byteview
January 2, 2026 ยท View on GitHub
An immutable byte slice that may be inlined, and can be partially cloned without heap allocation.
Think of it as a specialized Arc<[u8]> that can be inlined (skip allocation for small values) and no weak count.
byteview was originally designed to speed up deserialization in lsm-tree, allow inlining of small values and reduce memory usage compared to Arc'd slices.
Values with a known length can be constructed 2-2.5x faster than using Arc<[u8]>:
Sponsors
Memory usage
Allocating 200M "" (len=0) strings:
| Struct | Memory Usage |
|---|---|
Arc<[u8]> | 9.6 GB |
tokio::Bytes | 6.4 GB |
ByteView | 4.8 GB |
Allocating 200M "helloworld" (len=10) strings:
| Struct | Memory Usage |
|---|---|
Arc<[u8]> | 12.8 GB |
tokio::Bytes | 12.8 GB |
ByteView | 4.8 GB |
Allocating 100M "helloworldhelloworld" (len=20) strings:
| Struct | Memory Usage |
|---|---|
Arc<[u8]> | 6.4 GB |
tokio::Bytes | 6.4 GB |
ByteView | 2.4 GB |
Allocating 50M "helloworldhelloworldhelloworldhelloworld" (len=30) strings:
| Struct | Memory Usage |
|---|---|
Arc<[u8]> | 4.0 GB |
tokio::Bytes | 4.0 GB |
ByteView | 3.6 GB |
Allocating 500k "helloworld".repeat(1000) (len=10'000) strings:
| Struct | Memory Usage |
|---|---|
Arc<[u8]> | 5 GB |
tokio::Bytes | 5 GB |
ByteView | 5 GB |
Run fuzz tests
cargo +nightly fuzz run fuzz_target_1