
 			 N_("write multi-pack index containing only given indexes")),
 		OPT_FILENAME(0, "refs-snapshot", &opts.refs_snapshot,
@@ -177,6 +180,15 @@ static int cmd_multi_pack_index_write(int argc, const char **argv,
 	if (argc)
 		usage_with_options(builtin_multi_pack_index_write_usage,
 				   options);
+
+	if (opts.flags & MIDX_WRITE_CHECKSUM_ONLY &&
+	    !(opts.flags & MIDX_WRITE_INCREMENTAL)) {
+		error(_("cannot use %s without %s"),
+		      "--checksum-only", "--incremental");
+		usage_with_options(builtin_multi_pack_index_write_usage,
+				   options);
+	}
+
 	source = handle_object_dir_option(repo);
 
 	FREE_AND_NULL(options);
@@ -220,6 +232,9 @@ static int cmd_multi_pack_index_compact(int argc, const char **argv,
 			MIDX_WRITE_BITMAP | MIDX_WRITE_REV_INDEX),
 		OPT_BIT(0, "incremental", &opts.flags,
 			N_("write a new incremental MIDX"), MIDX_WRITE_INCREMENTAL),
+		OPT_BIT(0, "checksum-only", &opts.flags,
+			N_("write a MIDX layer without updating the MIDX chain"),
+			MIDX_WRITE_CHECKSUM_ONLY),
 		OPT_END(),
 	};
 
@@ -238,6 +253,15 @@ static int cmd_multi_pack_index_compact(int argc, const char **argv,
 	if (argc != 2)
 		usage_with_options(builtin_multi_pack_index_compact_usage,
 				   options);
+
+	if (opts.flags & MIDX_WRITE_CHECKSUM_ONLY &&
+	    !(opts.flags & MIDX_WRITE_INCREMENTAL)) {
+		error(_("cannot use %s without %s"),
+		      "--checksum-only", "--incremental");
+		usage_with_options(builtin_multi_pack_index_compact_usage,
+				   options);
+	}
+
 	source = handle_object_dir_option(the_repository);
 
 	FREE_AND_NULL(options);
diff --git a/midx-write.c b/midx-write.c
index 7fdc4d31243..9cf085ca333 100644
--- a/midx-write.c
+++ b/midx-write.c
@@ -1598,11 +1598,14 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	}
 
 	if (ctx.incremental) {
-		struct strbuf lock_name = STRBUF_INIT;
+		if (!(opts->flags & MIDX_WRITE_CHECKSUM_ONLY)) {
+			struct strbuf lock_name = STRBUF_INIT;
 
-		get_midx_chain_filename(opts->source, &lock_name);
-		hold_lock_file_for_update(&lk, lock_name.buf, LOCK_DIE_ON_ERROR);
-		strbuf_release(&lock_name);
+			get_midx_chain_filename(opts->source, &lock_name);
+			hold_lock_file_for_update(&lk, lock_name.buf,
+						  LOCK_DIE_ON_ERROR);
+			strbuf_release(&lock_name);
+		}
 
 		incr = mks_tempfile_m(midx_name.buf, 0444);
 		if (!incr) {
@@ -1723,14 +1726,19 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	}
 	strvec_init_alloc(&keep_hashes, keep_hashes_nr);
 
+	if (opts->flags & MIDX_WRITE_CHECKSUM_ONLY)
+		printf("%s\n", hash_to_hex_algop(midx_hash, r->hash_algo));
+
 	if (ctx.incremental) {
-		FILE *chainf = fdopen_lock_file(&lk, "w");
 		struct strbuf final_midx_name = STRBUF_INIT;
 		struct multi_pack_index *m = ctx.base_midx;
 
-		if (!chainf) {
-			error_errno(_("unable to open multi-pack-index chain file"));
-			goto cleanup;
+		if (!(opts->flags & MIDX_WRITE_CHECKSUM_ONLY)) {
+			FILE *chainf = fdopen_lock_file(&lk, "w");
+			if (!chainf) {
+				error_errno(_("unable to open multi-pack-index chain file"));
+				goto cleanup;
+			}
 		}
 
 		if (link_midx_to_chain(ctx.base_midx) < 0)
@@ -1791,8 +1799,10 @@ static int write_midx_internal(struct write_midx_opts *opts)
 			}
 		}
 
-		for (uint32_t i = 0; i < keep_hashes_nr; i++)
-			fprintf(get_lock_file_fp(&lk), "%s\n", keep_hashes.v[i]);
+		if (!(opts->flags & MIDX_WRITE_CHECKSUM_ONLY))
+			for (uint32_t i = 0; i < keep_hashes_nr; i++)
+				fprintf(get_lock_file_fp(&lk), "%s\n",
+					keep_hashes.v[i]);
 	} else {
 		keep_hashes.v[ctx.num_multi_pack_indexes_before] =
 			xstrdup(hash_to_hex_algop(midx_hash, r->hash_algo));
@@ -1802,10 +1812,12 @@ static int write_midx_internal(struct write_midx_opts *opts)
 	if (ctx.m || ctx.base_midx)
 		odb_close(ctx.repo->objects);
 
-	if (commit_lock_file(&lk) < 0)
-		die_errno(_("could not write multi-pack-index"));
+	if (!(opts->flags & MIDX_WRITE_CHECKSUM_ONLY)) {
+		if (commit_lock_file(&lk) < 0)
+			die_errno(_("could not write multi-pack-index"));
 
-	clear_midx_files(opts->source, &keep_hashes, ctx.incremental);
+		clear_midx_files(opts->source, &keep_hashes, ctx.incremental);
+	}
 	result = 0;
 
 cleanup:
diff --git a/midx.h b/midx.h
index 08f3728e520..9f1acd7ace4 100644
--- a/midx.h
+++ b/midx.h
@@ -83,6 +83,7 @@ struct multi_pack_index {
 #define MIDX_WRITE_BITMAP_LOOKUP_TABLE (1 << 4)
 #define MIDX_WRITE_INCREMENTAL (1 << 5)
 #define MIDX_WRITE_COMPACT (1 << 6)
+#define MIDX_WRITE_CHECKSUM_ONLY (1 << 7)
 
 #define MIDX_EXT_REV "rev"
 #define MIDX_EXT_BITMAP "bitmap"
diff --git a/t/t5334-incremental-multi-pack-index.sh b/t/t5334-incremental-multi-pack-index.sh
index d30d7253d6f..96449178c07 100755
--- a/t/t5334-incremental-multi-pack-index.sh
+++ b/t/t5334-incremental-multi-pack-index.sh
@@ -95,6 +95,23 @@ test_expect_success 'show object from second pack' '
 	git cat-file -p 2.2
 '
 
+test_expect_success 'write MIDX layer with --checksum-only' '
+	test_commit checksum-only &&
+	git repack -d &&
+
+	cp "$midx_chain" "$midx_chain.bak" &&
+	layer="$(git multi-pack-index write --bitmap --incremental \
+		--checksum-only)" &&
+
+	test_cmp "$midx_chain.bak" "$midx_chain" &&
+	test_path_is_file "$midxdir/multi-pack-index-$layer.midx"
+'
+
+test_expect_success 'write non-incremental MIDX layer with --checksum-only' '
+	test_must_fail git multi-pack-index write --bitmap --checksum-only 2>err &&
+	test_grep "cannot use --checksum-only without --incremental" err
+'
+
 for reuse in false single multi
 do
 	test_expect_success "full clone (pack.allowPackReuse=$reuse)" '
diff --git a/t/t5335-compact-multi-pack-index.sh b/t/t5335-compact-multi-pack-index.sh
index 40f3844282f..55b9773568b 100755
--- a/t/t5335-compact-multi-pack-index.sh
+++ b/t/t5335-compact-multi-pack-index.sh
@@ -290,4 +290,38 @@ test_expect_success 'MIDX compaction with bitmaps (non-trivial)' '
 	)
 '
 
+test_expect_success 'MIDX compaction with --checksum-only' '
+	git init midx-compact-with--checksum-only &&
+	(
+		cd midx-compact-with--checksum-only &&
+
+		write_packs A B C D &&
+
+		test_line_count = 4 $midx_chain &&
+		cp "$midx_chain" "$midx_chain".bak &&
+
+		layer="$(git multi-pack-index compact --incremental \
+			--checksum-only \
+			"$(nth_line 2 "$midx_chain")" \
+			"$(nth_line 3 "$midx_chain")")" &&
+
+		test_cmp "$midx_chain.bak" "$midx_chain" &&
+
+		# After writing the new layer, insert it into the chain
+		# manually. This is done in order to make $layer visible
+		# to the read-midx test helper below, and matches what
+		# the MIDX command would do without --checksum-only.
+		{
+			nth_line 1 "$midx_chain.bak" &&
+			echo $layer &&
+			nth_line 4 "$midx_chain.bak"
+		} >$midx_chain &&
+
+		test-tool read-midx $objdir $layer >midx.data &&
+		grep "^pack-B-.*\.idx" midx.data &&
+		grep "^pack-C-.*\.idx" midx.data
+
+	)
+'
+
 test_done
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 05/14] midx: support custom `--base` for incremental MIDX writes

[Taylor Blau]

Both `compact` and `write --incremental` fix the base of the resulting
MIDX layer: `compact` always places the compacted result on top of
"from's" immediate parent in the chain, and `write --incremental` always
appends a new layer to the existing tip. In both cases the base is not
configurable.

Future callers need additional flexibility. For instance, the incremental
MIDX-based repacking code may wish to write a layer based on some
intermediate ancestor rather than the current tip, or produce a root
layer when replacing the bottommost entries in the chain.

Introduce a new `--base` option for both subcommands to specify the
checksum of the MIDX layer to use as the base. The given checksum must
refer to a valid layer in the MIDX chain that is an ancestor of the
topmost layer being written or compacted.

The special value "none" is accepted to produce a root layer with no
parent. This will be needed when the incremental repacking machinery
determines that the bottommost layers of the chain should be replaced.

If no `--base` is given, behavior is unchanged: `compact` uses "from's"
immediate parent in the chain, and `write` appends to the existing tip.

For the `write` subcommand, `--base` requires `--checksum-only`. A plain
`write --incremental` appends a new layer to the live chain tip with no
mechanism to atomically replace it; overriding the base would produce a
layer that does not extend the tip, breaking chain invariants. With
`--checksum-only` the chain is left unmodified and the caller is
responsible for assembling a valid chain.

For `compact`, no such restriction applies. The compaction operation
atomically replaces the compacted range in the chain file, so writing
the result on top of any valid ancestor preserves chain invariants.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 Documentation/git-multi-pack-index.adoc | 17 +++++-
 builtin/multi-pack-index.c              | 24 +++++++--
 midx-write.c                            | 34 ++++++++++--
 midx.h                                  |  5 +-
 t/t5334-incremental-multi-pack-index.sh | 30 +++++++++++
 t/t5335-compact-multi-pack-index.sh     | 71 +++++++++++++++++++++++++
 6 files changed, 172 insertions(+), 9 deletions(-)

diff --git a/Documentation/git-multi-pack-index.adoc b/Documentation/git-multi-pack-index.adoc
index 657e0639f6a..635105ad801 100644
--- a/Documentation/git-multi-pack-index.adoc
+++ b/Documentation/git-multi-pack-index.adoc
@@ -12,8 +12,10 @@ SYNOPSIS
 'git multi-pack-index' [<options>] write [--preferred-pack=<pack>]
 		         [--[no-]bitmap] [--[no-]incremental] [--[no-]stdin-packs]
 		         [--refs-snapshot=<path>] [--[no-]checksum-only]
+			 [--base=<checksum>]
 'git multi-pack-index' [<options>] compact [--[no-]incremental]
-		         [--[no-]bitmap] [--[no-]checksum-only] <from> <to>
+		         [--[no-]bitmap] [--base=<checksum>] [--[no-]checksum-only]
+			 <from> <to>
 'git multi-pack-index' [<options>] verify
 'git multi-pack-index' [<options>] expire
 'git multi-pack-index' [<options>] repack [--batch-size=<size>]
@@ -83,6 +85,13 @@ marker).
 		and packs not present in an existing MIDX layer.
 		Migrates non-incremental MIDXs to incremental ones when
 		necessary.
+
+	--base=<checksum>::
+		Specify the checksum of an existing MIDX layer to use
+		as the base when writing a new incremental layer.
+		The special value `none` indicates that the new layer
+		should have no base (i.e., it becomes a root layer).
+		Requires `--checksum-only`.
 --
 
 compact::
@@ -97,6 +106,12 @@ compact::
 
 	--[no-]bitmap::
 		Control whether or not a multi-pack bitmap is written.
+
+	--base=<checksum>::
+		Specify the checksum of an existing MIDX layer to use
+		as the base for the compacted result, instead of using
+		the immediate parent of `<from>`. The special value
+		`none` indicates that the result should have no base.
 --
 
 verify::
diff --git a/builtin/multi-pack-index.c b/builtin/multi-pack-index.c
index ee1ddf1386f..4fc53a5971c 100644
--- a/builtin/multi-pack-index.c
+++ b/builtin/multi-pack-index.c
@@ -15,11 +15,13 @@
 #define BUILTIN_MIDX_WRITE_USAGE \
 	N_("git multi-pack-index [<options>] write [--preferred-pack=<pack>]\n" \
 	   "  [--[no-]bitmap] [--[no-]incremental] [--[no-]stdin-packs]\n" \
-	   "  [--refs-snapshot=<path>] [--[no-]checksum-only]")
+	   "  [--refs-snapshot=<path>] [--[no-]checksum-only]\n" \
+	   "  [--base=<checksum>]")
 
 #define BUILTIN_MIDX_COMPACT_USAGE \
 	N_("git multi-pack-index [<options>] compact [--[no-]incremental]\n" \
-	   "  [--[no-]bitmap] [--[no-]checksum-only] <from> <to>")
+	   "  [--[no-]bitmap] [--base=<checksum>] [--[no-]checksum-only]\n" \
+	   "  <from> <to>")
 
 #define BUILTIN_MIDX_VERIFY_USAGE \
 	N_("git multi-pack-index [<options>] verify")
@@ -62,6 +64,7 @@ static char const * const builtin_multi_pack_index_usage[] = {
 static struct opts_multi_pack_index {
 	char *object_dir;
 	const char *preferred_pack;
+	const char *incremental_base;
 	char *refs_snapshot;
 	unsigned long batch_size;
 	unsigned flags;
@@ -150,6 +153,8 @@ static int cmd_multi_pack_index_write(int argc, const char **argv,
 			   N_("pack for reuse when computing a multi-pack bitmap")),
 		OPT_BIT(0, "bitmap", &opts.flags, N_("write multi-pack bitmap"),
 			MIDX_WRITE_BITMAP | MIDX_WRITE_REV_INDEX),
+		OPT_STRING(0, "base", &opts.incremental_base, N_("checksum"),
+			   N_("base MIDX for incremental writes")),
 		OPT_BIT(0, "incremental", &opts.flags,
 			N_("write a new incremental MIDX"), MIDX_WRITE_INCREMENTAL),
 		OPT_BIT(0, "checksum-only", &opts.flags,
@@ -189,6 +194,13 @@ static int cmd_multi_pack_index_write(int argc, const char **argv,
 				   options);
 	}
 
+	if (opts.incremental_base &&
+	    !(opts.flags & MIDX_WRITE_CHECKSUM_ONLY)) {
+		error(_("cannot use --base without --checksum-only"));
+		usage_with_options(builtin_multi_pack_index_write_usage,
+				   options);
+	}
+
 	source = handle_object_dir_option(repo);
 
 	FREE_AND_NULL(options);
@@ -200,7 +212,8 @@ static int cmd_multi_pack_index_write(int argc, const char **argv,
 
 		ret = write_midx_file_only(source, &packs,
 					   opts.preferred_pack,
-					   opts.refs_snapshot, opts.flags);
+					   opts.refs_snapshot,
+					   opts.incremental_base, opts.flags);
 
 		string_list_clear(&packs, 0);
 		free(opts.refs_snapshot);
@@ -228,6 +241,8 @@ static int cmd_multi_pack_index_compact(int argc, const char **argv,
 
 	struct option *options;
 	static struct option builtin_multi_pack_index_compact_options[] = {
+		OPT_STRING(0, "base", &opts.incremental_base, N_("checksum"),
+			   N_("base MIDX for incremental writes")),
 		OPT_BIT(0, "bitmap", &opts.flags, N_("write multi-pack bitmap"),
 			MIDX_WRITE_BITMAP | MIDX_WRITE_REV_INDEX),
 		OPT_BIT(0, "incremental", &opts.flags,
@@ -289,7 +304,8 @@ static int cmd_multi_pack_index_compact(int argc, const char **argv,
 			die(_("MIDX %s must be an ancestor of %s"), argv[0], argv[1]);
 	}
 
