| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
A geometric repack writes a new pack and multi-pack-index and then
deletes the packs the new one subsumes. A process still using the
previous MIDX keeps seeing a removed pack listed as the owner of some
objects. Since a MIDX attributes each object to exactly one pack, such
an object is served only through its recorded owner; if that owner was
just removed, find_pack_entry() cannot serve it -- the MIDX lookup routes
to the missing pack, and the regular pack fallback deliberately skips
every MIDX-covered pack, so a surviving copy in another covered pack
(e.g. a kept base pack) is never consulted.
Unlike the ordinary "a pack's .idx is mapped but its .pack is gone"
race, the second read does not rescue us. Reloading the on-disk pack set
does not reload the borrowed, cached MIDX (freeing it under the code that
caches the "struct multi_pack_index *" would be a use-after-free), so the
stale MIDX keeps routing to the removed pack and the surviving copy stays
hidden behind the covered-pack skip. cat-file, rev-list and pack-objects
can thus all spuriously fail with "unable to read object".
Teach find_pack_entry() to recover. The MIDX lookup now returns a
tri-state, distinguishing an object absent from the MIDX from one it owns
via a pack that can no longer be opened; in the latter case, once the
regular fallback has also missed, scan the MIDX's packs directly for a
surviving copy. Because the return value is no longer a boolean, rename
fill_midx_entry() to midx_fill_entry() so callers must reckon with the
new enum rather than silently treat MIDX_FILL_OWNER_UNAVAILABLE as a hit.
Do the scan only on the second read (OBJECT_INFO_SECOND_READ): by then
the cheaper on-disk reload has run, so an object merely relocated into a
new (uncovered) pack has already been found by the regular fallback, and
only a genuine hidden duplicate reaches the rescan. A QUICK caller that
skips the second read simply accepts the false negative, as QUICK is
designed to.
Reloading the stale MIDX would be a more complete fix but is much more
involved (the borrowers above need proper invalidation), so leave that
for later.
Assisted-by: Claude Opus 4.8 & GPT-5.6 Sol
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When reading packed objects we know to tell apart missing objects and
corrupt objects by returning a positive error code in the former case,
and a negative one in the latter case. We do that by distinguishing
between errors returned by `find_pack_entry()`, which yields the offset
of the object, and `packed_object_info()`, which reads the object
contents.
But even though we already distinguish those cases when reading packed
objects, the logic is broken in case a caller tries to read an object
that has been marked as corrupt. In that case, `find_pack_entry()` will
tell us that the object in question does not exist, and consequently
we'll not flag the object as corrupt but as missing.
Fix this issue by bubbling up whether the object is corrupt and, if so,
which packfile contains the corrupted object.
Note that we don't yet need the information about the specific packfile,
so we could've just as well made this a `bool *corrupted` pointer. But
we'll need information about the containing packfile in a subsequent
commit so that we can generate a proper error message telling the user
which packfile contains the broken object.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `reprepare()` callback function can be used to flush caches of a
given object source and then prepare it anew. This is for example used
when a concurrent process may have written new objects. Ultimately, this
can be seen as doing two separate steps:
1. We drop any caches.
2. We prepare the source.
We have one callsite in git-grep(1) though that really only want to do
(2). This is done by reaching into the "files" backend directly and then
calling `odb_source_packed_prepare()`, which of course may not work with
alternate backends.
We could in theory just call `reprepare()` here, and that would likely
not have any significant downside. But this would certainly feel like a
code smell.
Instead, generalize the `reprepare()` callback to `prepare()` with a
flag that optionally instructs the backend to also flush the caches,
which allows us to drop the external `odb_source_packed_prepare()`
declaration.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Our interfaces used to interact with MIDXs all work on top of the
generic `struct odb_source`. This doesn't make much sense though: a MIDX
is strictly tied to the "packed" source, so passing in a generic source
gives the false sense that it may also work with a different type of
source.
Fix this conceptual weirdness and instead require the caller to pass in
a "packed" source explicitly. This also makes the next commit easier to
implement, where we drop the pointer to the "files" source in the
"packed" source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Move the logic to prepare and reprepare the "packed" source into
"odb/source-packed.c" and wire it up as the `reprepare()` callback.
Note that "preparing" a source is not yet generic. Eventually, it would
probably make sense to turn the existing `reprepare()` callback into a
`prepare()` callback with an optional flag to force re-preparing. But
this step will be handled in a separate patch series.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Expose the incremental MIDX repacking mode (implemented in an earlier
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>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
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>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
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 index. 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".
