NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
crates.io, #1394 most downloaded on crates.io
Implements git packs and related data structures
Latest stable
0.77.0
Last release 26 days ago
22 Oct 2026
Ships on a steady schedule
a new release about every 4 weeks
Nearly every release is documented
notes for 60 of 60 stable releases
1 version withdrawn
withdrawn after publishing
4 years old
61 releases · first in 2023
add wasm feature toggle to let parts of git-pack build on wasm32. It's a breaking change because we also start using the dep: syntax for declaring ref…
<csr-id-ebc7f47708a63c3df4415ba0e702660d976dfb3e/> <csr-id-2290d006705ff47ad780b009fe58ee422b3285af/> <csr-id-598698b88c194bc0e6ef69539f9fa7246ebfab70/> <csr-id-b46347fd3d50886eeca500e31e1e12b354711309/> <csr-id-591afd56d9862a6348ef8b3af61798004b36aa19/> <csr-id-2f2d856efe733d3cf81110c0e0607d2e7c40d968/> <csr-id-9b9f10ad862b5e097c836c51df1eb98607df5ae1/> <csr-id-e0b8636f96e4bfe1bc72b5aa6ad4c4c8538ff92c/> <csr-id-71c628d46088ab455b54eb2330d24dcff96c911d/> <csr-id-8fe461281842b58aa11437445637c6e587bedd63/> <csr-id-e6ff1a885889cf88f6b34b1193aa03d8bce16af5/> <csr-id-f48630ba8f745c2ec61a1e3c51fa63a1789a088c/> <csr-id-f7f136dbe4f86e7dee1d54835c420ec07c96cd78/> <csr-id-25209454d3f7e27e12e8ddca92e43b1ff01d58aa/> <csr-id-c800fdd331e6d7a0b8d756ba822915259f26e9e8/>
<csr-id-6c4c196c9bc6c2171dc4dc58b69bd5ef53226e29/> add wasm feature toggle to let parts of git-pack build on wasm32.
It's a breaking change because we also start using the dep: syntax for declaring
references to optional dependencies, which will prevent them from being automatically
available as features.
Besides that, it adds the wasm feature toggle to allow compiling to wasm32 targets.
<csr-id-95210cb2ba85f75148b4ef48ccea9d9f8a0a0114/> Provide optional candidates for ambigious entries during lookup_prefix()
The candidate entries are all entries matching a given prefix.
<csr-id-37f3a675d7fa931f5d3f38e91df74ec0b517422b/> move data::ResolveBase into data::decode_entry::; unify decode_entry|decode_header::Error into decode::Error.
<csr-id-99905bacace8aed42b16d43f0f04cae996cb971c/> upgrade bstr to 1.0.1
<csr-id-8c5ae77f06a64c57df9a9ad1190266896a223dbe/> Remove deprecated compound and linked object databases The dynamic/general store is the only maintained can-do-it-all DB now.
<csr-id-a79a7fb638b45df88af0d0d5fc9ada6d824bc328/> Improve method signatures of cache::Tree::*
<csr-id-91d047658b114f372735116c9d8e6962a3873137/> cleanup and unify verify_integrity() method signature
Previously they used many different ways of handling their parameters
despite all boiling down to calling the same 'index::File::traverse()`
method.
This allows for more reuse of Options structs and generally makes
clearer how these optinos are used.
<csr-id-2cf7727228e1d8094ffd2eec6746006348c39eab/> index::File::traverse() now returns an Outcome struct instead of tuple of 3 fields
<csr-id-bf04644ab75ed1969507f957dc8d4868790d462d/> remove Option<impl Progress> in favor of impl Progress
<csr-id-6829e5e5d6aed1e6c87647144e2dd76a1e4b9f1f/> multi-index integrity check; use integrity::Outcome for various integrity checks
<csr-id-d851bede97801096d188ff6af06c98a79fe276db/> remove unnecessary Arc around should_interrupt flag
<csr-id-c2679a03358b9c19d63ed1af1cd57324c6381447/> remove Sha1 mentions in index::verify::Mode::* variants
The hash is repository defined and not hard-coded
<csr-id-80b120d3278e46429f848df7af3db13413c36649/> introduce index::File::verify_integrity(…, pack: Option<PackContext>, …), replacing tuple
This allows for more documentation on what input is required there and
generally makes for an easier to use API.