-	ret = write_midx_file_compact(source, from_midx, to_midx, opts.flags);
+	ret = write_midx_file_compact(source, from_midx, to_midx,
+				      opts.incremental_base, opts.flags);
 
 	return ret;
 }
diff --git a/midx-write.c b/midx-write.c
index 9cf085ca333..c3e70d76d7c 100644
--- a/midx-write.c
+++ b/midx-write.c
@@ -1245,6 +1245,7 @@ struct write_midx_opts {
 
 	const char *preferred_pack_name;
 	const char *refs_snapshot;
+	const char *incremental_base;
 	unsigned flags;
 };
 
@@ -1327,11 +1328,32 @@ static int write_midx_internal(struct write_midx_opts *opts)
 
 	/*
 	 * If compacting MIDX layer(s) in the range [from, to], then the
-	 * compacted MIDX will share the same base MIDX as 'from'.
+	 * compacted MIDX will share the same base MIDX as 'from',
+	 * unless a custom --base is specified (see below).
 	 */
 	if (ctx.compact)
 		ctx.base_midx = ctx.compact_from->base_midx;
 
+	if (opts->incremental_base) {
+		if (!strcmp(opts->incremental_base, "none")) {
+			ctx.base_midx = NULL;
+		} else {
+			while (ctx.base_midx) {
+				const char *cmp = midx_get_checksum_hex(ctx.base_midx);
+				if (!strcmp(opts->incremental_base, cmp))
+					break;
+
+				ctx.base_midx = ctx.base_midx->base_midx;
+			}
+
+			if (!ctx.base_midx) {
+				error(_("could not find base MIDX '%s'"),
+				      opts->incremental_base);
+				goto cleanup;
+			}
+		}
+	}
+
 	ctx.nr = 0;
 	ctx.alloc = ctx.m ? ctx.m->num_packs + ctx.m->num_packs_in_base : 16;
 	ctx.info = NULL;
@@ -1844,7 +1866,8 @@ static int write_midx_internal(struct write_midx_opts *opts)
 
 int write_midx_file(struct odb_source *source,
 		    const char *preferred_pack_name,
-		    const char *refs_snapshot, unsigned flags)
+		    const char *refs_snapshot,
+		    unsigned flags)
 {
 	struct write_midx_opts opts = {
 		.source = source,
@@ -1859,13 +1882,16 @@ int write_midx_file(struct odb_source *source,
 int write_midx_file_only(struct odb_source *source,
 			 struct string_list *packs_to_include,
 			 const char *preferred_pack_name,
-			 const char *refs_snapshot, unsigned flags)
+			 const char *refs_snapshot,
+			 const char *incremental_base,
+			 unsigned flags)
 {
 	struct write_midx_opts opts = {
 		.source = source,
 		.packs_to_include = packs_to_include,
 		.preferred_pack_name = preferred_pack_name,
 		.refs_snapshot = refs_snapshot,
+		.incremental_base = incremental_base,
 		.flags = flags,
 	};
 
@@ -1875,12 +1901,14 @@ int write_midx_file_only(struct odb_source *source,
 int write_midx_file_compact(struct odb_source *source,
 			    struct multi_pack_index *from,
 			    struct multi_pack_index *to,
+			    const char *incremental_base,
 			    unsigned flags)
 {
 	struct write_midx_opts opts = {
 		.source = source,
 		.compact_from = from,
 		.compact_to = to,
+		.incremental_base = incremental_base,
 		.flags = flags | MIDX_WRITE_COMPACT,
 	};
 
diff --git a/midx.h b/midx.h
index 9f1acd7ace4..e4a75ff2bef 100644
--- a/midx.h
+++ b/midx.h
@@ -132,10 +132,13 @@ int write_midx_file(struct odb_source *source,
 int write_midx_file_only(struct odb_source *source,
 			 struct string_list *packs_to_include,
 			 const char *preferred_pack_name,
-			 const char *refs_snapshot, unsigned flags);
+			 const char *refs_snapshot,
+			 const char *incremental_base,
+			 unsigned flags);
 int write_midx_file_compact(struct odb_source *source,
 			    struct multi_pack_index *from,
 			    struct multi_pack_index *to,
+			    const char *incremental_base,
 			    unsigned flags);
 void clear_midx_file(struct repository *r);
 int verify_midx_file(struct odb_source *source, unsigned flags);
diff --git a/t/t5334-incremental-multi-pack-index.sh b/t/t5334-incremental-multi-pack-index.sh
index 96449178c07..77fb40ade01 100755
--- a/t/t5334-incremental-multi-pack-index.sh
+++ b/t/t5334-incremental-multi-pack-index.sh
@@ -112,6 +112,36 @@ test_expect_success 'write non-incremental MIDX layer with --checksum-only' '
 	test_grep "cannot use --checksum-only without --incremental" err
 '
 
+test_expect_success 'write MIDX layer with --base without --checksum-only' '
+	test_must_fail git multi-pack-index write --bitmap --incremental \
+		--base=none 2>err &&
+	test_grep "cannot use --base without --checksum-only" err
+'
+
+test_expect_success 'write MIDX layer with --base=none and --checksum-only' '
+	test_commit base-none &&
+	git repack -d &&
+
+	cp "$midx_chain" "$midx_chain.bak" &&
+	layer="$(git multi-pack-index write --bitmap --incremental \
+		--checksum-only --base=none)" &&
+
+	test_cmp "$midx_chain.bak" "$midx_chain" &&
+	test_path_is_file "$midxdir/multi-pack-index-$layer.midx"
+'
+
+test_expect_success 'write MIDX layer with --base=<hash> and --checksum-only' '
+	test_commit base-hash &&
+	git repack -d &&
+
+	cp "$midx_chain" "$midx_chain.bak" &&
+	layer="$(git multi-pack-index write --bitmap --incremental \
+		--checksum-only --base="$(nth_line 1 "$midx_chain")")" &&
+
+	test_cmp "$midx_chain.bak" "$midx_chain" &&
+	test_path_is_file "$midxdir/multi-pack-index-$layer.midx"
+'
+
 for reuse in false single multi
 do
 	test_expect_success "full clone (pack.allowPackReuse=$reuse)" '
diff --git a/t/t5335-compact-multi-pack-index.sh b/t/t5335-compact-multi-pack-index.sh
index 55b9773568b..f0d1d68d26c 100755
--- a/t/t5335-compact-multi-pack-index.sh
+++ b/t/t5335-compact-multi-pack-index.sh
@@ -302,6 +302,7 @@ test_expect_success 'MIDX compaction with --checksum-only' '
 
 		layer="$(git multi-pack-index compact --incremental \
 			--checksum-only \
+			--base="$(nth_line 1 "$midx_chain")" \
 			"$(nth_line 2 "$midx_chain")" \
 			"$(nth_line 3 "$midx_chain")")" &&
 
@@ -324,4 +325,74 @@ test_expect_success 'MIDX compaction with --checksum-only' '
 	)
 '
 
+test_expect_success 'MIDX compaction with --base' '
+	git init midx-compact-with--base &&
+	(
+		cd midx-compact-with--base &&
+
+		write_packs A B C D &&
+
+		test_line_count = 4 "$midx_chain" &&
+
+		cp "$midx_chain" "$midx_chain.bak" &&
+
+		git multi-pack-index compact --incremental \
+			--base="$(nth_line 1 "$midx_chain")" \
+			"$(nth_line 3 "$midx_chain")" \
+			"$(nth_line 4 "$midx_chain")" &&
+		test_line_count = 2 $midx_chain &&
+
+		nth_line 1 "$midx_chain.bak" >expect &&
+		nth_line 1 "$midx_chain" >actual &&
+
+		test_cmp expect actual
+	)
+'
+
+test_expect_success 'MIDX compaction with --base=none' '
+	git init midx-compact-base-none &&
+	(
+		cd midx-compact-base-none &&
+
+		write_packs A B C D &&
+
+		test_line_count = 4 $midx_chain &&
+
+		cp "$midx_chain" "$midx_chain".bak &&
+
+		# Compact the two bottommost layers (A and B) into a new
+		# root layer with no parent.
+		git multi-pack-index compact --incremental \
+			--base=none \
+			"$(nth_line 1 "$midx_chain")" \
+			"$(nth_line 2 "$midx_chain")" &&
+
+		test_line_count = 3 $midx_chain &&
+
+		# The upper layers (C and D) should be preserved
+		# unchanged.
+		nth_line 3 "$midx_chain.bak" >expect &&
+		nth_line 4 "$midx_chain.bak" >>expect &&
+		nth_line 2 "$midx_chain" >actual &&
+		nth_line 3 "$midx_chain" >>actual &&
+
+		test_cmp expect actual
+	)
+'
+
+test_expect_success 'MIDX compaction with bogus --base checksum' '
+	git init midx-compact-bogus-base &&
+	(
+		cd midx-compact-bogus-base &&
+
+		write_packs A B C &&
+
+		test_must_fail git multi-pack-index compact --incremental \
+			--base=deadbeef \
+			"$(nth_line 2 "$midx_chain")" \
+			"$(nth_line 3 "$midx_chain")" 2>err &&
+		test_grep "could not find base MIDX" err
+	)
+'
+
 test_done
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 06/14] repack: track the ODB source via existing_packs

[Taylor Blau]

Store the ODB source in the `existing_packs` struct and use that in
place of the raw `repo->objects->sources` access within `cmd_repack()`.

The source used is still assigned from the first source in the list, so
there are no functional changes in this commit. The changes instead
serve two purposes (one immediate, one not):

 - The incremental MIDX-based repacking machinery will need to know what
   source is being used to read the existing MIDX/chain from that source
   (should one exist).

 - In the future, if "git repack" is taught how to operate on other
   object sources, this field will serve as the authoritative value for
   that source.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 builtin/repack.c | 5 ++---
 repack.c         | 2 ++
 repack.h         | 1 +
 3 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/builtin/repack.c b/builtin/repack.c
index f6bb04bef72..44a95b56f23 100644
--- a/builtin/repack.c
+++ b/builtin/repack.c
@@ -402,7 +402,7 @@ int cmd_repack(int argc,
 		 * midx_has_unknown_packs() will make the decision for
 		 * us.
 		 */
-		if (!get_multi_pack_index(repo->objects->sources))
+		if (!get_multi_pack_index(existing.source))
 			midx_must_contain_cruft = 1;
 	}
 
@@ -549,8 +549,7 @@ int cmd_repack(int argc,
 		unsigned flags = 0;
 		if (git_env_bool(GIT_TEST_MULTI_PACK_INDEX_WRITE_INCREMENTAL, 0))
 			flags |= MIDX_WRITE_INCREMENTAL;
-		write_midx_file(repo->objects->sources,
-				NULL, NULL, flags);
+		write_midx_file(existing.source, NULL, NULL, flags);
 	}
 
 cleanup:
diff --git a/repack.c b/repack.c
index 596841027af..2ee6b51420a 100644
--- a/repack.c
+++ b/repack.c
@@ -154,6 +154,8 @@ void existing_packs_collect(struct existing_packs *existing,
 			string_list_append(&existing->non_kept_packs, buf.buf);
 	}
 
+	existing->source = existing->repo->objects->sources;
+
 	string_list_sort(&existing->kept_packs);
 	string_list_sort(&existing->non_kept_packs);
 	string_list_sort(&existing->cruft_packs);
diff --git a/repack.h b/repack.h
index bc9f2e1a5de..c0e9f0ca647 100644
--- a/repack.h
+++ b/repack.h
@@ -56,6 +56,7 @@ struct packed_git;
 
 struct existing_packs {
 	struct repository *repo;
+	struct odb_source *source;
 	struct string_list kept_packs;
 	struct string_list non_kept_packs;
 	struct string_list cruft_packs;
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 07/14] midx: expose `midx_layer_contains_pack()`

[Taylor Blau]

Rename the function `midx_contains_pack_1()` to instead be called
`midx_layer_contains_pack()` and make it accessible. Unlike
`midx_contains_pack()` (which recurses through the entire chain), this
function checks only a single MIDX layer.

This will be used by a subsequent commit to determine whether a given
pack belongs to the tip MIDX layer specifically, rather than to any
layer in the chain.

No functional changes are present in this commit.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 midx.c | 6 +++---
 midx.h | 2 ++
 2 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/midx.c b/midx.c
index 20732c256a6..709fe7c3fd7 100644
--- a/midx.c
+++ b/midx.c
@@ -665,8 +665,8 @@ static int midx_pack_names_cmp(const void *a, const void *b, void *m_)
 		      m->pack_names[*(const size_t *)b]);
 }
 
-static int midx_contains_pack_1(struct multi_pack_index *m,
-				const char *idx_or_pack_name)
+int midx_layer_contains_pack(struct multi_pack_index *m,
+			     const char *idx_or_pack_name)
 {
 	uint32_t first = 0, last = m->num_packs;
 
@@ -707,7 +707,7 @@ static int midx_contains_pack_1(struct multi_pack_index *m,
 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
 {
 	for (; m; m = m->base_midx)
-		if (midx_contains_pack_1(m, idx_or_pack_name))
+		if (midx_layer_contains_pack(m, idx_or_pack_name))
 			return 1;
 	return 0;
 }
diff --git a/midx.h b/midx.h
index e4a75ff2bef..f211a38b9e7 100644
--- a/midx.h
+++ b/midx.h
@@ -119,6 +119,8 @@ struct object_id *nth_midxed_object_oid(struct object_id *oid,
 int fill_midx_entry(struct multi_pack_index *m, const struct object_id *oid, struct pack_entry *e);
 int midx_contains_pack(struct multi_pack_index *m,
 		       const char *idx_or_pack_name);
+int midx_layer_contains_pack(struct multi_pack_index *m,
+			     const char *idx_or_pack_name);
 int midx_preferred_pack(struct multi_pack_index *m, uint32_t *pack_int_id);
 int prepare_multi_pack_index_one(struct odb_source *source);
 
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 08/14] repack-midx: factor out `repack_prepare_midx_command()`

[Taylor Blau]

The `write_midx_included_packs()` function assembles and executes a
`git multi-pack-index write` command, constructing the argument list
inline.

Future commits will introduce additional callers that need to construct
similar `git multi-pack-index` commands (for both `write` and `compact`
subcommands), so extract the common portions of the command setup into a
reusable `repack_prepare_midx_command()` helper.

The extracted helper sets `git_cmd`, pushes the `multi-pack-index`
subcommand and verb, and handles `--progress`/`--no-progress` and
`--bitmap` flags. The remaining arguments that are specific to the
`write` subcommand (such as `--stdin-packs`) are left to the caller.

No functional changes are included in this patch.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 repack-midx.c | 30 +++++++++++++++++++-----------
 1 file changed, 19 insertions(+), 11 deletions(-)

diff --git a/repack-midx.c b/repack-midx.c
index 74bdfa3a6e9..7547efc0651 100644
--- a/repack-midx.c
+++ b/repack-midx.c
@@ -283,6 +283,23 @@ static void remove_redundant_bitmaps(struct string_list *include,
 	strbuf_release(&path);
 }
 
+static void repack_prepare_midx_command(struct child_process *cmd,
+					struct repack_write_midx_opts *opts,
+					const char *verb)
+{
+	cmd->git_cmd = 1;
+
+	strvec_pushl(&cmd->args, "multi-pack-index", verb, NULL);
+
+	if (opts->show_progress)
+		strvec_push(&cmd->args, "--progress");
+	else
+		strvec_push(&cmd->args, "--no-progress");
+
+	if (opts->write_bitmaps)
+		strvec_push(&cmd->args, "--bitmap");
+}
+
 int write_midx_included_packs(struct repack_write_midx_opts *opts)
 {
 	struct child_process cmd = CHILD_PROCESS_INIT;
@@ -297,18 +314,9 @@ int write_midx_included_packs(struct repack_write_midx_opts *opts)
 		goto done;
 
 	cmd.in = -1;
-	cmd.git_cmd = 1;
 
-	strvec_push(&cmd.args, "multi-pack-index");
-	strvec_pushl(&cmd.args, "write", "--stdin-packs", NULL);
-
-	if (opts->show_progress)
-		strvec_push(&cmd.args, "--progress");
-	else
-		strvec_push(&cmd.args, "--no-progress");
-
-	if (opts->write_bitmaps)
-		strvec_push(&cmd.args, "--bitmap");
+	repack_prepare_midx_command(&cmd, opts, "write");
+	strvec_push(&cmd.args, "--stdin-packs");
 
 	if (preferred)
 		strvec_pushf(&cmd.args, "--preferred-pack=%s",
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 09/14] repack-midx: extract `repack_fill_midx_stdin_packs()`

[Taylor Blau]

The function `write_midx_included_packs()` manages the lifecycle of
writing packs to stdin when running `git multi-pack-index write` as a
child process.