Since the previous commit already builds the array in forward order, the
conversion is straightforward: replace indexed assignments with
`strvec_push()`, drop the pre-counting and `CALLOC_ARRAY()`, and
simplify cleanup via `strvec_clear()`.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
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 `strset`, thus requiring us
to pass an additional length parameter, and makes lookups linear.
Replace the bare array with a `strset` 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>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Further work on incremental repacking using MIDX/bitmap
* tb/incremental-midx-part-3.2:
midx: enable reachability bitmaps during MIDX compaction
midx: implement MIDX compaction
t/helper/test-read-midx.c: plug memory leak when selecting layer
midx-write.c: factor fanout layering from `compute_sorted_entries()`
midx-write.c: enumerate `pack_int_id` values directly
midx-write.c: extract `fill_pack_from_midx()`
midx-write.c: introduce `midx_pack_perm()` helper
midx: do not require packs to be sorted in lexicographic order
midx-write.c: introduce `struct write_midx_opts`
midx-write.c: don't use `pack_perm` when assigning `bitmap_pos`
t/t5319-multi-pack-index.sh: fix copy-and-paste error in t5319.39
git-multi-pack-index(1): align SYNOPSIS with 'git multi-pack-index -h'
git-multi-pack-index(1): remove non-existent incompatibility
builtin/multi-pack-index.c: make '--progress' a common option
midx: introduce `midx_get_checksum_hex()`
midx: rename `get_midx_checksum()` to `midx_get_checksum_hash()`
midx: mark `get_midx_checksum()` arguments as const
|
|
The "files" backend is implemented as a pointer in the `struct
odb_source`. This contradicts our typical pattern for pluggable backends
like we use it for example in the ref store or for object database
streams, where we typically embed the generic base structure in the
specialized implementation. This pattern has a couple of small benefits:
- We avoid an extra allocation.
- We hide implementation details in the generic structure.
- We can easily downcast from a generic backend to the specialized
structure and vice versa because the offsets are known at compile
time.
- It becomes trivial to identify locations where we depend on backend
specific logic because the cast needs to be explicit.
Refactor our "files" object database source to do the same and embed the
`struct odb_source` in the `struct odb_source_files`.
There are still a bunch of sites in our code base where we do have to
access internals of the "files" backend. The intent is that those will
go away over time, but this will certainly take a while. Meanwhile,
provide a `odb_source_files_downcast()` function that can convert a
generic source into a "files" source.
As we only have a single source the downcast succeeds unconditionally
for now. Eventually though the intent is to make the cast `BUG()` in
case the caller requests to downcast a non-"files" backend to a "files"
backend.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Introduce a new "files" object database source. This source encapsulates
access to both loose object files and the packfile store, similar to how
the "files" backend for refs encapsulates access to loose refs and the
packed-refs file.
Note that for now the "files" source is still a direct member of a
`struct odb_source`. This architecture will be reversed in the next
commit so that the files source contains a `struct odb_source`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The MIDX file format currently requires that pack files be identified by
the lexicographic ordering of their names (that is, a pack having a
checksum beginning with "abc" would have a numeric pack_int_id which is
smaller than the same value for a pack beginning with "bcd").
As a result, it is impossible to combine adjacent MIDX layers together
without permuting bits from bitmaps that are in more recent layer(s).
To see why, consider the following example:
| packs | preferred pack
--------+-------------+---------------
MIDX #0 | { X, Y, Z } | Y
MIDX #1 | { A, B, C } | B
MIDX #2 | { D, E, F } | D
, where MIDX #2's base MIDX is MIDX #1, and so on. Suppose that we want
to combine MIDX layers #0 and #1, to create a new layer #0' containing
the packs from both layers. With the original three MIDX layers, objects
are laid out in the bitmap in the order they appear in their source
pack, and the packs themselves are arranged according to the pseudo-pack
order. In this case, that ordering is Y, X, Z, B, A, C.
But recall that the pseudo-pack ordering is defined by the order that
packs appear in the MIDX, with the exception of the preferred pack,
which sorts ahead of all other packs regardless of its position within
the MIDX. In the above example, that means that pack 'Y' could be placed
anywhere (so long as it is designated as preferred), however, all other
packs must be placed in the location listed above.