<csr-id-79dc0d5ba6fa31ddd5c075693ffdc6496c1eaded/> rename oid::try_from() to try_from_bytes(), add from_bytes_unchecked()
This change was done in the name of consistency, as from_bytes() is
used in many other git-* crates
<csr-id-2ef9a8424af51310db8c1e6df31dde9953ed3d21/> Change accessors named hash_kind() to object_hash() for consistency
<csr-id-b76f6be6c5baa6cf613a174241f007e92bf5ba36/> consistently use object_hash instead of hash_kind
<csr-id-629412b4cb192614b7eff08dbf203e3448c902c1/> data::Entry::from_read() now takes a hash lengths as parameter That way ref-deltas can be interepreted without hard-coding SHA1
<csr-id-851dc2c52fa8e204ba2d5ced8fb0959a889869d8/> data::Entry::from_bytes(…, hash_len) takes new parameter The hash-len tells it how to interpret ref-delta objects, which store the complete hash of the base object.
This is now entirely configurable.
<csr-id-db8c8c41b4ced0fc296d3877883d801e77d550ae/> index::File::at() with git_hash::Kind parameter
It will allow to assume different hashes even in the index file format
which isn't yet capable of storing this information.
<csr-id-e6a3c9f72332b524b143bc94ee9df0a6db11e864/> data::File::at() and Bundle::at() now have hash_kind parameter
It's used to configure the kind of hash to assume when reading packs and
indices.
<csr-id-82b9b33bd5f4c3c1721a5093de2cedc62cb10565/> move bundle::Location to data::entry::Location
The latter place best describes its purpose.
<csr-id-3f05fea55dc8acce1ed62ecbe4e0a1394f2720b7/> remove make_object_cache parameter from git_pack::data::output::count::objects()
It now is an implementation detail of the Find trait.
<csr-id-e7526b2a7b51cbac4018e1ab3b623a85987fadc2/> parallel utilities now use Send + Clone insted of Send + Sync
This helps to assure that thread-local computations always work with the
kind of types we provide. The ones that are carrying out actions are
notably not Sync anymore.
We cater to that by defining our bounds accordingly, but for those
who want to use other utilities that need Sync, using types like
Repository and thread_local!() is the only way to make this
work.
<csr-id-e8b091943f0c9a26317da0003f7fcdf5a56ef21a/> Rename gix->ein and gixp->gix
<csr-id-ec58bef84d620360dc52e34e173ace1310f74c85/> reduce memory usage for pack resolution by 18%… …compared to where it started out before.
This is the result of using u32 intead of usize where possible, leveraging the object limit of u32 in packs.
This change makes the biggest difference by not keeping the decompressed memory of leaf nodes alive for longer than needed, at the cost of some code duplication which could be extracted into a function if one was inclined to deal with the boilerplate and lots of generics.
<csr-id-49d168a8859c6b5a9e7ef58cd836093212c2c7ad/> Allow verification of empty packs and indices. Empty packs are fine, even though the implementation should probably assure to not write them, or remove them after having been written.
<csr-id-a745512185fb0a46e35daaa6d28829aec05edb55/> increase pack-receive performance using a BufWriter Previously the NamedTempFile would receive every little write request for millions of objects, consuming considerable amounts of time.
Now a buf writer alleviates this issue entirely.
<csr-id-0b6ed60f842f0a36f61f187651080540a358758e/> bundle::write::Error is now publicly available
<csr-id-d9451e8d7fc39c252042f9d2447061262c16ae7a/> downgrade dashmap to 4.0 to avoid unsoundness. See https://github.com/xacrimon/dashmap/issues/167 for tracking progress on resolving the issue.