Extract a standalone `repack_fill_midx_stdin_packs()` helper, which
handles `--stdin-packs` argument setup, starting the command, writing
pack names to its standard input, and finishing the command.

This simplifies `write_midx_included_packs()` and prepares for a
subsequent commit where the same helper is called with `cmd->out = -1`
to capture the MIDX's checksum from the command's standard output,
which is needed when writing MIDX layers with `--checksum-only`.

No functional changes are included in this patch.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 repack-midx.c | 38 ++++++++++++++++++++++++--------------
 1 file changed, 24 insertions(+), 14 deletions(-)

diff --git a/repack-midx.c b/repack-midx.c
index 7547efc0651..bc5059927f7 100644
--- a/repack-midx.c
+++ b/repack-midx.c
@@ -300,23 +300,42 @@ static void repack_prepare_midx_command(struct child_process *cmd,
 		strvec_push(&cmd->args, "--bitmap");
 }
 
+static int repack_fill_midx_stdin_packs(struct child_process *cmd,
+					struct string_list *include)
+{
+	struct string_list_item *item;
+	FILE *in;
+	int ret;
+
+	cmd->in = -1;
+
+	strvec_push(&cmd->args, "--stdin-packs");
+
+	ret = start_command(cmd);
+	if (ret)
+		return ret;
+
+	in = xfdopen(cmd->in, "w");
+	for_each_string_list_item(item, include)
+		fprintf(in, "%s\n", item->string);
+	fclose(in);
+
+	return finish_command(cmd);
+}
+
 int write_midx_included_packs(struct repack_write_midx_opts *opts)
 {
 	struct child_process cmd = CHILD_PROCESS_INIT;
 	struct string_list include = STRING_LIST_INIT_DUP;
 	struct string_list_item *item;
 	struct packed_git *preferred = pack_geometry_preferred_pack(opts->geometry);
-	FILE *in;
 	int ret = 0;
 
 	midx_included_packs(&include, opts);
 	if (!include.nr)
 		goto done;
 
-	cmd.in = -1;
-
 	repack_prepare_midx_command(&cmd, opts, "write");
-	strvec_push(&cmd.args, "--stdin-packs");
 
 	if (preferred)
 		strvec_pushf(&cmd.args, "--preferred-pack=%s",
@@ -358,16 +377,7 @@ int write_midx_included_packs(struct repack_write_midx_opts *opts)
 		strvec_pushf(&cmd.args, "--refs-snapshot=%s",
 			     opts->refs_snapshot);
 
-	ret = start_command(&cmd);
-	if (ret)
-		goto done;
-
-	in = xfdopen(cmd.in, "w");
-	for_each_string_list_item(item, &include)
-		fprintf(in, "%s\n", item->string);
-	fclose(in);
-
-	ret = finish_command(&cmd);
+	ret = repack_fill_midx_stdin_packs(&cmd, &include);
 done:
 	if (!ret && opts->write_bitmaps)
 		remove_redundant_bitmaps(&include, opts->packdir);
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 10/14] repack-geometry: prepare for incremental MIDX repacking

[Taylor Blau]

Teach `pack_geometry_init()` to optionally restrict the set of
repacking candidates to only packs in the tip MIDX layer when a
`midx_layer_threshold` is configured. If the tip layer has fewer packs
than the threshold, those packs are excluded entirely; otherwise only
packs in that layer participate in the geometric repack.

Also track whether any tip-layer packs were included in the rollup
(`midx_tip_rewritten`), which a subsequent commit will use to decide
how to update the MIDX chain after repacking.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 repack-geometry.c | 37 +++++++++++++++++++++++++++++++++++++
 repack.h          |  4 ++++
 2 files changed, 41 insertions(+)

diff --git a/repack-geometry.c b/repack-geometry.c
index 7cebd0cb45f..d2065205f87 100644
--- a/repack-geometry.c
+++ b/repack-geometry.c
@@ -4,6 +4,7 @@
 #include "repack.h"
 #include "repository.h"
 #include "hex.h"
+#include "midx.h"
 #include "packfile.h"
 
 static uint32_t pack_geometry_weight(struct packed_git *p)
@@ -31,8 +32,30 @@ void pack_geometry_init(struct pack_geometry *geometry,
 {
 	struct packed_git *p;
 	struct strbuf buf = STRBUF_INIT;
+	struct multi_pack_index *m = get_multi_pack_index(existing->source);
 
 	repo_for_each_pack(existing->repo, p) {
+		if (geometry->midx_layer_threshold_set && m &&
+		    p->multi_pack_index) {
+			/*
+			 * When writing MIDX layers incrementally,
+			 * ignore packs unless they are in the most
+			 * recent MIDX layer *and* there are at least
+			 * 'midx_layer_threshold' packs in that layer.
+			 *
+			 * Otherwise 'p' is either in an older layer, or
+			 * the youngest layer does not have enough packs
+			 * to consider its packs as candidates for
+			 * repacking. In either of those cases we want
+			 * to ignore the pack.
+			 */
+			if (m->num_packs > geometry->midx_layer_threshold &&
+			    midx_layer_contains_pack(m, pack_basename(p)))
+				;
+			else
+				continue;
+		}
+
 		if (args->local && !p->pack_local)
 			/*
 			 * When asked to only repack local packfiles we skip
@@ -173,6 +196,20 @@ void pack_geometry_split(struct pack_geometry *geometry)
 	geometry->promisor_split = compute_pack_geometry_split(geometry->promisor_pack,
 							       geometry->promisor_pack_nr,
 							       geometry->split_factor);
+	for (uint32_t i = 0; i < geometry->split; i++) {
+		struct packed_git *p = geometry->pack[i];
+		/*
+		 * During incremental MIDX/bitmap repacking, any packs
+		 * included in the rollup are either (a) not MIDX'd, or
+		 * (b) contained in the tip layer iff it has more than
+		 * the threshold number of packs.
+		 *
+		 * In the latter case, we can safely conclude that the
+		 * tip of the MIDX chain will be rewritten.
+		 */
+		if (p->multi_pack_index)
+			geometry->midx_tip_rewritten = true;
+	}
 }
 
 struct packed_git *pack_geometry_preferred_pack(struct pack_geometry *geometry)
diff --git a/repack.h b/repack.h
index c0e9f0ca647..77d24ee45fb 100644
--- a/repack.h
+++ b/repack.h
@@ -108,6 +108,10 @@ struct pack_geometry {
 	uint32_t promisor_pack_nr, promisor_pack_alloc;
 	uint32_t promisor_split;
 
+	uint32_t midx_layer_threshold;
+	bool midx_layer_threshold_set;
+	bool midx_tip_rewritten;
+
 	int split_factor;
 };
 
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 11/14] builtin/repack.c: convert `--write-midx` to an `OPT_CALLBACK`

[Taylor Blau]

Change the --write-midx (-m) flag from an OPT_BOOL to an OPT_CALLBACK
that accepts an optional mode argument. Introduce an enum with
REPACK_WRITE_MIDX_NONE and REPACK_WRITE_MIDX_DEFAULT to distinguish
between the two states, and update all existing boolean checks
accordingly.

For now, passing no argument (or just `-m`) selects the default mode,
preserving existing behavior. A subsequent commit will add a new mode
for writing incremental MIDXs.

Extract repack_write_midx() as a dispatcher that selects the
appropriate MIDX-writing implementation based on the mode.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 builtin/repack.c | 50 ++++++++++++++++++++++++++++++++++++------------
 repack-midx.c    | 14 +++++++++++++-
 repack.h         |  8 +++++++-
 3 files changed, 58 insertions(+), 14 deletions(-)

diff --git a/builtin/repack.c b/builtin/repack.c
index 44a95b56f23..3a5042491d6 100644
--- a/builtin/repack.c
+++ b/builtin/repack.c
@@ -97,6 +97,24 @@ static int repack_config(const char *var, const char *value,
 	return git_default_config(var, value, ctx, cb);
 }
 
+static int option_parse_write_midx(const struct option *opt, const char *arg,
+				   int unset)
+{
+	enum repack_write_midx_mode *cfg = opt->value;
+
+	if (unset) {
+		*cfg = REPACK_WRITE_MIDX_NONE;
+		return 0;
+	}
+
+	if (!arg || !*arg)
+		*cfg = REPACK_WRITE_MIDX_DEFAULT;
+	else
+		return error(_("unknown value for %s: %s"), opt->long_name, arg);
+
+	return 0;
+}
+
 int cmd_repack(int argc,
 	       const char **argv,
 	       const char *prefix,
@@ -119,7 +137,7 @@ int cmd_repack(int argc,
 	struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
 	struct pack_objects_args po_args = PACK_OBJECTS_ARGS_INIT;
 	struct pack_objects_args cruft_po_args = PACK_OBJECTS_ARGS_INIT;
-	int write_midx = 0;
+	enum repack_write_midx_mode write_midx = REPACK_WRITE_MIDX_NONE;
 	const char *cruft_expiration = NULL;
 	const char *expire_to = NULL;
 	const char *filter_to = NULL;
@@ -185,8 +203,14 @@ int cmd_repack(int argc,
 				N_("do not repack this pack")),
 		OPT_INTEGER('g', "geometric", &geometry.split_factor,
 			    N_("find a geometric progression with factor <N>")),
-		OPT_BOOL('m', "write-midx", &write_midx,
-			   N_("write a multi-pack index of the resulting packs")),
+		OPT_CALLBACK_F(0, "write-midx", &write_midx,
+			   N_("mode"),
+			   N_("write a multi-pack index of the resulting packs"),
+			   PARSE_OPT_OPTARG, option_parse_write_midx),
+		OPT_SET_INT_F('m', NULL, &write_midx,
+			   N_("write a multi-pack index of the resulting packs"),
+			   REPACK_WRITE_MIDX_DEFAULT,
+			   PARSE_OPT_HIDDEN),
 		OPT_STRING(0, "expire-to", &expire_to, N_("dir"),
 			   N_("pack prefix to store a pack containing pruned objects")),
 		OPT_STRING(0, "filter-to", &filter_to, N_("dir"),
@@ -221,14 +245,16 @@ int cmd_repack(int argc,
 		pack_everything |= ALL_INTO_ONE;
 
 	if (write_bitmaps < 0) {
-		if (!write_midx &&
+		if (write_midx == REPACK_WRITE_MIDX_NONE &&
 		    (!(pack_everything & ALL_INTO_ONE) || !is_bare_repository()))
 			write_bitmaps = 0;
 	}
 	if (po_args.pack_kept_objects < 0)
-		po_args.pack_kept_objects = write_bitmaps > 0 && !write_midx;
+		po_args.pack_kept_objects = write_bitmaps > 0 &&
+			write_midx == REPACK_WRITE_MIDX_NONE;
 
-	if (write_bitmaps && !(pack_everything & ALL_INTO_ONE) && !write_midx)
+	if (write_bitmaps && !(pack_everything & ALL_INTO_ONE) &&
+	    write_midx == REPACK_WRITE_MIDX_NONE)
 		die(_(incremental_bitmap_conflict_error));
 
 	if (write_bitmaps && po_args.local &&
@@ -244,7 +270,7 @@ int cmd_repack(int argc,
 		write_bitmaps = 0;
 	}
 
-	if (write_midx && write_bitmaps) {
+	if (write_midx != REPACK_WRITE_MIDX_NONE && write_bitmaps) {
 		struct strbuf path = STRBUF_INIT;
 
 		strbuf_addf(&path, "%s/%s_XXXXXX",
@@ -297,7 +323,7 @@ int cmd_repack(int argc,
 	}
 	if (repo_has_promisor_remote(repo))
 		strvec_push(&cmd.args, "--exclude-promisor-objects");
-	if (!write_midx) {
+	if (write_midx == REPACK_WRITE_MIDX_NONE) {
 		if (write_bitmaps > 0)
 			strvec_push(&cmd.args, "--write-bitmap-index");
 		else if (write_bitmaps < 0)
@@ -504,7 +530,7 @@ int cmd_repack(int argc,
 	if (delete_redundant && pack_everything & ALL_INTO_ONE)
 		existing_packs_mark_for_deletion(&existing, &names);
 
-	if (write_midx) {
+	if (write_midx != REPACK_WRITE_MIDX_NONE) {
 		struct repack_write_midx_opts opts = {
 			.existing = &existing,
 			.geometry = &geometry,
@@ -513,11 +539,11 @@ int cmd_repack(int argc,
 			.packdir = packdir,
 			.show_progress = show_progress,
 			.write_bitmaps = write_bitmaps > 0,
-			.midx_must_contain_cruft = midx_must_contain_cruft
+			.midx_must_contain_cruft = midx_must_contain_cruft,
+			.mode = write_midx,
 		};
 
-		ret = write_midx_included_packs(&opts);
-
+		ret = repack_write_midx(&opts);
 		if (ret)
 			goto cleanup;
 	}
diff --git a/repack-midx.c b/repack-midx.c
index bc5059927f7..6c28d9acef6 100644
--- a/repack-midx.c
+++ b/repack-midx.c
@@ -323,7 +323,7 @@ static int repack_fill_midx_stdin_packs(struct child_process *cmd,
 	return finish_command(cmd);
 }
 
-int write_midx_included_packs(struct repack_write_midx_opts *opts)
+static int write_midx_included_packs(struct repack_write_midx_opts *opts)
 {
 	struct child_process cmd = CHILD_PROCESS_INIT;
 	struct string_list include = STRING_LIST_INIT_DUP;
@@ -386,3 +386,15 @@ int write_midx_included_packs(struct repack_write_midx_opts *opts)
 
 	return ret;
 }
+
+int repack_write_midx(struct repack_write_midx_opts *opts)
+{
+	switch (opts->mode) {
+	case REPACK_WRITE_MIDX_NONE:
+		BUG("write_midx mode is NONE?");
+	case REPACK_WRITE_MIDX_DEFAULT:
+		return write_midx_included_packs(opts);
+	default:
+		BUG("unhandled write_midx mode: %d", opts->mode);
+	}
+}
diff --git a/repack.h b/repack.h
index 77d24ee45fb..81907fcce7f 100644
--- a/repack.h
+++ b/repack.h
@@ -134,6 +134,11 @@ void pack_geometry_release(struct pack_geometry *geometry);
 
 struct tempfile;
 
+enum repack_write_midx_mode {
+	REPACK_WRITE_MIDX_NONE,
+	REPACK_WRITE_MIDX_DEFAULT,
+};
+
 struct repack_write_midx_opts {
 	struct existing_packs *existing;
 	struct pack_geometry *geometry;
@@ -143,10 +148,11 @@ struct repack_write_midx_opts {
 	int show_progress;
 	int write_bitmaps;
 	int midx_must_contain_cruft;
+	enum repack_write_midx_mode mode;
 };
 
 void midx_snapshot_refs(struct repository *repo, struct tempfile *f);
-int write_midx_included_packs(struct repack_write_midx_opts *opts);
+int repack_write_midx(struct repack_write_midx_opts *opts);
 
 int write_filtered_pack(const struct write_pack_opts *opts,
 			struct existing_packs *existing,
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 12/14] repack: implement incremental MIDX repacking

[Taylor Blau]

Implement the `write_midx_incremental()` function, which builds and
maintains an incremental MIDX chain as part of the geometric repacking
process.

Unlike the default mode which writes a single flat MIDX, the incremental
mode constructs a compaction plan that determines which MIDX layers to
write, compact, or copy, and then executes each step using `git
multi-pack-index` subcommands with the --checksum-only flag.