Because that ordering isn't sorted lexicographically, it is impossible
to compact MIDX layers in the above configuration without permuting the
object-to-bit-position mapping. Changing this mapping would affect all
bitmaps belonging to newer layers, rendering the bitmaps associated with
MIDX #2 unreadable.
One of the goals of MIDX compaction is that we are able to shrink the
length of the MIDX chain *without* invalidating bitmaps that belong to
newer layers, and the lexicographic ordering constraint is at odds with
this goal.
However, packs do not *need* to be lexicographically ordered within the
MIDX. As far as I can gather, the only reason they are sorted lexically
is to make it possible to perform a binary search over the pack names in
a MIDX, necessary to make `midx_contains_pack()`'s performance
logarithmic in the number of packs rather than linear.
Relax this constraint by allowing MIDX writes to proceed with packs that
are not arranged in lexicographic order. `midx_contains_pack()` will
lazily instantiate a `pack_names_sorted` array on the MIDX, which will
be used to implement the binary search over pack names.
This change produces MIDXs which may not be correctly read with external
tools or older versions of Git. Though older versions of Git know how to
gracefully degrade and ignore any MIDX(s) they consider corrupt,
external tools may not be as robust. To avoid unintentionally breaking
any such tools, guard this change behind a version bump in the MIDX's
on-disk format.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When trying to print out, say, the hexadecimal representation of a
MIDX's hash, our code will do something like:
hash_to_hex_algop(midx_get_checksum_hash(m),
m->source->odb->repo->hash_algo);
, which is both cumbersome and repetitive. In fact, all but a handful of
callers to `midx_get_checksum_hash()` do exactly the above. Reduce the
repetitive nature of calling `midx_get_checksum_hash()` by having it
return a pointer into a static buffer containing the above result.
For the handful of callers that do need to compare the raw bytes and
don't want to deal with an encoded copy (e.g., because they are passing
it to hasheq() or similar), they may still rely on
`midx_get_checksum_hash()` which returns the raw bytes.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Since 541204aabea (Documentation: document naming schema for structs and
their functions, 2024-07-30), we have adopted a naming convention for
functions that would prefer a name like, say, `midx_get_checksum()` over
`get_midx_checksum()`.
Adopt this convention throughout the midx.h API. Since this function
returns a raw (that is, non-hex encoded) hash, let's suffix the function
with "_hash()" to make this clear. As a side effect, this prepares us
for the subsequent change which will introduce a "_hex()" variant that
encodes the checksum itself.
Suggested-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
To make clear that the function `get_midx_checksum()` does not do
anything to modify its argument, mark the MIDX pointer as const.
The following commit will rename this function altogether to make clear
that it returns the raw bytes of the checksum, not a hex-encoded copy of
it.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The packfile_store data structure is moved from object store to odb
source.
* ps/packfile-store-in-odb-source:
packfile: move MIDX into packfile store
packfile: refactor `find_pack_entry()` to work on the packfile store
packfile: inline `find_kept_pack_entry()`
packfile: only prepare owning store in `packfile_store_prepare()`
packfile: only prepare owning store in `packfile_store_get_packs()`
packfile: move packfile store into object source
packfile: refactor misleading code when unusing pack windows
packfile: refactor kept-pack cache to work with packfile stores
packfile: pass source to `prepare_pack()`
packfile: create store via its owning source
|
|
The multi-pack index still is tracked as a member of the object database
source, but ultimately the MIDX is always tied to one specific packfile
store.
Move the structure into `struct packfile_store` accordingly. This
ensures that the packfile store now keeps track of all data related to
packfiles.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The packfile store is a member of `struct object_database`, which means
that we have a single store per database. This doesn't really make much
sense though: each source connected to the database has its own set of
packfiles, so there is a conceptual mismatch here. This hasn't really
caused much of a problem in the past, but with the advent of pluggable
object databases this is becoming more of a problem because some of the
sources may not even use packfiles in the first place.
Move the packfile store down by one level from the object database into
the object database source. This ensures that each source now has its
own packfile store, and we can eventually start to abstract it away
entirely so that the caller doesn't even know what kind of store it
uses.
Note that we only need to adjust a relatively small number of callers,
way less than one might expect. This is because most callers are using
`repo_for_each_pack()`, which handles enumeration of all packfiles that
exist in the repository. So for now, none of these callers need to be
adapted. The remaining callers that iterate through the packfiles
directly and that need adjustment are those that are a bit more tangled
with packfiles. These will be adjusted over time.