<csr-id-42e0487286c1f745837c0ce337ed7c9d86b14516/> support Rust 1.52
<csr-id-84ade1d23060f10bf6c8529f8f693d06660b4f4e/> Allow resolution of in-pack ref-deltas This finally allows delta tree caches to be used on typical small packs returned by GitHub.
<csr-id-ba92cc09ba41fe4c9a9097bfeb8d18016408fcdf/> don't try to short-cut detection of large offsets when writing index files The code incorrectly assumed that the input is sorted by offsets, with the largest offset being last, even though by all means that's not the case.
<csr-id-6d3f52dc13d7243a6bce6dab89a985114a75d94b/> Avoid the dashmap being cloned for each thread Intead, share it by reference, it's sync after all.
This issue was introduced when switching to a Send + Clone model,
instead of Send + Sync, to allow thread-local caches in database
handles of all kinds.
<csr-id-b605c1fa0494b10872d3c2e6ecce0e39f1a90a9e/> linked::Store now assures unique IDs across compound stores
<csr-id-20b3994206aa5bc5e35cbbc9c8f8f99187077f79/> Adjust size-hints of resolving entries iterator and use the upper bound in delta tree The delta-tree is a data structure that actually heavily relies on favorable allocation and a known amount of objects in order to provide front and back buffers. However, this is an implementation detail and they don't have to stay consistent at all especially after growing the buffer by pushing to it.
Interestingly, the VecDeque internally over-allocates as well which
definitely helps the example of as_mut_slices(), otherwise
it could also suffer from the assertions that trigger here.
<csr-id-d8fe8141e80a9e9a433b5e1a072b850325c806c8/> don't put more objects into the pack cache than needed Previously when accessing a packed object, it would store the base object into the pack cache (if it wasn't retrieved from there) which is great if that operation is free.
Since it isn't, it's better not to stress the cache with puts and trash more objects than necessary.
Now only the last decompressed object will be put into the LRU cache.
<csr-id-faf6f813927720c5adf62102f9ce46606ff2617c/> don't include submodules in count… …to avoid dealing with missing objects.
It's still a good idea to handle these gracefully though, git itself seems to ignore them.
<csr-id-d792ea543246632bf1ca8d0e1d239bbe7f07e219/> use enumerations to advertise progress ids publicly. Previously these were an implementation detail which also means they couldn't be relied upon.
Thanks to an intermediate enumeration, they become part of the public API and their actual value is not exposed.
<csr-id-f0e40ecddaf1211f76ed60ef30cf03dcfd53a7ab/> add wasm feature toggle to allow compilation to wasm32-unknown-unknown
<csr-id-a85dcddba29a453bbea87ac865b8aedc97f41aed/> add data::File::resolve_header() to obtain object information without decompressing it.
<csr-id-c8835c6edae784c9ffcb69a674c0a6545dbb2af3/> upgrade to prodash 21.1 and add Ids to all progress instances.
That way callers can identify progress they are interested in, say, for
selective visualizations.
<csr-id-450257eb596465352fb363a5c8c514f544e7f9ac/> Bundle::write… also produces a .keep file
These files are placed before moving the coresponding pack and index
files into place to prevent them from being picked up for collection
while their refs are still being placed.
The caller is responsible for deleting them once refs integrate the contained objects into the commit graph.
<csr-id-6836cb148cbadf919bdac2e0e77bdec34caa9307/> more efficient distribution of tasks to threads during pack resolution. This clearly is a peak-performance optimization as it will now efficiently distribut tree-root nodes to threads one at a time by means of shared memory, which is accessed mutably yet safely.
This change can also further reduce peak memory usage as it will not keep buffers of more resolved deltas then necessary, as it only handles one at a time per thread.