The repacking strategy works as follows:

 * Acquire the lock guarding the multi-pack-index-chain.

 * A new MIDX layer is always written containing the newly created
   pack(s). If the tip MIDX layer was rewritten during geometric
   repacking, any surviving packs from that layer are also included.

 * Starting from the new layer, adjacent MIDX layers are merged together
   as long as the accumulated object count exceeds half the object count
   of the next deeper layer (controlled by 'repack.midxSplitFactor').

 * Remaining layers in the chain are evaluated pairwise and either
   compacted or copied as-is, following the same merging condition.

 * Write the contents of the new multi-pack-index chain, atomically move
   it into place, and then release the lock.

 * Delete any now-unused MIDX layers.

After writing the new layer, the strategy is evaluated among the
existing MIDX layers in order from oldest to newest. Each step that
writes a new MIDX layer uses "--checksum-only" to avoid updating the
multi-pack-index-chain file. After all steps are complete, the new chain
file is written and then atomically moved into place.

At present, this functionality is exposed behind a new enum value,
`REPACK_WRITE_MIDX_INCREMENTAL`, but has no external callers. A
subsequent commit will expose this mode via `git repack
--write-midx=incremental`.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 repack-midx.c | 475 +++++++++++++++++++++++++++++++++++++++++++++++++-
 repack.h      |   3 +
 2 files changed, 476 insertions(+), 2 deletions(-)

diff --git a/repack-midx.c b/repack-midx.c
index 6c28d9acef6..ad78a10378b 100644
--- a/repack-midx.c
+++ b/repack-midx.c
@@ -2,9 +2,12 @@
 #include "repack.h"
 #include "hash.h"
 #include "hex.h"
+#include "lockfile.h"
+#include "midx.h"
 #include "odb.h"
 #include "oidset.h"
 #include "pack-bitmap.h"
+#include "path.h"
 #include "refs.h"
 #include "run-command.h"
 #include "tempfile.h"
@@ -301,13 +304,16 @@ static void repack_prepare_midx_command(struct child_process *cmd,
 }
 
 static int repack_fill_midx_stdin_packs(struct child_process *cmd,
-					struct string_list *include)
+					struct string_list *include,
+					struct string_list *out)
 {
 	struct string_list_item *item;
 	FILE *in;
 	int ret;
 
 	cmd->in = -1;
+	if (out)
+		cmd->out = -1;
 
 	strvec_push(&cmd->args, "--stdin-packs");
 
@@ -320,6 +326,17 @@ static int repack_fill_midx_stdin_packs(struct child_process *cmd,
 		fprintf(in, "%s\n", item->string);
 	fclose(in);
 
+	if (out) {
+		struct strbuf buf = STRBUF_INIT;
+		FILE *outf = xfdopen(cmd->out, "r");
+
+		while (strbuf_getline(&buf, outf) != EOF)
+			string_list_append(out, buf.buf);
+		strbuf_release(&buf);
+
+		fclose(outf);
+	}
+
 	return finish_command(cmd);
 }
 
@@ -377,7 +394,7 @@ static int write_midx_included_packs(struct repack_write_midx_opts *opts)
 		strvec_pushf(&cmd.args, "--refs-snapshot=%s",
 			     opts->refs_snapshot);
 
-	ret = repack_fill_midx_stdin_packs(&cmd, &include);
+	ret = repack_fill_midx_stdin_packs(&cmd, &include, NULL);
 done:
 	if (!ret && opts->write_bitmaps)
 		remove_redundant_bitmaps(&include, opts->packdir);
@@ -387,6 +404,458 @@ static int write_midx_included_packs(struct repack_write_midx_opts *opts)
 	return ret;
 }
 
+struct midx_compaction_step {
+	union {
+		struct multi_pack_index *copy;
+		struct string_list write;
+		struct {
+			struct multi_pack_index *from;
+			struct multi_pack_index *to;
+		} compact;
+	} u;
+
+	uint32_t objects_nr;
+	const char *csum;
+
+	enum {
+		MIDX_COMPACTION_STEP_UNKNOWN,
+		MIDX_COMPACTION_STEP_COPY,
+		MIDX_COMPACTION_STEP_WRITE,
+		MIDX_COMPACTION_STEP_COMPACT,
+	} type;
+};
+
+static const char *midx_compaction_step_base(const struct midx_compaction_step *step)
+{
+	switch (step->type) {
+	case MIDX_COMPACTION_STEP_UNKNOWN:
+		BUG("cannot use UNKNOWN step as a base");
+	case MIDX_COMPACTION_STEP_COPY:
+		return midx_get_checksum_hex(step->u.copy);
+	case MIDX_COMPACTION_STEP_WRITE:
+		BUG("cannot use WRITE step as a base");
+	case MIDX_COMPACTION_STEP_COMPACT:
+		return midx_get_checksum_hex(step->u.compact.to);
+	default:
+		BUG("unhandled midx compaction step type %d", step->type);
+	}
+}
+
+static int midx_compaction_step_exec_copy(struct midx_compaction_step *step)
+{
+	step->csum = xstrdup(midx_get_checksum_hex(step->u.copy));
+	return 0;
+}
+
+static int midx_compaction_step_exec_write(struct midx_compaction_step *step,
+					   struct repack_write_midx_opts *opts,
+					   const char *base)
+{
+	struct child_process cmd = CHILD_PROCESS_INIT;
+	struct string_list hash = STRING_LIST_INIT_DUP;
+	struct string_list_item *item;
+	const char *preferred_pack = NULL;
+	int ret = 0;
+
+	if (!step->u.write.nr) {
+		ret = error(_("no packs to write MIDX during compaction"));
+		goto out;
+	}
+
+	for_each_string_list_item(item, &step->u.write) {
+		if (item->util)
+			preferred_pack = item->string;
+	}
+
+	repack_prepare_midx_command(&cmd, opts, "write");
+	strvec_pushl(&cmd.args, "--incremental", "--checksum-only", NULL);
+	strvec_pushf(&cmd.args, "--base=%s", base ? base : "none");
+
+	if (preferred_pack) {
+		struct strbuf buf = STRBUF_INIT;
+
+		strbuf_addstr(&buf, preferred_pack);
+		strbuf_strip_suffix(&buf, ".idx");
+		strbuf_addstr(&buf, ".pack");
+
+		strvec_pushf(&cmd.args, "--preferred-pack=%s", buf.buf);
+
+		strbuf_release(&buf);
+	}
+
+	ret = repack_fill_midx_stdin_packs(&cmd, &step->u.write, &hash);
+	if (hash.nr != 1) {
+		ret = error(_("expected exactly one line during MIDX write, "
+			      "got: %"PRIuMAX),
+			    (uintmax_t)hash.nr);
+		goto out;
+	}
+
+	step->csum = xstrdup(hash.items[0].string);
+
+out:
+	string_list_clear(&hash, 0);
+
+	return ret;
+}
+
+static int midx_compaction_step_exec_compact(struct midx_compaction_step *step,
+					     struct repack_write_midx_opts *opts)
+{
+	struct child_process cmd = CHILD_PROCESS_INIT;
+	struct strbuf buf = STRBUF_INIT;
+	FILE *out = NULL;
+	int ret;
+
+	repack_prepare_midx_command(&cmd, opts, "compact");
+	strvec_pushl(&cmd.args, "--incremental", "--checksum-only",
+		     midx_get_checksum_hex(step->u.compact.from),
+		     midx_get_checksum_hex(step->u.compact.to), NULL);
+
+	cmd.out = -1;
+
+	ret = start_command(&cmd);
+	if (ret)
+		goto out;
+
+	out = xfdopen(cmd.out, "r");
+	while (strbuf_getline_lf(&buf, out) != EOF) {
+		if (step->csum) {
+			ret = error(_("unexpected MIDX output: '%s'"), buf.buf);
+			goto out;
+		}
+		step->csum = strbuf_detach(&buf, NULL);
+	}
+
+	ret = finish_command(&cmd);
+
+out:
+	if (out)
+		fclose(out);
+	strbuf_release(&buf);
+
+	return ret;
+}
+
+static int midx_compaction_step_exec(struct midx_compaction_step *step,
+				     struct repack_write_midx_opts *opts,
+				     const char *base)
+{
+	switch (step->type) {
+	case MIDX_COMPACTION_STEP_UNKNOWN:
+		BUG("cannot execute UNKNOWN midx compaction step");
+	case MIDX_COMPACTION_STEP_COPY:
+		return midx_compaction_step_exec_copy(step);
+	case MIDX_COMPACTION_STEP_WRITE:
+		return midx_compaction_step_exec_write(step, opts, base);
+	case MIDX_COMPACTION_STEP_COMPACT:
+		return midx_compaction_step_exec_compact(step, opts);
+	default:
+		BUG("unhandled midx compaction step type %d", step->type);
+	}
+}
+
+static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
+					    struct midx_compaction_step **steps_p,
+					    size_t *steps_nr_p)
+{
+	struct multi_pack_index *m;
+	struct midx_compaction_step *steps = NULL;
+	struct midx_compaction_step step = { 0 };
+	struct strbuf buf = STRBUF_INIT;
+	size_t steps_nr = 0, steps_alloc = 0;
+	uint32_t i;
+	int ret = 0;
+
+	odb_reprepare(opts->existing->repo->objects);
+	m = get_multi_pack_index(opts->existing->source);
+
+	for (i = 0; m && i < m->num_packs + m->num_packs_in_base; i++) {
+		if (prepare_midx_pack(m, i)) {
+			ret = error(_("could not load pack %"PRIu32" from MIDX"),
+				    i);
+			goto out;
+		}
+	}
+
+	/*
+	 * The first MIDX in the resulting chain is always going to be
+	 * new.
+	 *
+	 * At a minimum, it will include all of the newly written packs.
+	 * If there is an existing MIDX whose tip layer contain packs
+	 * that were repacked, it will also include any of its pack
+	 * which were *not* rolled up as part of the geometric repack
+	 * (if any), and the previous tip will be replaced.
+	 *
+	 * It may grow to include the packs from zero or more MIDXs from
+	 * the old chain, beginning either at the old tip (if the MIDX
+	 * was *not* rewritten) or the old tip's base MIDX layer
+	 * (otherwise).
+	 */
+
+	step.type = MIDX_COMPACTION_STEP_WRITE;
+	string_list_init_nodup(&step.u.write);
+
+	for (i = 0; i < opts->names->nr; i++) {
+		strbuf_addf(&buf, "pack-%s.idx", opts->names->items[i].string);
+		string_list_append(&step.u.write, strbuf_detach(&buf, NULL));
+	}
+	for (i = 0; i < opts->geometry->split; i++) {
+		struct packed_git *p = opts->geometry->pack[i];
+		if (unsigned_add_overflows(step.objects_nr, p->num_objects)) {
+			ret = error(_("too many objects in MIDX compaction step"));
+			goto out;
+		}
+
+		step.objects_nr += p->num_objects;
+	}
+
+	/*
+	 * Now handle any existing packs which were *not* rewritten.
+	 *
+	 * The list of packs in opts->geometry only contains MIDX'd
+	 * packs from the newest layer when that layer has more than
+	 * 'repack.midxNewLayerThreshold' number of packs.
+	 *
+	 * If the MIDX tip was rewritten (that is, one or more of those
+	 * packs appear below the split line), then add all packs above
+	 * the split line to the new layer, as the old one is no longer
+	 * usable.
+	 *
+	 * If the MIDX tip was not rewritten (that is, all MIDX'd packs
+	 * from the youngest layer appear below the split line, or were
+	 * not included in the geometric repack at all because there
+	 * were too few of them), ignore them since we'll retain the
+	 * existing layer as-is.
+	 */
+	for (i = opts->geometry->split; i < opts->geometry->pack_nr; i++) {
+		struct packed_git *p = opts->geometry->pack[i];
+		struct string_list_item *item;
+
+		strbuf_reset(&buf);
+		strbuf_addstr(&buf, pack_basename(p));
+		strbuf_strip_suffix(&buf, ".pack");
+		strbuf_addstr(&buf, ".idx");
+
+		if (p->multi_pack_index && !opts->geometry->midx_tip_rewritten)
+			continue;
+
+		item = string_list_append(&step.u.write,
+					  strbuf_detach(&buf, NULL));
+		if (p->multi_pack_index || i == opts->geometry->pack_nr - 1)
+			item->util = (void *)1; /* mark as preferred */
+
+		if (unsigned_add_overflows(step.objects_nr, p->num_objects)) {
+			ret = error(_("too many objects in MIDX compaction step"));
+			goto out;
+		}
+		step.objects_nr += p->num_objects;
+	}
+
+	/*
+	 * If the MIDX tip was rewritten, then we no longer consider it
+	 * a candidate for compaction, since it will not exist in the
+	 * MIDX chain being built.
+	 */
+	if (opts->geometry->midx_tip_rewritten)
+		m = m->base_midx;
+
+	/*
+	 * Compact additional MIDX layers into this proposed one until
+	 * the merging condition is violated.
+	 */
+	while (m) {
+		uint32_t preferred_pack_idx;
+
+		if (step.objects_nr < m->num_objects / opts->midx_split_factor) {
+			/*
+			 * Stop compacting MIDX layer as soon as the
+			 * merged size is less than half the size of the
+			 * next layer in the chain.
+			 */
+			break;
+		}
+
+		if (midx_preferred_pack(m, &preferred_pack_idx) < 0) {
+			ret = error(_("could not find preferred pack for MIDX "
+				      "%s"), midx_get_checksum_hex(m));
+			goto out;
+		}
+
+		for (i = 0; i < m->num_packs; i++) {
+			struct string_list_item *item;
+			uint32_t pack_int_id = i + m->num_packs_in_base;
+			struct packed_git *p = nth_midxed_pack(m, pack_int_id);
+
+			strbuf_reset(&buf);
+			strbuf_addstr(&buf, pack_basename(p));
+			strbuf_strip_suffix(&buf, ".pack");
+			strbuf_addstr(&buf, ".idx");
+
+			item = string_list_append(&step.u.write,
+						  strbuf_detach(&buf, NULL));
+			if (pack_int_id == preferred_pack_idx)
+				item->util = (void *)1; /* mark as preferred */
+		}
+
+		if (unsigned_add_overflows(step.objects_nr, m->num_objects)) {
+			ret = error(_("too many objects in MIDX compaction step"));
+			goto out;
+		}
+		step.objects_nr += m->num_objects;
+
+		m = m->base_midx;
+	}
+
+	if (step.u.write.nr > 0) {
+		/*
+		 * As long as there is at least one new pack to write
+		 * (and thus the MIDX is non-empty), add it to the plan.
+		 */
+		ALLOC_GROW(steps, steps_nr + 1, steps_alloc);
+		steps[steps_nr++] = step;
+	}
+
+	/*
+	 * Then start over, repeat, and either compact or keep as-is
+	 * each MIDX layer until we have exhausted the chain.
+	 *
+	 * Finally, evaluate the remainder of the chain (if any) and
+	 * either compact a sequence of adjacent layers, or keep
+	 * individual layers as-is according to the same merging
+	 * condition as above.
+	 */
+	while (m) {
+		struct multi_pack_index *next = m;
+
+		ALLOC_GROW(steps, steps_nr + 1, steps_alloc);
+
+		memset(&step, 0, sizeof(step));
+		step.type = MIDX_COMPACTION_STEP_UNKNOWN;
+
+		while (next) {
+			struct multi_pack_index *base = next->base_midx;
+			uint32_t proposed_objects_nr;
+
+			if (unsigned_add_overflows(step.objects_nr, next->num_objects)) {
+				ret = error(_("too many objects in MIDX compaction step"));
+				goto out;
+			}
+
+			proposed_objects_nr = step.objects_nr + next->num_objects;
+
+			if (!base) {
+				/*
+				 * If we are at the end of the MIDX
+				 * chain, there is nothing to compact,
+				 * so mark it and stop.
+				 */
+				step.objects_nr = proposed_objects_nr;
+				break;
+			}
+
+			if (proposed_objects_nr < base->num_objects / opts->midx_split_factor) {
+				/*
+				 * If there is a MIDX following this
+				 * one, but our accumulated size is less
+				 * than half of its size, compacting
+				 * them would violate the merging
+				 * condition, so stop here.
+				 */
+				break;
+			}
+
+			/*
+			 * Otherwise, it is OK to compact the next layer
+			 * into this one. Do so, and then continue
+			 * through the remainder of the chain.
+			 */
+			step.objects_nr = proposed_objects_nr;
+			next = base;
+		}
+
+		if (m == next) {
+			step.type = MIDX_COMPACTION_STEP_COPY;
+			step.u.copy = m;
+		} else {
+			step.type = MIDX_COMPACTION_STEP_COMPACT;
+			step.u.compact.from = next;
+			step.u.compact.to = m;
+		}
+
+		m = next->base_midx;
+
+		steps[steps_nr++] = step;
+	}
+
+out:
+	*steps_p = steps;
+	*steps_nr_p = steps_nr;
+
+	return ret;
+}
+
+static int write_midx_incremental(struct repack_write_midx_opts *opts)
+{
+	struct midx_compaction_step *steps = NULL;
+	struct strbuf lock_name = STRBUF_INIT;
+	struct lock_file lf;
+	size_t steps_nr = 0;
+	size_t i;
+	int ret = 0;
+
+	get_midx_chain_filename(opts->existing->source, &lock_name);
+	if (safe_create_leading_directories(opts->existing->repo,
+					    lock_name.buf))
+		die_errno(_("unable to create leading directories of %s"),
+			  lock_name.buf);
+	hold_lock_file_for_update(&lf, lock_name.buf, LOCK_DIE_ON_ERROR);
+
+	if (!fdopen_lock_file(&lf, "w")) {
+		ret = error_errno(_("unable to open multi-pack-index chain file"));
+		goto done;
+	}
+
+	if (repack_make_midx_compaction_plan(opts, &steps, &steps_nr) < 0) {
+		ret = error(_("unable to generate compaction plan"));
+		goto done;
+	}
+
+	for (i = 0; i < steps_nr; i++) {
+		struct midx_compaction_step *step = &steps[i];
+		char *base = NULL;
+
+		if (i + 1 < steps_nr)
+			base = xstrdup(midx_compaction_step_base(&steps[i + 1]));
+
+		if (midx_compaction_step_exec(step, opts, base) < 0) {
+			ret = error(_("unable to execute compaction step %"PRIuMAX),
+				    (uintmax_t)i);
+			free(base);
+			goto done;
+		}
+
+		free(base);
+	}
+
+	i = steps_nr;
+	while (i--) {
+		struct midx_compaction_step *step = &steps[i];
+		if (!step->csum)
+			BUG("missing result for compaction step %"PRIuMAX,
+			    (uintmax_t)i);
+		fprintf(get_lock_file_fp(&lf), "%s\n", step->csum);
+	}
+
+	commit_lock_file(&lf);
+
+done:
+	strbuf_release(&lock_name);
+	free(steps);
+	return ret;
+}
+
 int repack_write_midx(struct repack_write_midx_opts *opts)
 {
 	switch (opts->mode) {
@@ -394,6 +863,8 @@ int repack_write_midx(struct repack_write_midx_opts *opts)
 		BUG("write_midx mode is NONE?");
 	case REPACK_WRITE_MIDX_DEFAULT:
 		return write_midx_included_packs(opts);
+	case REPACK_WRITE_MIDX_INCREMENTAL:
+		return write_midx_incremental(opts);
 	default:
 		BUG("unhandled write_midx mode: %d", opts->mode);
 	}
diff --git a/repack.h b/repack.h
index 81907fcce7f..831ccfb1c6c 100644
--- a/repack.h
+++ b/repack.h
@@ -137,6 +137,7 @@ struct tempfile;
 enum repack_write_midx_mode {
 	REPACK_WRITE_MIDX_NONE,
 	REPACK_WRITE_MIDX_DEFAULT,
+	REPACK_WRITE_MIDX_INCREMENTAL,
 };
 