Note that this patch only moves the packfile store, and there is still a
bunch of functions that seemingly operate on a packfile store but that
end up iterating over all sources. These will be adjusted in subsequent
commits.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Even when there is no changes in the packfile and no need to
recompute bitmaps, "git repack" recomputed and updated the MIDX
file, which has been corrected.
* ps/repack-avoid-noop-midx-rewrite:
midx-write: skip rewriting MIDX with `--stdin-packs` unless needed
midx-write: extract function to test whether MIDX needs updating
midx: fix `BUG()` when getting preferred pack without a reverse index
|
|
The function `midx_preferred_pack()` returns the preferred pack for a
given multi-pack index. To compute the preferred pack we:
1. Take the first position indexed by the MIDX in pseudo-pack order.
2. Convert this pseudo-pack position into the MIDX position.
3. We then look up the pack that corresponds to this MIDX position.
This reliably returns the preferred pack given that all of its contained
objects will be up front in pseudo-pack order.
The second step that turns the pseudo-pack order into MIDX order
requires the reverse index though, which may not exist for example when
the MIDX does not have a bitmap. And in that case one may easily hit a
bug:
BUG: ../pack-revindex.c:491: pack_pos_to_midx: reverse index not yet loaded
In theory, `midx_preferred_pack()` already knows to handle the case
where no reverse index exists, as it calls `load_midx_revindex()` before
calling into `midx_preferred_pack()`. But we only check for negative
return values there, even though the function returns a positive error
code in case the reverse index does not exist.
Fix the issue by testing for a non-zero return value instead, same as
all the other callers of this function already do. While at it, document
the return value of `load_midx_revindex()`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When preparing the packfile store we know to also prepare the MRU list
of packfiles with all packs that are currently loaded in the store via
`packfile_store_prepare_mru()`. So we know that the list of packs in the
MRU list should match the list of packs in the non-MRU list.
But there are some direct or indirect callsites that add a packfile to
the store via `packfile_store_add_pack()` without adding the pack to the
MRU. And while functions that access the MRU (e.g. `find_pack_entry()`)
know to call `packfile_store_prepare()`, which knows to prepare the MRU
via `packfile_store_prepare_mru()`, that operation will be turned into a
no-op because the packfile store is already prepared. So this will not
cause us to add the packfile to the MRU, and consequently we won't be
able to find the packfile in our MRU list.
There are only a handful of callers outside of "packfile.c" that add a
packfile to the store:
- "builtin/fast-import.c" adds multiple packs of imported objects, but
it knows to look up objects via `packfile_store_get_packs()`. This
function does not use the MRU, so we're good.
- "builtin/index-pack.c" adds the indexed pack to the store in case it
needs to perform consistency checks on its objects.
- "http.c" adds the fetched pack to the store so that we can access
its objects.
In all of these cases we actually want to access the contained objects.
And luckily, reading these objects works as expected:
1. We eventually end up in `do_oid_object_info_extended()`.
2. Calling `find_pack_entry()` fails because the MRU list doesn't
contain the newly added packfile.
3. The callers don't pass `OBJECT_INFO_QUICK`, so we end up
repreparing the object database. This will also cause us to
reprepare the MRU list.
4. We now retry reading the object via `find_pack_entry()`, and now we
succeed because the MRU list got populated.
This logic feels quite fragile: we intentionally add the packfile to the
store, but we then ultimately rely on repreparing the entire store only
to make the packfile accessible. While we do the correct thing in
`do_oid_object_info_extended()`, other sites that access the MRU may not
know to reprepare.
But besides being fragile it's also a waste of resources: repreparing
the object database requires us to re-read the alternates file and
discard any caches.
Refactor the code so that we unconditionally add packfiles to the MRU
when adding them to a packfile store. This makes the logic less fragile
and ensures that we don't have to reprepare the store to make the pack
accessible.
Note that this does not allow us to drop `packfile_store_prepare_mru()`
just yet: while the MRU list is already populated with all packs now,
the order in which we add these packs is indeterministic for most of the
part. So by first calling `sort_pack()` on the other packfile list and
then re-preparing the MRU list we inherit its sorting.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Packfiles have two lists associated to them:
- A list that keeps track of packfiles in the order that they were
added to a packfile store.