<csr-id-b1c40b0364ef092cd52d03b34f491b254816b18d/> use docsrs feature in code to show what is feature-gated automatically on docs.rs
<csr-id-517677147f1c17304c62cf97a1dd09f232ebf5db/> pass --cfg docsrs when compiling for https://docs.rs
<csr-id-503b1a1f8d4f39b44c166209d7a8ba8d74137859/> index::File::lookup_prefix(…)
<csr-id-cb83beedd1aa389f6774e2296f79273e8c8f14f4/> git-hash::Prefix::from_id() A way to obtain a prefix of an object id, with all non-prefix bytes set to zero.
<csr-id-16208306ab49ade30d8ffd6b067ebd8eefd84cd4/> in-manifest and in-lib documentation of feature toggles
<csr-id-b80dec2323b81fb2172df76c7d897a4b5e6bdfea/> zero-objects check for index and multi-index integrity validation
<csr-id-56fc99fb9c1cab61abd03c10e1b4af0d6e491bbf/> support for fan-checking in index and multi-index integrity verification
<csr-id-28e3ea8612112f6a04cfaff591565eca5a1ffba2/> introduce type for entry indices within an index or multi-index That way it's a littl emore descriptive than a bare u32.
<csr-id-58c2edb76755ab71e10eef4cd9a51533825c291f/> git_pack::Find::try_find_cached(…, pack_cache) With this method it's easier to bypass local caches and control the cache oneself entirely.
<csr-id-e25f4eadec679406aad6df10026e27e4832c2482/> A simplified version of the Find trait
It's meant for the next generation of object db handles which keep a
local cache of all the details of the actual object database.
<csr-id-60c9fad8002b4e3f6b9607bba6361871752f4d3d/> control pack and object cache size in megabytes
<csr-id-50cf610e8939812c3d2268c48835e2dac67d0c31/> cache::Object trait for caching and retrieving whole objects
<csr-id-5a8c2da6cb1e2accf7cfdccc16bc3a1d0b2a7dbc/> object cache size is configurable
<csr-id-d6c44e6ab8f436020d4fb235e423b018fd1e7a9f/> dynamically sized full-object speeds up diff-based object counting… …which is what happens when counting objects for fetches where only changed objects should be sent.
<csr-read-only-do-not-edit/>
<csr-read-only-do-not-edit/>
Clippy helped 26 times to make code idiomatic.
<csr-read-only-do-not-edit/>
<details><summary>view details</summary>
2f2d856)52b38bc)e695eda)0c81769)a517e39)4eebaac)4a9a05f)f536bad)a0bc98c)d8fe814)2f75db2)e1861ca)58ab2ae)c857d61)6b5c394)9d6f263)719b6bd)2f43a54)e8b0919)b605c1f)c800fdd)52d4fe5)b88f253)52a4dcd)dde5c6b)8ad25c5)ad36fb9)62d3f10)22d35bd)7eff6bf)bundle::Location to data::entry::Location (82b9b33)ed0c266)12db899)make_object_cache parameter from git_pack::data::output::count::objects() (3f05fea)f48630b)5fa1a9d)58c2edb)3310d8f)Find::try_find(…) (ebc7f47)1bb4253)2290d00)Find trait (e25f4ea)dfdd6fb)7517482)91e6d38)f223ecb)235a27a)2e16f13)4f6b030)5dc1f81)4137327)1bea1d4)542c0df)50fde01)f7d5c7f)bfc8069)565a7ae)6a68ed7)93dc660)2fc6751)5457761)9ce1e7f)index::File::traverse() now returns an Outcome struct instead of tuple of 3 fields (2cf7727)c361ee3)3a76a28)b80dec2)56fc99f)8f9a55b)d2bea27)e3dfa12)bb35c69)Option<impl Progress> in favor of impl Progress (bf04644)integrity::Outcome for various integrity checks (6829e5e)Arc