 struct repack_write_midx_opts {
@@ -148,6 +149,8 @@ struct repack_write_midx_opts {
 	int show_progress;
 	int write_bitmaps;
 	int midx_must_contain_cruft;
+	int midx_split_factor;
+	int midx_new_layer_threshold;
 	enum repack_write_midx_mode mode;
 };
 
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 13/14] repack: introduce `--write-midx=incremental`

[Taylor Blau]

Expose the incremental MIDX repacking mode (implemented in the previous
commit) via a new --write-midx=incremental option for `git repack`.

Add "incremental" as a recognized argument to the --write-midx
OPT_CALLBACK, mapping it to REPACK_WRITE_MIDX_INCREMENTAL. When this
mode is active and --geometric is in use, set the midx_layer_threshold
on the pack geometry so that only packs in sufficiently large tip layers
are considered for repacking.

Two new configuration options control the compaction behavior:

 - repack.midxSplitFactor (default: 2): the factor used in the
   geometric merging condition for MIDX layers.

 - repack.midxNewLayerThreshold (default: 8): the minimum number of
   packs in the tip MIDX layer before its packs are considered as
   candidates for geometric repacking.

Add tests exercising the new mode across a variety of scenarios
including basic geometric violations, multi-round chain integrity,
branching and merging histories, cross-layer object uniqueness, and
threshold-based compaction.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 Documentation/config/repack.adoc   |  18 ++
 Documentation/git-repack.adoc      |  39 ++-
 builtin/repack.c                   |  38 ++-
 midx.c                             |  30 ++
 midx.h                             |   3 +
 repack-geometry.c                  |  13 +-
 repack-midx.c                      | 113 +++++++-
 repack.c                           |  21 +-
 repack.h                           |   9 +-
 t/meson.build                      |   1 +
 t/t7705-repack-incremental-midx.sh | 436 +++++++++++++++++++++++++++++
 11 files changed, 692 insertions(+), 29 deletions(-)
 create mode 100755 t/t7705-repack-incremental-midx.sh

diff --git a/Documentation/config/repack.adoc b/Documentation/config/repack.adoc
index e9e78dcb198..054de9f8795 100644
--- a/Documentation/config/repack.adoc
+++ b/Documentation/config/repack.adoc
@@ -46,3 +46,21 @@ repack.midxMustContainCruft::
 	`--write-midx`. When false, cruft packs are only included in the MIDX
 	when necessary (e.g., because they might be required to form a
 	reachability closure with MIDX bitmaps). Defaults to true.
+
+repack.midxSplitFactor::
+	The factor used in the geometric merging condition when
+	compacting incremental MIDX layers during `git repack` when
+	invoked with the `--write-midx=incremental` option.
++
+Adjacent layers are merged when the accumulated object count of the
+newer layer exceeds `1/<N>` of the object count of the next deeper
+layer. Defaults to 2.
+
+repack.midxNewLayerThreshold::
+	The minimum number of packs in the tip MIDX layer before those
+	packs are considered as candidates for geometric repacking
+	during `git repack --write-midx=incremental`.
++
+When the tip layer has fewer packs than this threshold, those packs are
+excluded from the geometric repack entirely, and are thus left
+unmodified. Defaults to 8.
diff --git a/Documentation/git-repack.adoc b/Documentation/git-repack.adoc
index 673ce910837..27a99cc46f4 100644
--- a/Documentation/git-repack.adoc
+++ b/Documentation/git-repack.adoc
@@ -11,7 +11,7 @@ SYNOPSIS
 [verse]
 'git repack' [-a] [-A] [-d] [-f] [-F] [-l] [-n] [-q] [-b] [-m]
 	[--window=<n>] [--depth=<n>] [--threads=<n>] [--keep-pack=<pack-name>]
-	[--write-midx] [--name-hash-version=<n>] [--path-walk]
+	[--write-midx[=<mode>]] [--name-hash-version=<n>] [--path-walk]
 
 DESCRIPTION
 -----------
@@ -250,9 +250,42 @@ pack as the preferred pack for object selection by the MIDX (see
 linkgit:git-multi-pack-index[1]).
 
 -m::
---write-midx::
+--write-midx[=<mode>]::
 	Write a multi-pack index (see linkgit:git-multi-pack-index[1])
-	containing the non-redundant packs.
+	containing the non-redundant packs. The following modes are
+	available:
++
+--
+	`default`;;
+		Write a single MIDX covering all packs. This is the
+		default when `--write-midx` is given without an
+		explicit mode.
+
+	`incremental`;;
+		Write an incremental MIDX chain instead of a single
+		flat MIDX. This mode requires `--geometric`.
++
+The incremental mode maintains a chain of MIDX layers that is compacted
+over time using a geometric merging strategy. Each repack creates a new
+tip layer containing the newly written pack(s). Adjacent layers are then
+merged whenever the newer layer's object count exceeds
+`1/repack.midxSplitFactor` of the next deeper layer's count. Layers
+that do not meet this condition are retained as-is.
++
+The result is that newer (tip) layers tend to contain many small packs
+with relatively few objects, while older (deeper) layers contain fewer,
+larger packs covering more objects. Because compaction is driven by the
+tip of the chain, newer layers are also rewritten more frequently than
+older ones, which are only touched when enough objects have accumulated
+to justify merging into them. This keeps the total number of layers
+logarithmic relative to the total number of objects.
++
+Only packs in the tip MIDX layer are considered as candidates for the
+geometric repack; packs in deeper layers are left untouched. If the tip
+layer contains fewer packs than `repack.midxNewLayerThreshold`, those
+packs are excluded from the geometry entirely, and a new layer is
+created for any new pack(s) without disturbing the existing chain.
+--
 
 --name-hash-version=<n>::
 	Provide this argument to the underlying `git pack-objects` process.
diff --git a/builtin/repack.c b/builtin/repack.c
index 3a5042491d6..9e070f35868 100644
--- a/builtin/repack.c
+++ b/builtin/repack.c
@@ -33,7 +33,7 @@ static int midx_must_contain_cruft = 1;
 static const char *const git_repack_usage[] = {
 	N_("git repack [-a] [-A] [-d] [-f] [-F] [-l] [-n] [-q] [-b] [-m]\n"
 	   "[--window=<n>] [--depth=<n>] [--threads=<n>] [--keep-pack=<pack-name>]\n"
-	   "[--write-midx] [--name-hash-version=<n>] [--path-walk]"),
+	   "[--write-midx[=<mode>]] [--name-hash-version=<n>] [--path-walk]"),
 	NULL
 };
 
@@ -42,9 +42,14 @@ static const char incremental_bitmap_conflict_error[] = N_(
 "--no-write-bitmap-index or disable the pack.writeBitmaps configuration."
 );
 
+#define DEFAULT_MIDX_SPLIT_FACTOR 2
+#define DEFAULT_MIDX_NEW_LAYER_THRESHOLD 8
+
 struct repack_config_ctx {
 	struct pack_objects_args *po_args;
 	struct pack_objects_args *cruft_po_args;
+	int midx_split_factor;
+	int midx_new_layer_threshold;
 };
 
 static int repack_config(const char *var, const char *value,
@@ -94,6 +99,16 @@ static int repack_config(const char *var, const char *value,
 		midx_must_contain_cruft = git_config_bool(var, value);
 		return 0;
 	}
+	if (!strcmp(var, "repack.midxsplitfactor")) {
+		repack_ctx->midx_split_factor = git_config_int(var, value,
+							       ctx->kvi);
+		return 0;
+	}
+	if (!strcmp(var, "repack.midxnewlayerthreshold")) {
+		repack_ctx->midx_new_layer_threshold = git_config_int(var, value,
+								      ctx->kvi);
+		return 0;
+	}
 	return git_default_config(var, value, ctx, cb);
 }
 
@@ -109,6 +124,8 @@ static int option_parse_write_midx(const struct option *opt, const char *arg,
 
 	if (!arg || !*arg)
 		*cfg = REPACK_WRITE_MIDX_DEFAULT;
+	else if (!strcmp(arg, "incremental"))
+		*cfg = REPACK_WRITE_MIDX_INCREMENTAL;
 	else
 		return error(_("unknown value for %s: %s"), opt->long_name, arg);
 
@@ -223,6 +240,8 @@ int cmd_repack(int argc,
 	memset(&config_ctx, 0, sizeof(config_ctx));
 	config_ctx.po_args = &po_args;
 	config_ctx.cruft_po_args = &cruft_po_args;
+	config_ctx.midx_split_factor = DEFAULT_MIDX_SPLIT_FACTOR;
+	config_ctx.midx_new_layer_threshold = DEFAULT_MIDX_NEW_LAYER_THRESHOLD;
 
 	repo_config(repo, repack_config, &config_ctx);
 
@@ -244,6 +263,9 @@ int cmd_repack(int argc,
 	if (pack_everything & PACK_CRUFT)
 		pack_everything |= ALL_INTO_ONE;
 
+	if (write_midx == REPACK_WRITE_MIDX_INCREMENTAL && !geometry.split_factor)
+		die(_("--write-midx=incremental requires --geometric"));
+
 	if (write_bitmaps < 0) {
 		if (write_midx == REPACK_WRITE_MIDX_NONE &&
 		    (!(pack_everything & ALL_INTO_ONE) || !is_bare_repository()))
@@ -293,6 +315,10 @@ int cmd_repack(int argc,
 	if (geometry.split_factor) {
 		if (pack_everything)
 			die(_("options '%s' and '%s' cannot be used together"), "--geometric", "-A/-a");
+		if (write_midx == REPACK_WRITE_MIDX_INCREMENTAL) {
+			geometry.midx_layer_threshold = config_ctx.midx_new_layer_threshold;
+			geometry.midx_layer_threshold_set = true;
+		}
 		pack_geometry_init(&geometry, &existing, &po_args);
 		pack_geometry_split(&geometry);
 	}
@@ -540,6 +566,8 @@ int cmd_repack(int argc,
 			.show_progress = show_progress,
 			.write_bitmaps = write_bitmaps > 0,
 			.midx_must_contain_cruft = midx_must_contain_cruft,
+			.midx_split_factor = config_ctx.midx_split_factor,
+			.midx_new_layer_threshold = config_ctx.midx_new_layer_threshold,
 			.mode = write_midx,
 		};
 
@@ -552,11 +580,15 @@ int cmd_repack(int argc,
 
 	if (delete_redundant) {
 		int opts = 0;
-		existing_packs_remove_redundant(&existing, packdir);
+		bool wrote_incremental_midx = write_midx == REPACK_WRITE_MIDX_INCREMENTAL;
+
+		existing_packs_remove_redundant(&existing, packdir,
+						wrote_incremental_midx);
 
 		if (geometry.split_factor)
 			pack_geometry_remove_redundant(&geometry, &names,
-						       &existing, packdir);
+						       &existing, packdir,
+						       wrote_incremental_midx);
 		if (show_progress)
 			opts |= PRUNE_PACKED_VERBOSE;
 		prune_packed_objects(opts);
diff --git a/midx.c b/midx.c
index 709fe7c3fd7..5f024d061e6 100644
--- a/midx.c
+++ b/midx.c
@@ -837,6 +837,36 @@ void clear_midx_file(struct repository *r)
 	strbuf_release(&midx);
 }
 
+void clear_incremental_midx_files(struct repository *r,
+				  const struct strvec *keep_hashes)
+{
+	struct strbuf chain = STRBUF_INIT;
+
+	get_midx_chain_filename(r->objects->sources, &chain);
+
+	if (r->objects) {
+		struct odb_source *source = r->objects->sources;
+		for (; source; source = source->next) {
+			if (source->packfiles->midx)
+				close_midx(source->packfiles->midx);
+			source->packfiles->midx = NULL;
+		}
+	}
+
+	if (!keep_hashes && remove_path(chain.buf))
+		die(_("failed to clear multi-pack-index chain at %s"),
+		    chain.buf);
+
+	clear_incremental_midx_files_ext(r->objects->sources, MIDX_EXT_BITMAP,
+					 keep_hashes);
+	clear_incremental_midx_files_ext(r->objects->sources, MIDX_EXT_REV,
+					 keep_hashes);
+	clear_incremental_midx_files_ext(r->objects->sources, MIDX_EXT_MIDX,
+					 keep_hashes);
+
+	strbuf_release(&chain);
+}
+
 static int verify_midx_error;
 
 __attribute__((format (printf, 1, 2)))
diff --git a/midx.h b/midx.h
index f211a38b9e7..b45da0a3144 100644
--- a/midx.h
+++ b/midx.h
@@ -9,6 +9,7 @@ struct repository;
 struct bitmapped_pack;
 struct git_hash_algo;
 struct odb_source;
+struct strvec;
 