- A list that keeps track of packfiles in most-recently-used order so
that packfiles that are more likely to contain a specific object are
ordered towards the front.
Both of these lists are hosted by `struct packed_git` itself, So to
identify all packfiles in a repository you simply need to grab the first
packfile and then iterate the `->next` pointers or the MRU list. This
pattern has the problem that all packfiles are part of the same list,
regardless of whether or not they belong to the same object source.
With the upcoming pluggable object database effort this needs to change:
packfiles should be contained by a single object source, and reading an
object from any such packfile should use that source to look up the
object. Consequently, we need to break up the global lists of packfiles
into per-object-source lists.
A first step towards this goal is to move those lists out of `struct
packed_git` and into the packfile store. While the packfile store is
currently sitting on the `struct object_database` level, the intent is
to push it down one level into the `struct odb_source` in a subsequent
patch series.
Introduce a new `struct packfile_list` that is used to manage lists of
packfiles and use it to store the list of most-recently-used packfiles
in `struct packfile_store`. For now, the new list type is only used in a
single spot, but we'll expand its usage in subsequent patches.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `get_multi_pack_index()` function is declared and implemented in the
packfile subsystem, even though it really belongs into the multi-pack
index subsystem. The reason for this is likely that it needs to call
`packfile_store_prepare()`, which is not exposed by the packfile system.
In a subsequent commit we're about to add another caller outside of the
packfile system though, so we'll have to expose the function anyway.
Do so now already and move `get_multi_pack_index()` into the MIDX
subsystem.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
We have a recurring pattern where we essentially perform an upsert of a
packfile in case it isn't yet known by the packfile store. The logic to
do so is non-trivial as we have to reconstruct the packfile's key, check
the map of packfiles, then create the new packfile and finally add it to
the store.
Introduce a new function that does this dance for us. Refactor callsites
to use it.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `install_packed_git()` functions adds a packfile to a specific
object store. Refactor it to accept a packfile store instead of a
repository to clarify its scope.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The object database tracks the list of packfiles in most-recently-used
order, which is mostly used to favor reading from packfiles that contain
most of the objects that we're currently accessing. With the
introduction of the `struct packfile_store` we have a better place to
host this list though.
Move the list accordingly.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The object database tracks a map of packfiles by their respective paths,
which is used to figure out whether a given packfile has already been
loaded. With the introduction of the `struct packfile_store` we have a
better place to host this list though.
Move the map accordingly.
`pack_map_entry_cmp()` isn't used anywhere but in "packfile.c" anymore
after this change, so we convert it to a static function, as well. Note
that we also drop the `inline` hint: the function is used as a callback
function exclusively, and callbacks cannot be inlined.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
With the preceding commits we started to always have the object database
source available when we load, write or access multi-pack indices. With
this in place we can change how MIDX paths are computed so that we don't
have to pass in the combination of a hash algorithm and object directory
anymore, but only the object database source.
Refactor the code accordingly.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Multi-pack indices store some information that is redundant with their
owning source:
- The locality bit that tracks whether the source is the primary
object source or an alternate.
- The object directory path the multi-pack index is located in.
- The pointer to the owning parent directory.
All of this information is already contained in `struct odb_source`. So
now that we always have that struct available when loading a multi-pack
index we have it readily accessible.
Drop the redundant information and instead store a pointer to the object
source.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
To load a multi-pack index the caller is expected to pass both the
repository and the object directory where the multi-pack index is
located. While this works, this layout has a couple of downsides:
- We need to pass in information reduntant with the owning source,
namely its object directory and whether the source is local or not.
- We don't have access to the source when loading the multi-pack
index. If we had that access, we could store a pointer to the owning
source in the MIDX and thus deduplicate some information.
- Multi-pack indices are inherently specific to the object source and
its format. With the goal of pluggable object backends in mind we
will eventually want the backends to own the logic of reading and
writing multi-pack indices. Making the logic work on top of object
sources is a step into that direction.
Refactor loading of multi-pack indices accordingly.
This surfaces one small problem though: git-multi-pack-index(1) and our
MIDX test helper both know to read and write multi-pack-indices located
in a different object directory. This issue is addressed by adding the
user-provided object directory as an in-memory alternate.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
There are a couple of functions that take both a `struct repository` and
a `struct multi_pack_index`. This provides redundant information though
without much benefit given that the multi-pack index already has a
pointer to its owning repository.