around should_interrupt flag (d851bed)index::verify::Mode::* variants (c2679a0)index::File::verify_integrity(…, pack: Option<PackContext>, …), replacing tuple (80b120d)853d468)ce044ef)28e3ea8)data::Id typedef (e0b8636)710780c)10a24c1)73fbc91)eafdff4)1c99903)c2a6918)254f618)a54f552)oid::try_from() to try_from_bytes(), add from_bytes_unchecked() (79dc0d5)hash_kind() to object_hash() for consistency (2ef9a84)754a663)Default implementation for user of Tree::traverse (9da20e9)d548f72)9bf25cc)object_hash instead of hash_kind (b76f6be)ca35246)0cae25b)629412b)851dc2c)7331e99)07aa1bc)e6ff1a8)index::File::at() with git_hash::Kind parameter (db8c8c4)data::File::at() and Bundle::at() now have hash_kind parameter (e6a3c9f)b71ea6a)82fec95)bf62067)69f1d2a)8b8b4c5)8c9c7fc)44ea5c3)ac46765)743d696)a363de9)6ae49e3)e7e40c3)24a9790)3ca04e3)99023bb)fac8efa)b2d2ae2)edf02ae)1a2a049)5e085ec)4b4f9f8)47860b7)831397c)84ade1d)38426a1)7db7195)80a4a7a)6c06659)cache::Tree::* (a79a7fb)verify_integrity() method signature (91d0476)7b2ab26)e3977fe)7721b5f)35bdee4)7558844)52e3c6f)4122306)memmap2 instead of filebuffer (d9011c7) ([d71bd9d`](https://github.com/GitoxideLabs/gitoxide/commit/d71bd9ded1e5e5a61a27be3d55f4b85ee4049bcf))8240622)index::File::lookup_prefix(…) (503b1a1)d09fd9b)927b2ac)b558f11)69cb6d1)cff6f9f)5bc548e)75efa79)cb83bee)467453a)814de07)f95c1a0)84cb256)cc2d810)git-path crate instead of git_features::path (47e607d)7e76796)f86eacc)e5f6943)bba4c68)de2d587)4b320e7)c28404b)5ffe54f)4db4754)8a63076)a76a0e1)0340896)candidates for ambigious entries during lookup_prefix() (95210cb)a2f9fa4)b593806)prodash v21 (a0655dc)pack- (e489b10)a745512)index::write::Outcome::index_kind -> ::index_version. (b46347f)72ce7fd)bundle::write::Error is now publicly available (0b6ed60)bundle::write::Options::index_kind -> ::index_version. (591afd5)bstr to 1.0.1 (99905ba)caa7a1b)3877920)6d3f52d)bcb3d37)f9232ac)70179e2)faf6f81)60c9fad)71c628d)8fe4612)50cf610)5a8c2da)d6c44e6)80c6994)2664d73)rust-version to 1.64 (55066ce)8c0bca3)59e9fac)48f5bd2)a5869e0)41d57b9)e313112)6efd0d3)6ccc88a)c9275b9)git-testtools to gix-testtools (b65c33d)git-pack to gix-pack (1ee81ad)git-pack to gix-pack (134d70d)git-odb to gix-odb (476e2ad)git-index to gix-index (86db5e0)git-diff to gix-diff (49a163e)git-commitgraph to gix-commitgraph (f1dd0a3)git-mailmap to gix-mailmap (2e28c56)git-discover to gix-discover (53adfe1)git-lfs to gix-lfs (b9225c8)git-chunk to gix-chunk (59194e3)git-bitmap to gix-bitmap (75f2a07)git-protocol to gix-protocol (823795a)git-refspec to gix-refspec (c958802)git-revision to gix-revision (ee0ee84)git-transport to gix-transport (b2ccf71)git-credentials to gix-credentials (6b18abc)git-prompt to gix-prompt (6a4654e)git-command to gix-command (d26b8e0)git-packetline to gix-packetline (5cbd22c)git-worktree to gix-worktree (73a1282)git-hashtable to gix-hashtable (26a0c98)git-worktree to gix-worktree (108bb1a)git-url to `giYour coding agent can read these notes before it upgrades. Set up the MCP server →
One column per quarter.