 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
 #define MIDX_VERSION_V1 1
@@ -143,6 +144,8 @@ int write_midx_file_compact(struct odb_source *source,
 			    const char *incremental_base,
 			    unsigned flags);
 void clear_midx_file(struct repository *r);
+void clear_incremental_midx_files(struct repository *r,
+				  const struct strvec *keep_hashes);
 int verify_midx_file(struct odb_source *source, unsigned flags);
 int expire_midx_packs(struct odb_source *source, unsigned flags);
 int midx_repack(struct odb_source *source, size_t batch_size, unsigned flags);
diff --git a/repack-geometry.c b/repack-geometry.c
index d2065205f87..5b554da89f1 100644
--- a/repack-geometry.c
+++ b/repack-geometry.c
@@ -251,7 +251,8 @@ static void remove_redundant_packs(struct packed_git **pack,
 				   uint32_t pack_nr,
 				   struct string_list *names,
 				   struct existing_packs *existing,
-				   const char *packdir)
+				   const char *packdir,
+				   bool wrote_incremental_midx)
 {
 	const struct git_hash_algo *algop = existing->repo->hash_algo;
 	struct strbuf buf = STRBUF_INIT;
@@ -271,7 +272,8 @@ static void remove_redundant_packs(struct packed_git **pack,
 		    (string_list_has_string(&existing->kept_packs, buf.buf)))
 			continue;
 
-		repack_remove_redundant_pack(existing->repo, packdir, buf.buf);
+		repack_remove_redundant_pack(existing->repo, packdir, buf.buf,
+					     wrote_incremental_midx);
 	}
 
 	strbuf_release(&buf);
@@ -280,12 +282,13 @@ static void remove_redundant_packs(struct packed_git **pack,
 void pack_geometry_remove_redundant(struct pack_geometry *geometry,
 				    struct string_list *names,
 				    struct existing_packs *existing,
-				    const char *packdir)
+				    const char *packdir,
+				    bool wrote_incremental_midx)
 {
 	remove_redundant_packs(geometry->pack, geometry->split,
-			       names, existing, packdir);
+			       names, existing, packdir, wrote_incremental_midx);
 	remove_redundant_packs(geometry->promisor_pack, geometry->promisor_split,
-			       names, existing, packdir);
+			       names, existing, packdir, wrote_incremental_midx);
 }
 
 void pack_geometry_release(struct pack_geometry *geometry)
diff --git a/repack-midx.c b/repack-midx.c
index ad78a10378b..5da67f854c7 100644
--- a/repack-midx.c
+++ b/repack-midx.c
@@ -11,6 +11,7 @@
 #include "refs.h"
 #include "run-command.h"
 #include "tempfile.h"
+#include "trace2.h"
 
 struct midx_snapshot_ref_data {
 	struct repository *repo;
@@ -567,6 +568,9 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 	uint32_t i;
 	int ret = 0;
 
+	trace2_region_enter("repack", "make_midx_compaction_plan",
+			    opts->existing->repo);
+
 	odb_reprepare(opts->existing->repo->objects);
 	m = get_multi_pack_index(opts->existing->source);
 
@@ -578,6 +582,8 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 		}
 	}
 
+	trace2_region_enter("repack", "steps:write", opts->existing->repo);
+
 	/*
 	 * The first MIDX in the resulting chain is always going to be
 	 * new.
@@ -593,13 +599,16 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 	 * was *not* rewritten) or the old tip's base MIDX layer
 	 * (otherwise).
 	 */
-
 	step.type = MIDX_COMPACTION_STEP_WRITE;
 	string_list_init_nodup(&step.u.write);
 
 	for (i = 0; i < opts->names->nr; i++) {
 		strbuf_addf(&buf, "pack-%s.idx", opts->names->items[i].string);
 		string_list_append(&step.u.write, strbuf_detach(&buf, NULL));
+
+		trace2_data_string("repack", opts->existing->repo,
+				   "include:fresh",
+				   step.u.write.items[step.u.write.nr - 1].string);
 	}
 	for (i = 0; i < opts->geometry->split; i++) {
 		struct packed_git *p = opts->geometry->pack[i];
@@ -610,6 +619,9 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 
 		step.objects_nr += p->num_objects;
 	}
+	trace2_data_intmax("repack", opts->existing->repo,
+			   "include:fresh:objects_nr",
+			   (uintmax_t)step.objects_nr);
 
 	/*
 	 * Now handle any existing packs which were *not* rewritten.
@@ -638,8 +650,18 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 		strbuf_strip_suffix(&buf, ".pack");
 		strbuf_addstr(&buf, ".idx");
 
-		if (p->multi_pack_index && !opts->geometry->midx_tip_rewritten)
+		if (p->multi_pack_index &&
+		    !opts->geometry->midx_tip_rewritten) {
+			trace2_data_string("repack", opts->existing->repo,
+					   "exclude:unmodified", buf.buf);
 			continue;
+		}
+
+		trace2_data_string("repack", opts->existing->repo,
+				   "include:unmodified", buf.buf);
+		trace2_data_string("repack", opts->existing->repo,
+				   "include:unmodified:midx",
+				   p->multi_pack_index ? "true" : "false");
 
 		item = string_list_append(&step.u.write,
 					  strbuf_detach(&buf, NULL));
@@ -650,8 +672,12 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 			ret = error(_("too many objects in MIDX compaction step"));
 			goto out;
 		}
+
 		step.objects_nr += p->num_objects;
 	}
+	trace2_data_intmax("repack", opts->existing->repo,
+			   "include:unmodified:objects_nr",
+			   (uintmax_t)step.objects_nr);
 
 	/*
 	 * If the MIDX tip was rewritten, then we no longer consider it
@@ -661,6 +687,11 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 	if (opts->geometry->midx_tip_rewritten)
 		m = m->base_midx;
 
+	trace2_data_string("repack", opts->existing->repo, "midx:rewrote-tip",
+			   opts->geometry->midx_tip_rewritten ? "true" : "false");
+
+	trace2_region_enter("repack", "compact", opts->existing->repo);
+
 	/*
 	 * Compact additional MIDX layers into this proposed one until
 	 * the merging condition is violated.
@@ -668,12 +699,27 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 	while (m) {
 		uint32_t preferred_pack_idx;
 
+		trace2_data_string("repack", opts->existing->repo,
+				   "candidate", midx_get_checksum_hex(m));
+
 		if (step.objects_nr < m->num_objects / opts->midx_split_factor) {
 			/*
 			 * Stop compacting MIDX layer as soon as the
 			 * merged size is less than half the size of the
 			 * next layer in the chain.
 			 */
+			trace2_data_string("repack", opts->existing->repo,
+					   "compact", "violated");
+			trace2_data_intmax("repack", opts->existing->repo,
+					   "objects_nr",
+					   (uintmax_t)step.objects_nr);
+			trace2_data_intmax("repack", opts->existing->repo,
+					   "next_objects_nr",
+					   (uintmax_t)m->num_objects);
+			trace2_data_intmax("repack", opts->existing->repo,
+					   "split_factor",
+					   (uintmax_t)opts->midx_split_factor);
+
 			break;
 		}
 
@@ -693,6 +739,9 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 			strbuf_strip_suffix(&buf, ".pack");
 			strbuf_addstr(&buf, ".idx");
 
+			trace2_data_string("repack", opts->existing->repo,
+					   "midx:pack", buf.buf);
+
 			item = string_list_append(&step.u.write,
 						  strbuf_detach(&buf, NULL));
 			if (pack_int_id == preferred_pack_idx)
@@ -717,6 +766,16 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 		steps[steps_nr++] = step;
 	}
 
+	trace2_data_intmax("repack", opts->existing->repo,
+			   "step:objects_nr", (uintmax_t)step.objects_nr);
+	trace2_data_intmax("repack", opts->existing->repo,
+			   "step:packs_nr", (uintmax_t)step.u.write.nr);
+
+	trace2_region_leave("repack", "compact", opts->existing->repo);
+	trace2_region_leave("repack", "steps:write", opts->existing->repo);
+
+	trace2_region_enter("repack", "steps:rest", opts->existing->repo);
+
 	/*
 	 * Then start over, repeat, and either compact or keep as-is
 	 * each MIDX layer until we have exhausted the chain.
@@ -734,18 +793,29 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 		memset(&step, 0, sizeof(step));
 		step.type = MIDX_COMPACTION_STEP_UNKNOWN;
 
+		trace2_region_enter("repack", "step", opts->existing->repo);
+
+		trace2_data_string("repack", opts->existing->repo,
+				   "from", midx_get_checksum_hex(m));
+
 		while (next) {
-			struct multi_pack_index *base = next->base_midx;
 			uint32_t proposed_objects_nr;
-
 			if (unsigned_add_overflows(step.objects_nr, next->num_objects)) {
 				ret = error(_("too many objects in MIDX compaction step"));
+				trace2_region_leave("repack", "step", opts->existing->repo);
 				goto out;
 			}
 
 			proposed_objects_nr = step.objects_nr + next->num_objects;
 
-			if (!base) {
+			trace2_data_string("repack", opts->existing->repo,
+					   "proposed",
+					   midx_get_checksum_hex(next));
+			trace2_data_intmax("repack", opts->existing->repo,
+					   "proposed:objects_nr",
+					   (uintmax_t)next->num_objects);
+
+			if (!next->base_midx) {
 				/*
 				 * If we are at the end of the MIDX
 				 * chain, there is nothing to compact,
@@ -755,7 +825,7 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 				break;
 			}
 
-			if (proposed_objects_nr < base->num_objects / opts->midx_split_factor) {
+			if (proposed_objects_nr < next->base_midx->num_objects / opts->midx_split_factor) {
 				/*
 				 * If there is a MIDX following this
 				 * one, but our accumulated size is less
@@ -763,6 +833,13 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 				 * them would violate the merging
 				 * condition, so stop here.
 				 */
+
+				trace2_data_string("repack", opts->existing->repo,
+						   "compact:violated:at",
+						   midx_get_checksum_hex(next->base_midx));
+				trace2_data_intmax("repack", opts->existing->repo,
+						   "compact:violated:at:objects_nr",
+						   (uintmax_t)next->base_midx->num_objects);
 				break;
 			}
 
@@ -772,27 +849,43 @@ static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 			 * through the remainder of the chain.
 			 */
 			step.objects_nr = proposed_objects_nr;
-			next = base;
+			trace2_data_intmax("repack", opts->existing->repo,
+					   "step:objects_nr",
+					   (uintmax_t)step.objects_nr);
+			next = next->base_midx;
 		}
 
 		if (m == next) {
 			step.type = MIDX_COMPACTION_STEP_COPY;
 			step.u.copy = m;
+
+			trace2_data_string("repack", opts->existing->repo,
+					   "type", "copy");
 		} else {
 			step.type = MIDX_COMPACTION_STEP_COMPACT;
 			step.u.compact.from = next;
 			step.u.compact.to = m;
+
+			trace2_data_string("repack", opts->existing->repo,
+					   "to", midx_get_checksum_hex(m));
+			trace2_data_string("repack", opts->existing->repo,
+					   "type", "compact");
 		}
 
 		m = next->base_midx;
-
 		steps[steps_nr++] = step;
+		trace2_region_leave("repack", "step", opts->existing->repo);
 	}
 
+	trace2_region_leave("repack", "steps:rest", opts->existing->repo);
+
 out:
 	*steps_p = steps;
 	*steps_nr_p = steps_nr;
 
+	trace2_region_leave("repack", "make_midx_compaction_plan",
+			    opts->existing->repo);
+
 	return ret;
 }
 
@@ -801,6 +894,7 @@ static int write_midx_incremental(struct repack_write_midx_opts *opts)
 	struct midx_compaction_step *steps = NULL;
 	struct strbuf lock_name = STRBUF_INIT;
 	struct lock_file lf;
+	struct strvec keep_hashes = STRVEC_INIT;
 	size_t steps_nr = 0;
 	size_t i;
 	int ret = 0;
@@ -846,10 +940,13 @@ static int write_midx_incremental(struct repack_write_midx_opts *opts)
 			BUG("missing result for compaction step %"PRIuMAX,
 			    (uintmax_t)i);
 		fprintf(get_lock_file_fp(&lf), "%s\n", step->csum);
+		strvec_push(&keep_hashes, step->csum);
 	}
 
 	commit_lock_file(&lf);
 
+	clear_incremental_midx_files(opts->existing->repo, &keep_hashes);
+
 done:
 	strbuf_release(&lock_name);
 	free(steps);
diff --git a/repack.c b/repack.c
index 2ee6b51420a..be2cc0e9d44 100644
--- a/repack.c
+++ b/repack.c
@@ -55,14 +55,18 @@ void pack_objects_args_release(struct pack_objects_args *args)
 }
 
 void repack_remove_redundant_pack(struct repository *repo, const char *dir_name,
-				  const char *base_name)
+				  const char *base_name,
+				  bool wrote_incremental_midx)
 {
 	struct strbuf buf = STRBUF_INIT;
 	struct odb_source *source = repo->objects->sources;
 	struct multi_pack_index *m = get_multi_pack_index(source);
 	strbuf_addf(&buf, "%s.pack", base_name);
-	if (m && source->local && midx_contains_pack(m, buf.buf))
+	if (m && source->local && midx_contains_pack(m, buf.buf)) {
 		clear_midx_file(repo);
+		if (!wrote_incremental_midx)
+			clear_incremental_midx_files(repo, NULL);
+	}
 	strbuf_insertf(&buf, 0, "%s/", dir_name);
 	unlink_pack_path(buf.buf, 1);
 	strbuf_release(&buf);
@@ -252,23 +256,26 @@ void existing_packs_mark_for_deletion(struct existing_packs *existing,
 
 static void remove_redundant_packs_1(struct repository *repo,
 				     struct string_list *packs,
-				     const char *packdir)
+				     const char *packdir,
+				     bool wrote_incremental_midx)
 {
 	struct string_list_item *item;
 	for_each_string_list_item(item, packs) {
 		if (!existing_pack_is_marked_for_deletion(item))
 			continue;
-		repack_remove_redundant_pack(repo, packdir, item->string);
+		repack_remove_redundant_pack(repo, packdir, item->string,
+					     wrote_incremental_midx);
 	}
 }
 
 void existing_packs_remove_redundant(struct existing_packs *existing,
-				     const char *packdir)
+				     const char *packdir,
+				     bool wrote_incremental_midx)
 {
 	remove_redundant_packs_1(existing->repo, &existing->non_kept_packs,
-				 packdir);
+				 packdir, wrote_incremental_midx);
 	remove_redundant_packs_1(existing->repo, &existing->cruft_packs,
-				 packdir);
+				 packdir, wrote_incremental_midx);
 }
 
 void existing_packs_release(struct existing_packs *existing)
diff --git a/repack.h b/repack.h
index 831ccfb1c6c..d2876f569a6 100644
--- a/repack.h
+++ b/repack.h
@@ -34,7 +34,8 @@ void prepare_pack_objects(struct child_process *cmd,
 void pack_objects_args_release(struct pack_objects_args *args);
 
 void repack_remove_redundant_pack(struct repository *repo, const char *dir_name,
-				  const char *base_name);
+				  const char *base_name,
+				  bool wrote_incremental_midx);
 
 struct write_pack_opts {
 	struct pack_objects_args *po_args;
@@ -84,7 +85,8 @@ void existing_packs_retain_cruft(struct existing_packs *existing,
 void existing_packs_mark_for_deletion(struct existing_packs *existing,
 				      struct string_list *names);
 void existing_packs_remove_redundant(struct existing_packs *existing,
-				     const char *packdir);
+				     const char *packdir,
+				     bool wrote_incremental_midx);
 void existing_packs_release(struct existing_packs *existing);
 
 struct generated_pack;
@@ -129,7 +131,8 @@ struct packed_git *pack_geometry_preferred_pack(struct pack_geometry *geometry);
 void pack_geometry_remove_redundant(struct pack_geometry *geometry,
 				    struct string_list *names,
 				    struct existing_packs *existing,
-				    const char *packdir);
+				    const char *packdir,
+				    bool wrote_incremental_midx);
 void pack_geometry_release(struct pack_geometry *geometry);
 