Drop the `struct repository` parameter from such functions. While at it,
reorder the list of parameters of `fill_midx_entry()` so that the MIDX
comes first to better align with our coding guidelines.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Object database sources are classified either as:
- Local, which means that the source is the repository's primary
source. This is typically ".git/objects".
- Non-local, which is everything else. Most importantly this includes
alternates and quarantine directories.
This locality is often computed ad-hoc by checking whether a given
object source is the first one. This works, but it is quite roundabout.
Refactor the code so that we store locality when creating the sources in
the first place. This makes it both more accessible and robust.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In the preceding commits we have migrated all users of the linked list
of multi-pack indices to instead use those stored in the object database
sources. Remove those now-unused pointers.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When calling `close_midx()` we not only close the multi-pack index for
one object source, but instead we iterate through the whole linked list
of MIDXs to close all of them. This linked list is about to go away in
favor of using the new per-source pointer to its respective MIDX.
Refactor the function to iterate through sources instead.
Note that after this patch, there's a couple of callsites left that
continue to use `close_midx()` without iterating through all sources.
These are all cases where we don't care about the MIDX from other
sources though, so it's fine to keep them as-is.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Multi-pack indices are tracked via `struct multi_pack_index`. This data
structure is stored as a linked list inside `struct object_database`,
which is the global database that spans across all of the object
sources.
This layout causes two problems:
- Object databases consist of multiple object sources (e.g. one source
per alternate object directory), where each multi-pack index is
specific to one of those sources. Regardless of that though, the
MIDX is not tracked per source, but tracked globally for the whole
object database. This creates a mismatch between the on-disk layout
and how things are organized in the object database subsystems and
makes some parts, like figuring out whether a source has an MIDX,
quite awkward.
- Multi-pack indices are an implementation detail of how efficient
access for packfiles work. As such, they are neither relevant in the
context of loose objects, nor in a potential future where we have
pluggable backends.
Refactor `prepare_multi_pack_index_one()` so that it works on a specific
source, which allows us to easily store a pointer to the multi-pack
index inside of it. For now, this pointer exists next to the existing
linked list we have in the object database. Users will be adjusted in
subsequent patches to instead use the per-source pointers.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `object_directory` structure is used as an access point for a single
object directory like ".git/objects". While the structure isn't yet
fully self-contained, the intent is for it to eventually contain all
information required to access objects in one specific location.
While the name "object directory" is a good fit for now, this will
change over time as we continue with the agenda to make pluggable object
databases a thing. Eventually, objects may not be accessed via any kind
of directory at all anymore, but they could instead be backed by any
kind of durable storage mechanism. While it seems quite far-fetched for
now, it is thinkable that eventually this might even be some form of a
database, for example.
As such, the current name of this structure will become worse over time
as we evolve into the direction of pluggable ODBs. Immediate next steps
will start to carve out proper self-contained object directories, which
requires us to pass in these object directories as parameters. Based on
our modern naming schema this means that those functions should then be
named after their subsystem, which means that we would start to bake the
current name into the codebase more and more.
Let's preempt this by renaming the structure. There have been a couple
alternatives that were discussed:
- `odb_backend` was discarded because it led to the association that
one object database has a single backend, but the model is that one
alternate has one backend. Furthermore, "backend" is more about the
actual backing implementation and less about the high-level concept.
- `odb_alternate` was discarded because it is a bit of a stretch to
also call the main object directory an "alternate".
Instead, pick `odb_source` as the new name. It makes it sufficiently
clear that there can be multiple sources and does not cause confusion
when mixed with the already-existing "alternate" terminology.
In the future, this change allows us to easily introduce for example a
`odb_files_source` and other format-specific implementations.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The multi-pack index acts as a cache across a set of packfiles so that
we can quickly look up which of those packfiles contains a given object.
As such, the multi-pack index naturally needs to be updated every time
one of the packfiles goes away, or otherwise the multi-pack index has
grown stale.
A stale multi-pack index should be handled gracefully by Git though, and
in fact it is: if the indexed pack cannot be found we simply ignore it
and eventually we fall back to doing the object lookup by just iterating
through all packs, even if those aren't indexed.
But while this fallback works, it has one significant downside: we don't
cache the fact that a pack has vanished. This leads to us repeatedly
trying to look up the same pack only to realize that it (still) doesn't
exist.
