kache-fs

Filesystem measurements, file identity, native reflinks and writable copies. Applications choose where files belong and when sharing an inode is acceptable.

measure_file distinguishes logical length, allocation and private extents. Private allocation remains None when unavailable. probe_for selects a platform probe for directory walks; an inode ledger accounts for hardlinks. APFS uses attribute queries, including bulk directory reads. Linux uses a bounded FIEMAP walk and discards incomplete or ambiguous extent accounting.

try_reflink creates an independent file that shares blocks where supported. It refuses an existing destination and returns errors to the caller for fallback selection. copy_writable copies bytes with an explicit writable mode. Hardlinks use std::fs::hard_link; their content and metadata remain shared.

The staging feature adds StagedFile for writable temporary files. Callers write through its writer, then choose create-only publication or replacement. Unpublished stages use tempfile's cleanup on drop. Creation uses ordinary writable-file permissions, including the process umask on Unix. Callers retain destination checks and any flush requirements; publication does not flush file contents or the parent directory.

The default feature set is empty. serde enables measurement serialization; testing exposes filesystem fixtures for consumers. tempfile is required only with staging. The crate has no runtime, cache index or cleanup policy.