Crate arrow_graph_git 

arrow-git — Graph-native git primitives for Arrow-based knowledge graphs.

Provides 8 git primitives operating on Arrow RecordBatches, designed for versioning RDF-like knowledge graphs stored in arrow-graph-core.

Primitives

1. Object Store — the live GitObjectStore wrapping ArrowGraphStore
2. Commit — snapshot state to Parquet + record in CommitsTable
3. Checkout — restore state from a commit’s Parquet snapshot
4. History DAG — traverse parent_ids for log and ancestors
5. Branch/Head (Refs) — lightweight branch pointers via RefsTable
6. Diff — object-level comparison between commits via diff()
7. Merge — 3-way merge with conflict detection/resolution via merge()
8. Save — crash-safe persistence without creating a commit via save()

Performance (10K triples)

Operation	Measured	Gate
Commit	~3ms	< 25ms
Checkout	~1.3ms	< 25ms
Commit+Checkout (H-GIT-1)	~4.3ms	< 50ms
Save+Restore (M-SAVE)	~4.3ms	< 100ms
Awakening (M-119)	~1.3ms	< 200ms
Batch add 10K	~4.3ms	< 10ms

Merge Conflict Resolution

The merge() function detects conflicts (same subject+predicate with different objects across branches). Use merge_with_strategy for automatic resolution via MergeStrategy::Ours, MergeStrategy::Theirs, MergeStrategy::LastWriterWins, or a custom closure.

Crash-Safe Persistence

save() uses a WAL marker + atomic file rename pattern. If a save is interrupted, the previous Parquet files remain valid. save_with_options adds zstd compression and incremental saves (only dirty namespaces).

Re-exports

pub use blame::BlameEntry;

pub use blame::blame;

pub use checkout::checkout;

pub use cherry_pick::cherry_pick;

pub use commit::Commit;

pub use commit::CommitsTable;

pub use commit::create_commit;

pub use diff::DiffEntry;

pub use diff::DiffResult;

pub use diff::diff;

pub use diff::diff_nondestructive;

pub use history::ancestors;

pub use history::find_common_ancestor;

pub use history::log;

pub use merge::Conflict;

pub use merge::MergeResult;

pub use merge::MergeStrategy;

pub use merge::Resolution;

pub use merge::merge;

pub use merge::merge_with_strategy;

pub use object_store::GitConfig;

pub use object_store::GitObjectStore;

pub use rebase::RebaseResult;

pub use rebase::rebase;

pub use refs::Ref;

pub use refs::RefType;

pub use refs::RefsTable;

pub use remote::RemoteError;

pub use remote::Snapshot;

pub use remote::bytes_to_snapshot;

pub use remote::restore_snapshot;

pub use remote::snapshot_state;

pub use remote::snapshot_to_bytes;

pub use revert::revert;

pub use save::SaveMetrics;

pub use save::SaveOptions;

pub use save::persist_commits;

pub use save::restore;

pub use save::restore_commits;

pub use save::restore_full;

pub use save::restore_named_batches;

pub use save::save;

pub use save::save_full;

pub use save::save_named_batches;

pub use save::save_with_options;

Modules

blame

Blame — per-triple provenance history.

checkout

Checkout — load a Parquet snapshot back into the ArrowGraphStore.

cherry_pick

Cherry-pick — apply a specific commit’s changes to the current branch.

commit

Commit — snapshot current graph state to Parquet + record in CommitsTable.

compat

Compatibility types for Namespace and Layer.

diff

Diff — object-level comparison between two commits.

history

History DAG — traversal of commit parent_ids to form a directed acyclic graph.

merge

Merge — 3-way merge with conflict detection.

object_store

Object Store — the live in-memory Arrow tables from arrow-graph-core.

rebase

Rebase — replay a sequence of commits onto a new base.

refs

Refs — branch and HEAD management.

remote

Remote — Push/Pull Graph State Over Any Transport

revert

Revert — create a new commit that undoes the changes from a target commit.

save

Save — persist current state without creating a commit (crash recovery).