This issue can be easily demonstrated by creating a repository with a
stale multi-pack index and a couple of objects. We do so by creating a
repository with two packfiles, both of which are indexed by the
multi-pack index, and then repack those two packfiles. Note that we have
to move the multi-pack-index before doing the final repack, as Git knows
to delete it otherwise.
$ git init repo
$ cd repo/
$ git config set maintenance.auto false
$ for i in $(seq 1000); do printf "%d-original" $i >file-$i; done
$ git add .
$ git commit -moriginal
$ git repack -dl
$ for i in $(seq 1000); do printf "%d-modified" $i >file-$i; done
$ git commit -a -mmodified
$ git repack -dl
$ git multi-pack-index write
$ mv .git/objects/pack/multi-pack-index .
$ git repack -Adl
$ mv multi-pack-index .git/objects/pack/
Commands that cause a lot of objects lookups will now repeatedly invoke
`add_packed_git()`, which leads to three failed access(3p) calls as well
as one failed stat(3p) call. The following strace for example is done
for `git log --patch` in the above repository:
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
74.67 0.024693 1 18038 18031 access
25.33 0.008378 1 6045 6017 newfstatat
------ ----------- ----------- --------- --------- ----------------
100.00 0.033071 1 24083 24048 total
Fix the issue by introducing a negative lookup cache for indexed packs.
This cache works by simply storing an invalid pointer for a missing pack
when `prepare_midx_pack()` fails to look up the pack. Most users of the
`packs` array don't need to be adjusted, either, as they all know to
call `prepare_midx_pack()` before accessing the array.
With this change in place we can now see a significantly reduced number
of syscalls:
% time seconds usecs/call calls errors syscall
------ ----------- ----------- --------- --------- ----------------
73.58 0.000323 5 60 28 newfstatat
26.42 0.000116 5 23 16 access
------ ----------- ----------- --------- --------- ----------------
100.00 0.000439 5 83 44 total
Furthermore, this change also results in a speedup:
Benchmark 1: git log --patch (revision = HEAD~)
Time (mean ± σ): 50.4 ms ± 2.5 ms [User: 22.0 ms, System: 24.4 ms]
Range (min … max): 45.4 ms … 54.9 ms 53 runs
Benchmark 2: git log --patch (revision = HEAD)
Time (mean ± σ): 12.7 ms ± 0.4 ms [User: 11.1 ms, System: 1.6 ms]
Range (min … max): 12.4 ms … 15.0 ms 191 runs
Summary
git log --patch (revision = HEAD) ran
3.96 ± 0.22 times faster than git log --patch (revision = HEAD~)
In the end, it should in theory never be necessary to have this negative
lookup cache given that we know to update the multi-pack index together
with repacks. But as the change is quite contained and as the speedup
can be significant as demonstrated above, it does feel sensible to have
the negative lookup cache regardless.
Based-on-patch-by: Jeff King <peff@peff.net>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `git_open_cloexec()` wrapper function provides the ability to open a
file with `O_CLOEXEC` in a platform-agnostic way. This function is
provided by "object-file.c" even though it is not specific to the object
subsystem at all.
Move the file into "compat/open.c". This file already exists before this
commit, but has only been compiled conditionally depending on whether or
not open(3p) may return EINTR. With this change we now unconditionally
compile the object, but wrap `git_open_with_retry()` in an ifdef.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
There are multiple sites in "csum-file.c" where we use the global
`the_repository` variable, either explicitly or implicitly by using
`the_hash_algo`.
Refactor the code to stop using `the_repository` by adapting functions
to receive required data as parameters. Adapt callsites accordingly by
either using `the_repository->hash_algo`, or by using a context-provided
hash algorithm in case the subsystem already got rid of its dependency
on `the_repository`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Stop using `the_repository` in the "progress" subsystem by passing in a
repository when initializing `struct progress`. Furthermore, store a
pointer to the repository in that struct so that we can pass it to the
trace2 API when logging information.
