Struct PackReader 

pub struct PackReader<'a> { /* private fields */ }

Implementations

impl PackReader<'static>

pub fn open(pack_path: &Path, index_path: &Path) -> Result<Self>

Open a pack file. mmap-backed when the pack is large enough to benefit (the same threshold the loose-blob path uses for its own mmap decision); read-into-heap otherwise.

pub fn from_bytes( pack_data: impl Into<Bytes>, index_data: impl AsRef<[u8]>, ) -> Result<Self>

impl<'a> PackReader<'a>

pub fn from_slice( pack_data: &'a [u8], index_data: impl AsRef<[u8]>, ) -> Result<Self>

pub fn list_ids(&self) -> Vec<PackObjectId>

List all object ids in this pack.

pub fn list_hashes(&self) -> Vec<ContentHash>

pub fn has_object(&self, id: &PackObjectId) -> bool

pub fn get_object( &self, id: &PackObjectId, ) -> Result<Option<(ObjectType, Vec<u8>)>>

Get an object from the pack.

Verifies that the tagged id at the indexed offset matches id before returning. A stale .idx file (e.g., overwritten in place after a pack rebuild) can otherwise route a request for hash A to a record physically located at hash B’s offset — same shape, different content, no error signal. This cheap 32-byte id comparison catches that without paying a full content-hash recompute on every read; corruption strictly inside the record body is a separate failure mode surfaced via the consumer-side hash verify (see FsStore::loose_blob_path for the blob equivalent).

pub fn get_hashed_object( &self, hash: &ContentHash, ) -> Result<Option<(ObjectType, Vec<u8>)>>

pub fn get_object_bytes( &self, id: &PackObjectId, ) -> Result<Option<(ObjectType, Bytes)>>

Zero-copy fast path: when the entry is non-delta and stored uncompressed, returns Bytes::slice into the pack’s (mmap-backed) buffer — no allocation, no memcpy. Compressed or delta entries fall back to get_object and wrap the resulting Vec<u8> in a Bytes (one Arc, no body copy).

Use this from the hot read path. The 10 MB benchmark gap between the mount and vanilla FS at the 1 MB+ tier is the per-blob memcpy this method eliminates.

pub fn get_hashed_object_bytes( &self, hash: &ContentHash, ) -> Result<Option<(ObjectType, Bytes)>>

pub fn get_hashed_object_size(&self, hash: &ContentHash) -> Result<Option<u64>>

Read just the type+size header for an object without decompressing its payload. Returns Ok(None) when the object isn’t in this pack.

For non-delta entries this is one varint decode at the indexed offset — much cheaper than get_object. Delta entries fall back to a full read because their resolved size requires chasing the base; in practice deltas are rare in the directory listing hot path so the fallback is acceptable.