 struct tempfile;
diff --git a/t/meson.build b/t/meson.build
index 2421220917a..4f7c7d5691d 100644
--- a/t/meson.build
+++ b/t/meson.build
@@ -941,6 +941,7 @@ integration_tests = [
   't7702-repack-cyclic-alternate.sh',
   't7703-repack-geometric.sh',
   't7704-repack-cruft.sh',
+  't7705-repack-incremental-midx.sh',
   't7800-difftool.sh',
   't7810-grep.sh',
   't7811-grep-open.sh',
diff --git a/t/t7705-repack-incremental-midx.sh b/t/t7705-repack-incremental-midx.sh
new file mode 100755
index 00000000000..f81c2c67060
--- /dev/null
+++ b/t/t7705-repack-incremental-midx.sh
@@ -0,0 +1,436 @@
+#!/bin/sh
+
+test_description='git repack --write-midx=incremental'
+
+. ./test-lib.sh
+
+GIT_TEST_MULTI_PACK_INDEX=0
+GIT_TEST_MULTI_PACK_INDEX_WRITE_BITMAP=0
+GIT_TEST_MULTI_PACK_INDEX_WRITE_INCREMENTAL=0
+
+objdir=.git/objects
+packdir=$objdir/pack
+midxdir=$packdir/multi-pack-index.d
+midx_chain=$midxdir/multi-pack-index-chain
+
+# incrementally_repack N
+#
+# Make "N" new commits, each stored in their own pack, and then repacked
+# with the --write-midx=incremental strategy.
+incrementally_repack () {
+	for i in $(test_seq 1 "$1")
+	do
+		test_commit "$i" &&
+
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+		git multi-pack-index verify || return 1
+	done
+}
+
+# Create packs with geometrically increasing sizes so that they
+# satisfy the geometric progression and survive a --geometric=2
+# repack without being rolled up. Creates 3 packs containing 1,
+# 2, and 6 commits (3, 6, and 18 objects) respectively.
+create_geometric_packs () {
+	test_commit "small" &&
+	git repack -d &&
+
+	test_commit_bulk --message="medium" 2 &&
+	test_commit_bulk --message="large" 6 &&
+
+	git repack --geometric=2 -d --write-midx=incremental \
+		--write-bitmap-index
+}
+
+# create_layer <test_commit_bulk args>
+#
+# Creates a new MIDX layer with the contents of "test_commit_bulk $@".
+create_layer () {
+	test_commit_bulk "$@" &&
+
+	git multi-pack-index write --incremental --bitmap
+}
+
+# create_layers
+#
+# Reads lines of "<message> <nr>" from stdin and creates a new MIDX
+# layer for each line. See create_layer above for more.
+create_layers () {
+	while read msg nr
+	do
+		create_layer --message="$msg" "$nr" || return 1
+	done
+}
+
+test_expect_success '--write-midx=incremental requires --geometric' '
+	test_must_fail git repack --write-midx=incremental 2>err &&
+
+	test_grep -- "--write-midx=incremental requires --geometric" err
+'
+
+test_expect_success 'below layer threshold, tip packs excluded' '
+	git init below-layer-threshold-tip-packs-excluded &&
+	(
+		cd below-layer-threshold-tip-packs-excluded &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 4 &&
+		git config repack.midxsplitfactor 2 &&
+
+		# Create 3 packs forming a geometric progression by
+		# object count such that they are unmodified by the
+		# initial repack. The MIDX chain thusly contains a
+		# single layer with three packs.
+		create_geometric_packs &&
+		ls $packdir/pack-*.idx | sort >packs.before &&
+		test_line_count = 1 $midx_chain &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Repack a new commit. Since the layer threshold is
+		# unmet, a new MIDX layer is added on top of the
+		# existing one.
+		test_commit extra &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+		git multi-pack-index verify &&
+
+		ls $packdir/pack-*.idx | sort >packs.after &&
+		comm -13 packs.before packs.after >packs.new &&
+		test_line_count = 1 packs.new &&
+
+		test_line_count = 2 "$midx_chain" &&
+		head -n 1 "$midx_chain.before" >expect &&
+		head -n 1 "$midx_chain" >actual &&
+		test_cmp expect actual
+	)
+'
+
+test_expect_success 'above layer threshold, tip packs repacked' '
+	git init above-layer-threshold-tip-packs-repacked &&
+	(
+		cd above-layer-threshold-tip-packs-repacked &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 2 &&
+		git config repack.midxsplitfactor 2 &&
+
+		# Same setup, but with the layer threshold set to 2.
+		# Since the tip MIDX layer meets that threshold, its
+		# packs are considered repack candidates.
+		create_geometric_packs &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Perturb the existing progression such that it is
+		# rolled up into a single new pack, invalidating the
+		# existing MIDX layer and replacing it with a new one.
+		test_commit extra &&
+		git repack -d &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		! test_cmp $midx_chain.before $midx_chain &&
+		test_line_count = 1 $midx_chain &&
+
+		git multi-pack-index verify
+	)
+'
+
+test_expect_success 'above layer threshold, tip layer preserved' '
+	git init above-layer-threshold-tip-layer-preserved &&
+	(
+		cd above-layer-threshold-tip-layer-preserved &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 2 &&
+		git config repack.midxsplitfactor 2 &&
+
+		test_commit_bulk --message="medium" 2 &&
+		test_commit_bulk --message="large" 6 &&
+
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		test_line_count = 1 "$midx_chain" &&
+		ls $packdir/pack-*.idx | sort >packs.before &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Create objects to form a pack satisfying the geometric
+		# progression (thus preserving the tip layer), but not
+		# so large that it meets the layer merging condition.
+		test_commit_bulk --message="small" 1 &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		ls $packdir/pack-*.idx | sort >packs.after &&
+		comm -13 packs.before packs.after >packs.new &&
+
+		test_line_count = 1 packs.new &&
+		test_line_count = 3 packs.after &&
+		test_line_count = 2 "$midx_chain" &&
+		head -n 1 "$midx_chain.before" >expect &&
+		head -n 1 "$midx_chain" >actual &&
+		test_cmp expect actual &&
+
+		git multi-pack-index verify
+	)
+'
+
+test_expect_success 'above layer threshold, tip packs preserved' '
+	git init above-layer-threshold-tip-packs-preserved &&
+	(
+		cd above-layer-threshold-tip-packs-preserved &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 2 &&
+		git config repack.midxsplitfactor 2 &&
+
+		create_geometric_packs &&
+		ls $packdir/pack-*.idx | sort >packs.before &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Same setup as above, but this time the new objects do
+		# not satisfy the new layer merging condition, resulting
+		# in a new tip layer.
+		test_commit_bulk --message="huge" 18 &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		ls $packdir/pack-*.idx | sort >packs.after &&
+		comm -13 packs.before packs.after >packs.new &&
+
+		! test_cmp $midx_chain.before $midx_chain &&
+		test_line_count = 1 $midx_chain &&
+		test_line_count = 1 packs.new &&
+
+		git multi-pack-index verify
+	)
+'
+
+test_expect_success 'new tip absorbs multiple layers' '
+	git init new-tip-absorbs-multiple-layers &&
+	(
+		cd new-tip-absorbs-multiple-layers &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		# Build a 4-layer chain where each layer is too small to
+		# absorb the one below it. The sizes must satisfy L(n) <
+		# L(n-1)/2 for each adjacent pair:
+		#
+		#   L0 (oldest): 75 obj (25 commits)
+		#   L1:          21 obj  (7 commits, 21 < 75/2)
+		#   L2:           9 obj  (3 commits,  9 < 21/2)
+		#   L3 (tip):     3 obj  (1 commit,   3 <  9/2)
+		create_layers <<-\EOF &&
+		L0 25
+		L1 7
+		L2 3
+		L3 1
+		EOF
+
+		test_line_count = 4 "$midx_chain" &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Now add a new commit. The merging condition is
+		# satisfied between L3-L1, but violated at L0, which is
+		# too large relative to the accumulated size.
+		#
+		# As a result, the chain shrinks from 4 to 2 layers.
+		test_commit new &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		! test_cmp $midx_chain.before $midx_chain &&
+		test_line_count = 2 "$midx_chain" &&
+		git multi-pack-index verify
+	)
+'
+
+test_expect_success 'compaction of older layers' '
+	git init compaction-of-older-layers &&
+	(
+		cd compaction-of-older-layers &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		# Build a chain with two small layers at the bottom
+		# and a larger barrier layer on top, producing a
+		# chain that violates the compaction invariant, since
+		# the two small layers would normally have been merged.
+		create_layers <<-\EOF &&
+		one 2
+		two 4
+		barrier 54
+		EOF
+
+		cp $midx_chain $midx_chain.before &&
+
+		# Running an incremental repack compacts the two
+		# small layers at the bottom of the chain as a
+		# separate step in the compaction plan.
+		test_commit another &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		test_line_count = 2 "$midx_chain" &&
+		git multi-pack-index verify
+	)
+'
+
+test_expect_success 'geometric rollup with surviving tip packs' '
+	git init geometric-rollup-with-surviving-tip-packs &&
+	(
+		cd geometric-rollup-with-surviving-tip-packs &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		# Create a pack large enough to anchor the geometric
+		# progression when small packs are added alongside it.
+		create_layer --message="big" 5 &&
+
+		test_line_count = 1 "$midx_chain" &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Repack a small number of objects such that the
+		# progression is unbothered. Note that the existing pack
+		# is considered a repack candidate as the new layer
+		# threshold is set to 1.
+		test_commit small-1 &&
+		git repack -d &&
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		! test_cmp $midx_chain.before $midx_chain &&
+		cp $midx_chain $midx_chain.before
+	)
+'
+
+test_expect_success 'kept packs are excluded from repack' '
+	git init kept-packs-excluded-from-repack &&
+	(
+		cd kept-packs-excluded-from-repack &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		# Create two equal-sized packs, marking one as kept.
+		for i in A B
+		do
+			test_commit "$i" && git repack -d || return 1
+		done &&
+
+		keep=$(ls $packdir/pack-*.idx | head -n 1) &&
+		touch "${keep%.idx}.keep" &&
+
+		# The kept pack is excluded as a repacking candidate
+		# entirely, so no rollup occurs as there is only one
+		# non-kept pack. A new MIDX layer is written containing
+		# that pack.
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		test-tool read-midx $objdir >actual &&
+		grep "^pack-.*\.idx$" actual >actual.packs &&
+		test_line_count = 1 actual.packs &&
+		test_grep ! "$keep" actual.packs &&
+
+		git multi-pack-index verify &&
+
+		# All objects (from both kept and non-kept packs)
+		# must still be accessible.
+		git fsck
+	)
+'
+
+test_expect_success 'incremental MIDX with --max-pack-size' '
+	git init incremental-midx-with--max-pack-size &&
+	(
+		cd incremental-midx-with--max-pack-size &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		create_layer --message="base" 1 &&
+
+		# Now add enough data that a small --max-pack-size will
+		# cause pack-objects to split its output. Create objects
+		# large enough to fill multiple packs.
+		test-tool genrandom foo 1M >big1 &&
+		test-tool genrandom bar 1M >big2 &&
+		git add big1 big2 &&
+		test_tick &&
+		git commit -a -m "big blobs" &&
+		git repack -d &&
+
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index --max-pack-size=1M &&
+
+		test_line_count = 1 "$midx_chain" &&
+		test-tool read-midx $objdir >actual &&
+		grep "^pack-.*\.idx$" actual >actual.packs &&
+		test_line_count -gt 1 actual.packs &&
+
+		git multi-pack-index verify
+	)
+'
+
+test_expect_success 'noop repack preserves valid MIDX chain' '
+	git init noop-repack-preserves-valid-midx-chain &&
+	(
+		cd noop-repack-preserves-valid-midx-chain &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		create_layer --message="base" 1 &&
+
+		git multi-pack-index verify &&
+		cp $midx_chain $midx_chain.before &&
+
+		# Running again with no new objects should not break
+		# the MIDX chain. It produces "Nothing new to pack."
+		git repack --geometric=2 -d --write-midx=incremental \
+			--write-bitmap-index &&
+
+		test_cmp $midx_chain.before $midx_chain &&
+
+		git multi-pack-index verify &&
+		git fsck
+	)
+'
+
+test_expect_success 'repack -ad removes stale incremental chain' '
+	git init repack--ad-removes-stale-incremental-chain &&
+	(
+		cd repack--ad-removes-stale-incremental-chain &&
+
+		git config maintenance.auto false &&
+		git config repack.midxnewlayerthreshold 1 &&
+		git config repack.midxsplitfactor 2 &&
+
+		create_layers <<-\EOF &&
+		one 1
+		two 1
+		EOF
+
+		test_path_is_file $midx_chain &&
+		test_line_count = 2 $midx_chain &&
+
+		git repack -ad &&
+
+		test_path_is_missing $packdir/multi-pack-index &&
+		test_dir_is_empty $midxdir
+	)
+'
+
+test_done
-- 
2.53.0.185.g29bc4dff628

[RFC PATCH 14/14] repack: allow `--write-midx=incremental` without `--geometric`

[Taylor Blau]

Previously, `--write-midx=incremental` required `--geometric` and would
die() without it. Relax this restriction so that incremental MIDX
repacking can be used independently.

Without `--geometric`, the behavior is append-only: a single new MIDX
layer is created containing whatever packs were written by the repack
and appended to the existing chain (or a new chain is started). Existing
layers are preserved as-is with no compaction or merging.

Implement this via a new repack_make_midx_append_plan() that builds a
plan consisting of a WRITE step for the freshly written packs followed
by COPY steps for every existing MIDX layer. The existing compaction
plan (repack_make_midx_compaction_plan) is used only when `--geometric`
is active.

Update the documentation to describe the behavior with and without
`--geometric`, and replace the test that enforced the old restriction
with one exercising append-only incremental MIDX repacking.

Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
 Documentation/git-repack.adoc      | 19 +++++----
 builtin/repack.c                   |  3 --
 repack-midx.c                      | 62 ++++++++++++++++++++++++++++--
 t/t7705-repack-incremental-midx.sh | 35 ++++++++++++++---
 4 files changed, 101 insertions(+), 18 deletions(-)

diff --git a/Documentation/git-repack.adoc b/Documentation/git-repack.adoc
index 27a99cc46f4..72c42015e23 100644
--- a/Documentation/git-repack.adoc
+++ b/Documentation/git-repack.adoc
@@ -263,14 +263,19 @@ linkgit:git-multi-pack-index[1]).
 