Adjust callers accordingly by using `the_repository`. While there may be
some callers that have a repository available in their context, this
trivial conversion allows for easier verification and bubbles up the use
of `the_repository` by one level.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* ps/build-sign-compare:
t/helper: don't depend on implicit wraparound
scalar: address -Wsign-compare warnings
builtin/patch-id: fix type of `get_one_patchid()`
builtin/blame: fix type of `length` variable when emitting object ID
gpg-interface: address -Wsign-comparison warnings
daemon: fix type of `max_connections`
daemon: fix loops that have mismatching integer types
global: trivial conversions to fix `-Wsign-compare` warnings
pkt-line: fix -Wsign-compare warning on 32 bit platform
csum-file: fix -Wsign-compare warning on 32-bit platform
diff.h: fix index used to loop through unsigned integer
config.mak.dev: drop `-Wno-sign-compare`
global: mark code units that generate warnings with `-Wsign-compare`
compat/win32: fix -Wsign-compare warning in "wWinMain()"
compat/regex: explicitly ignore "-Wsign-compare" warnings
git-compat-util: introduce macros to disable "-Wsign-compare" warnings
|
|
Mark code units that generate warnings with `-Wsign-compare`. This
allows for a structured approach to get rid of all such warnings over
time in a way that can be easily measured.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `MIDX_MIN_SIZE` definition is used to check the midx_size in
`local_multi_pack_index_one`. This definition relies on the
`the_hash_algo` global variable. Inline this and remove the global
variable usage.
With this, remove `USE_THE_REPOSITORY_VARIABLE` usage from `midx.c`.
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The functions `get_split_midx_filename_ext()`, `get_midx_filename()` and
`get_midx_filename_ext()` use `hash_to_hex()` which internally uses the
`the_hash_algo` global variable.
Remove this dependency on global variables by passing down the
`hash_algo` through to the functions mentioned and instead calling
`hash_to_hex_algop()` along with the obtained `hash_algo`.
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `load_multi_pack_index` function in midx uses `the_repository`
variable to access the `repository` struct. Modify the function and its
callee's to send the `repository` field.
This moves usage of `the_repository` to the `test-read-midx.c` file.
While that is not optimal, it is okay, since the upcoming commits will
slowly move the usage of `the_repository` up the layers and remove it
eventually.
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In the `midx.c` file, there are multiple usages of `the_repository` and
`the_hash_algo` within static functions of the file. Some of the usages
can be simply swapped out with the available `repository` struct. While
some of them can be swapped out by passing the repository to the
required functions.
This leaves out only some other usages of `the_repository` and
`the_hash_algo` in the file in non-static functions, which we'll tackle
in upcoming commits.
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The `multi_pack_index` struct represents the MIDX for a repository.
Here, we add a pointer to the repository in this struct, allowing direct
use of the repository variable without relying on the global
`the_repository` struct.
With this addition, we can determine the repository associated with a
`bitmap_index` struct. A `bitmap_index` points to either a `packed_git`
or a `multi_pack_index`, both of which have direct repository
references. To support this, we introduce a static helper function,
`bitmap_repo`, in `pack-bitmap.c`, which retrieves a repository given a
`bitmap_index`.
With this, we clear up all usages of `the_repository` within
`pack-bitmap.c` and also remove the `USE_THE_REPOSITORY_VARIABLE`
definition. Bringing us another step closer to remove all global
variable usage.
Although this change also opens up the potential to clean up `midx.c`,
doing so would require additional refactoring to pass the repository
struct to functions where the MIDX struct is created: a task better
suited for future patches.
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The struct `packed_git` holds information regarding a packed object
file. Let's add the repository variable to this object, to represent the
repository that this packfile belongs to. This helps remove dependency
on the global `the_repository` object in `packfile.c` by simply using
repository information now readily available in the struct.
We do need to consider that a packfile could be part of the alternates
of a repository, but considering that we only have one repository struct
and also that we currently anyways use 'the_repository', we should be
OK with this change.
We also modify `alloc_packed_git` to ensure that the repository is added
to newly created `packed_git` structs. This requires modifying the
function and all its callee to pass the repository object down the
levels.
Helped-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Karthik Nayak <karthik.188@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The main function we use to search a pack index for an object is
find_pack_entry_one(). That function still takes a bare pointer to the
hash, despite the fact that its underlying bsearch_pack() function needs
an object_id struct. And so we end up making an extra copy of the hash
into the struct just to do a lookup.
As it turns out, all callers but one already have such an object_id. So
we can just take a pointer to that struct and use it directly. This
avoids the extra copy and provides a more type-safe interface.
The one exception is get_delta_base() in packfile.c, when we are chasing
a REF_DELTA from inside the pack (and thus we have a pointer directly to
the mmap'd pack memory, not a struct). We can just bump the hashcpy()
from inside find_pack_entry_one() to this one caller that needs it.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
|