 	`incremental`;;
 		Write an incremental MIDX chain instead of a single
-		flat MIDX. This mode requires `--geometric`.
+		flat MIDX.
 +
-The incremental mode maintains a chain of MIDX layers that is compacted
-over time using a geometric merging strategy. Each repack creates a new
-tip layer containing the newly written pack(s). Adjacent layers are then
-merged whenever the newer layer's object count exceeds
-`1/repack.midxSplitFactor` of the next deeper layer's count. Layers
-that do not meet this condition are retained as-is.
+Without `--geometric`, a new MIDX layer is appended to the existing
+chain (or a new chain is started) containing whatever packs were written
+by the repack. Existing layers are preserved as-is.
++
+When combined with `--geometric`, the incremental mode maintains a chain
+of MIDX layers that is compacted over time using a geometric merging
+strategy. Each repack creates a new tip layer containing the newly
+written pack(s). Adjacent layers are then merged whenever the newer
+layer's object count exceeds `1/repack.midxSplitFactor` of the next
+deeper layer's count. Layers that do not meet this condition are
+retained as-is.
 +
 The result is that newer (tip) layers tend to contain many small packs
 with relatively few objects, while older (deeper) layers contain fewer,
diff --git a/builtin/repack.c b/builtin/repack.c
index 9e070f35868..8e187322fe4 100644
--- a/builtin/repack.c
+++ b/builtin/repack.c
@@ -263,9 +263,6 @@ int cmd_repack(int argc,
 	if (pack_everything & PACK_CRUFT)
 		pack_everything |= ALL_INTO_ONE;
 
-	if (write_midx == REPACK_WRITE_MIDX_INCREMENTAL && !geometry.split_factor)
-		die(_("--write-midx=incremental requires --geometric"));
-
 	if (write_bitmaps < 0) {
 		if (write_midx == REPACK_WRITE_MIDX_NONE &&
 		    (!(pack_everything & ALL_INTO_ONE) || !is_bare_repository()))
diff --git a/repack-midx.c b/repack-midx.c
index 5da67f854c7..3ab64937c2e 100644
--- a/repack-midx.c
+++ b/repack-midx.c
@@ -556,6 +556,58 @@ static int midx_compaction_step_exec(struct midx_compaction_step *step,
 	}
 }
 
+/*
+ * Build an append-only MIDX plan: a single WRITE step for the freshly
+ * written packs, plus COPY steps for every existing layer.  No
+ * compaction or merging is performed.
+ */
+static void repack_make_midx_append_plan(struct repack_write_midx_opts *opts,
+					 struct midx_compaction_step **steps_p,
+					 size_t *steps_nr_p)
+{
+	struct multi_pack_index *m;
+	struct midx_compaction_step *steps = NULL;
+	struct midx_compaction_step *step;
+	size_t steps_nr = 0, steps_alloc = 0;
+
+	odb_reprepare(opts->existing->repo->objects);
+	m = get_multi_pack_index(opts->existing->source);
+
+	if (opts->names->nr) {
+		struct strbuf buf = STRBUF_INIT;
+		uint32_t i;
+
+		ALLOC_GROW(steps, st_add(steps_nr, 1), steps_alloc);
+
+		step = &steps[steps_nr++];
+		memset(step, 0, sizeof(*step));
+
+		step->type = MIDX_COMPACTION_STEP_WRITE;
+		string_list_init_nodup(&step->u.write);
+
+		for (i = 0; i < opts->names->nr; i++) {
+			strbuf_addf(&buf, "pack-%s.idx",
+				    opts->names->items[i].string);
+			string_list_append(&step->u.write,
+					   strbuf_detach(&buf, NULL));
+		}
+	}
+
+	for (; m; m = m->base_midx) {
+		ALLOC_GROW(steps, st_add(steps_nr, 1), steps_alloc);
+
+		step = &steps[steps_nr++];
+		memset(step, 0, sizeof(*step));
+
+		step->type = MIDX_COMPACTION_STEP_COPY;
+		step->u.copy = m;
+		step->objects_nr = m->num_objects;
+	}
+
+	*steps_p = steps;
+	*steps_nr_p = steps_nr;
+}
+
 static int repack_make_midx_compaction_plan(struct repack_write_midx_opts *opts,
 					    struct midx_compaction_step **steps_p,
 					    size_t *steps_nr_p)
@@ -911,9 +963,13 @@ static int write_midx_incremental(struct repack_write_midx_opts *opts)
 		goto done;
 	}
 
-	if (repack_make_midx_compaction_plan(opts, &steps, &steps_nr) < 0) {
-		ret = error(_("unable to generate compaction plan"));
-		goto done;
+	if (opts->geometry->split_factor) {
+		if (repack_make_midx_compaction_plan(opts, &steps, &steps_nr) < 0) {
+			ret = error(_("unable to generate compaction plan"));
+			goto done;
+		}
+	} else {
+		repack_make_midx_append_plan(opts, &steps, &steps_nr);
 	}
 
 	for (i = 0; i < steps_nr; i++) {
diff --git a/t/t7705-repack-incremental-midx.sh b/t/t7705-repack-incremental-midx.sh
index f81c2c67060..562554e69b4 100755
--- a/t/t7705-repack-incremental-midx.sh
+++ b/t/t7705-repack-incremental-midx.sh
@@ -63,10 +63,36 @@ create_layers () {
 	done
 }
 
-test_expect_success '--write-midx=incremental requires --geometric' '
-	test_must_fail git repack --write-midx=incremental 2>err &&
+test_expect_success '--write-midx=incremental without --geometric' '
+	git init incremental-without-geometric &&
+	(
+		cd incremental-without-geometric &&
 
-	test_grep -- "--write-midx=incremental requires --geometric" err
+		git config maintenance.auto false &&
+
+		test_commit first &&
+		git repack -d &&
+
+		test_commit second &&
+		git repack --write-midx=incremental &&
+
+		git multi-pack-index verify &&
+		test_line_count = 1 $midx_chain &&
+		cp $midx_chain $midx_chain.before &&
+
+		# A second repack appends a new layer without
+		# disturbing the existing one.
+		test_commit third &&
+		git repack --write-midx=incremental &&
+
+		git multi-pack-index verify &&
+		test_line_count = 2 $midx_chain &&
+		head -n 1 $midx_chain.before >expect &&
+		head -n 1 $midx_chain >actual &&
+		test_cmp expect actual &&
+
+		git fsck
+	)
 '
 
 test_expect_success 'below layer threshold, tip packs excluded' '
@@ -334,8 +360,7 @@ test_expect_success 'kept packs are excluded from repack' '
 		# entirely, so no rollup occurs as there is only one
 		# non-kept pack. A new MIDX layer is written containing
 		# that pack.
-		git repack --geometric=2 -d --write-midx=incremental \
-			--write-bitmap-index &&
+		git repack --geometric=2 -d --write-midx=incremental &&
 
 		test-tool read-midx $objdir >actual &&
 		grep "^pack-.*\.idx$" actual >actual.packs &&
-- 
2.53.0.185.g29bc4dff628

Re: [RFC PATCH 06/14] repack: track the ODB source via existing_packs

[Taylor Blau]

On Tue, Feb 24, 2026 at 07:20:46PM -0500, Taylor Blau wrote:
> Store the ODB source in the `existing_packs` struct and use that in
> place of the raw `repo->objects->sources` access within `cmd_repack()`.

I have no idea why my scripts sent this patch twice, but there are two
copies of [06/14], which are identical. I'll make sure to just send one
of these in the non-RFC version of this series ;-).

Thanks,
Taylor

Re: [RFC PATCH 01/14] midx: use `string_list` for retained MIDX files

[Junio C Hamano]

Taylor Blau <me@ttaylorr.com> writes:

> Both `clear_midx_files_ext()` and `clear_incremental_midx_files_ext()`
> build a list of filenames to keep while pruning stale MIDX files. Today
> they hand-roll an array instead of using a `string_list`, thus requiring
> us to pass an additional length parameter, and makes lookups linear.
>
> Replace the bare array with a `string_list` which can be passed around
> as a single parameter. Though it improves lookup performance, the
> difference is likely immeasurable given how small the keep_hashes array
> typically is.

And if it the lookup performance turns out to be an issue, we can
switch to strmap or something more appropriate.

>
> Signed-off-by: Taylor Blau <me@ttaylorr.com>
> ---
>  midx.c | 56 ++++++++++++++++++++++----------------------------------
>  1 file changed, 22 insertions(+), 34 deletions(-)
>
> diff --git a/midx.c b/midx.c
> index c1b9658240d..c5e3553e2bb 100644
> --- a/midx.c
> +++ b/midx.c
> @@ -755,8 +755,7 @@ int midx_checksum_valid(struct multi_pack_index *m)
>  }
>  
>  struct clear_midx_data {
> -	char **keep;
> -	uint32_t keep_nr;
> +	struct string_list keep;
>  	const char *ext;
>  };
>  
> @@ -764,15 +763,12 @@ static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUS
>  				const char *file_name, void *_data)
>  {
>  	struct clear_midx_data *data = _data;
> -	uint32_t i;
>  
>  	if (!(starts_with(file_name, "multi-pack-index-") &&
>  	      ends_with(file_name, data->ext)))
>  		return;
> -	for (i = 0; i < data->keep_nr; i++) {
> -		if (!strcmp(data->keep[i], file_name))
> -			return;
> -	}
> +	if (string_list_has_string(&data->keep, file_name))
> +		return;
>  	if (unlink(full_path))
>  		die_errno(_("failed to remove %s"), full_path);
>  }
> @@ -780,48 +776,40 @@ static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUS
>  void clear_midx_files_ext(struct odb_source *source, const char *ext,
>  			  const char *keep_hash)
>  {
> -	struct clear_midx_data data;
> -	memset(&data, 0, sizeof(struct clear_midx_data));
> -
> -	if (keep_hash) {
> -		ALLOC_ARRAY(data.keep, 1);
> -
> -		data.keep[0] = xstrfmt("multi-pack-index-%s.%s", keep_hash, ext);
> -		data.keep_nr = 1;
> -	}
> -	data.ext = ext;
> -
> -	for_each_file_in_pack_dir(source->path,
> -				  clear_midx_file_ext,
> -				  &data);
> +	struct clear_midx_data data = {
> +		.keep = STRING_LIST_INIT_NODUP,
> +		.ext = ext,
> +	};
>  
>  	if (keep_hash)
> -		free(data.keep[0]);
> -	free(data.keep);
> +		string_list_insert(&data.keep, xstrfmt("multi-pack-index-%s.%s",
> +						       keep_hash, ext));
> +
> +	for_each_file_in_pack_dir(source->path, clear_midx_file_ext, &data);
> +
> +	string_list_clear(&data.keep, 0);
>  }
>  
>  void clear_incremental_midx_files_ext(struct odb_source *source, const char *ext,
>  				      char **keep_hashes,
>  				      uint32_t hashes_nr)
>  {
> -	struct clear_midx_data data;
> +	struct clear_midx_data data = {
> +		.keep = STRING_LIST_INIT_NODUP,
> +		.ext = ext,
> +	};
>  	uint32_t i;
>  
> -	memset(&data, 0, sizeof(struct clear_midx_data));
> -
> -	ALLOC_ARRAY(data.keep, hashes_nr);
>  	for (i = 0; i < hashes_nr; i++)
> -		data.keep[i] = xstrfmt("multi-pack-index-%s.%s", keep_hashes[i],
> -				       ext);
> -	data.keep_nr = hashes_nr;
> -	data.ext = ext;
> +		string_list_append(&data.keep,
> +				   xstrfmt("multi-pack-index-%s.%s",
> +					   keep_hashes[i], ext));
> +	string_list_sort(&data.keep);
>  
>  	for_each_file_in_pack_subdir(source->path, "multi-pack-index.d",
>  				     clear_midx_file_ext, &data);
>  
> -	for (i = 0; i < hashes_nr; i++)
> -		free(data.keep[i]);
> -	free(data.keep);
> +	string_list_clear(&data.keep, 0);
>  }
>  
>  void clear_midx_file(struct repository *r)

Re: [RFC PATCH 02/14] strvec: introduce `strvec_init_alloc()`

[Junio C Hamano]

Taylor Blau <me@ttaylorr.com> writes:

> When the caller knows upfront how many elements will be pushed onto a
> `strvec`, it is useful to pre-allocate enough space in the array to fit
> that many elements (and one additional slot to store NULL, indicating
> the end of the list.)
>
> Introduce `strvec_init_alloc()`, which allocates the backing array large
> enough to hold `alloc` elements and the termination marker without
> further reallocation.
>
> Signed-off-by: Taylor Blau <me@ttaylorr.com>
> ---
>  strvec.c | 7 +++++++
>  strvec.h | 5 +++++
>  2 files changed, 12 insertions(+)
>
> diff --git a/strvec.c b/strvec.c
> index f8de79f5579..f7f32a53b56 100644
> --- a/strvec.c
> +++ b/strvec.c
> @@ -10,6 +10,13 @@ void strvec_init(struct strvec *array)
>  	memcpy(array, &blank, sizeof(*array));
>  }
>  
> +void strvec_init_alloc(struct strvec *array, size_t alloc)
> +{
> +	CALLOC_ARRAY(array->v, st_add(alloc, 1));
> +	array->nr = 0;
> +	array->alloc = alloc + 1;
> +}

It is not satisifying that strvec_init() does *not* become a thin
wrapper around this that says "my initial allocation is for zero
elements", but that cannot be done easily as a strvec that begins as
an empty one has a small optimization to avoid one-slot allocation
only to store NULL.  So, ... OK.

Re: [RFC PATCH 02/14] strvec: introduce `strvec_init_alloc()`

[Junio C Hamano]

Junio C Hamano <gitster@pobox.com> writes:

> Taylor Blau <me@ttaylorr.com> writes:
>
>> When the caller knows upfront how many elements will be pushed onto a
>> `strvec`, it is useful to pre-allocate enough space in the array to fit
>> that many elements (and one additional slot to store NULL, indicating
>> the end of the list.)
>>
>> Introduce `strvec_init_alloc()`, which allocates the backing array large
>> enough to hold `alloc` elements and the termination marker without
>> further reallocation.
>>
>> Signed-off-by: Taylor Blau <me@ttaylorr.com>
>> ---
>>  strvec.c | 7 +++++++
>>  strvec.h | 5 +++++
>>  2 files changed, 12 insertions(+)
>>
>> diff --git a/strvec.c b/strvec.c
>> index f8de79f5579..f7f32a53b56 100644
>> --- a/strvec.c
>> +++ b/strvec.c
>> @@ -10,6 +10,13 @@ void strvec_init(struct strvec *array)
>>  	memcpy(array, &blank, sizeof(*array));
>>  }
>>  
>> +void strvec_init_alloc(struct strvec *array, size_t alloc)
>> +{
>> +	CALLOC_ARRAY(array->v, st_add(alloc, 1));
>> +	array->nr = 0;
>> +	array->alloc = alloc + 1;
>> +}
>
> It is not satisifying that strvec_init() does *not* become a thin
> wrapper around this that says "my initial allocation is for zero
> elements", but that cannot be done easily as a strvec that begins as
> an empty one has a small optimization to avoid one-slot allocation
> only to store NULL.  So, ... OK.

Actually, we should do the same optimization if a caller explicitly
asks

	strvec_init_alloc(&array, 0);

So perhaps we could do this if we wanted to encapsulate the tricky
bits in a single place for maintainability.

 strvec.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git c/strvec.c w/strvec.c
index f8de79f557..cbe72e9411 100644
--- c/strvec.c
+++ w/strvec.c
@@ -4,10 +4,21 @@
 
 const char *empty_strvec[] = { NULL };
 
+void strvec_init_alloc(struct strvec *array, size_t alloc)
+{
+	if (!alloc) {
+		struct strvec blank = STRVEC_INIT;
+		memcpy(array, &blank, sizeof(*array));
+	} else {
+		CALLOC_ARRAY(array->v, st_add(alloc, 1));
+		array->nr = 0;
+		array->alloc = alloc + 1;
+	}
+}
+
 void strvec_init(struct strvec *array)
 {
-	struct strvec blank = STRVEC_INIT;
-	memcpy(array, &blank, sizeof(*array));
+	strvec_init(array, 0);
 }
 
 void strvec_push_nodup(struct strvec *array, char *value)

Re: [RFC PATCH 01/14] midx: use `string_list` for retained MIDX files

[Taylor Blau]

On Thu, Feb 26, 2026 at 12:29:22PM -0800, Junio C Hamano wrote:
> > Replace the bare array with a `string_list` which can be passed around
> > as a single parameter. Though it improves lookup performance, the
> > difference is likely immeasurable given how small the keep_hashes array
> > typically is.
>
> And if it the lookup performance turns out to be an issue, we can
> switch to strmap or something more appropriate.

True. In practice these should have at most tens of entries, so I
suspect the string_list will be completely fine.

If we are cleaning up so many stale MIDX files that we need a more
efficient data structure just to keep track of those files, we likely
have far greater problems to worry about ;-).

Thanks,
Taylor

Re: [RFC PATCH 02/14] strvec: introduce `strvec_init_alloc()`

[Taylor Blau]

On Thu, Feb 26, 2026 at 12:58:48PM -0800, Junio C Hamano wrote:
> > It is not satisifying that strvec_init() does *not* become a thin
> > wrapper around this that says "my initial allocation is for zero
> > elements", but that cannot be done easily as a strvec that begins as
> > an empty one has a small optimization to avoid one-slot allocation
> > only to store NULL.  So, ... OK.
>
> Actually, we should do the same optimization if a caller explicitly
> asks
>
> 	strvec_init_alloc(&array, 0);
>
> So perhaps we could do this if we wanted to encapsulate the tricky
> bits in a single place for maintainability.

Ah, that is a very satisfying change. I like it and squashed it into my
series with your:

    Helped-by: Junio C Hamano <gitster@pobox.com>

>  void strvec_init(struct strvec *array)
>  {
> -	struct strvec blank = STRVEC_INIT;
> -	memcpy(array, &blank, sizeof(*array));
> +	strvec_init(array, 0);

I suspect you meant `strvec_init_alloc()` here and not `strvec_init()`,
but I applied the change and adjusted the typo before squashing it in.

Thanks,
Taylor