How scanning works

Parallel scanning

oops uses rayon for parallel directory traversal. When scanning a directory’s children, each child is sized in parallel across available CPU cores.

For a directory with 100 children, all 100 size calculations happen concurrently. Within each child directory, the recursive traversal also parallelizes at each level. This makes scanning large directories (like your home directory) significantly faster than serial tools.

The scan pipeline

Every command follows the same pattern:

1. Read the directory — fs::read_dir() collects immediate children
2. Size each child in parallel — rayon maps over children, recursively summing block sizes
3. Sort by size — largest first
4. Render — pass the sorted entries to the ui module

The ui module owns all rendering. Commands never write to the terminal directly — they compute data and hand it off. This keeps commands thin and testable.

Block-level sizing

Every file size is computed via stat.blocks * 512, not metadata.len(). This gives you the actual on-disk footprint:

metadata.len()   → logical/apparent size (what the file "says" it is)
stat.blocks * 512 → physical/allocated size (what the disk actually stores)

For regular files these are nearly identical. For sparse files, compressed files, or files with holes, they can differ dramatically.

Volume detection

oops volumes parses df -Pk output (POSIX mode, 1K blocks). This gives a predictable 6-column layout across macOS and Linux, avoiding the extra inode columns that macOS df normally includes.

Pseudo-filesystems (devfs, map auto_home, none) and zero-size mounts are filtered out.

Waste detection

oops sweep walks the directory tree looking for well-known patterns:

• Name matches: node_modules, .git, target/ (with Cargo.toml), venv/, __pycache__
• Platform caches: ~/Library/Caches, Xcode DerivedData, .cargo/registry, Docker containers
• Size thresholds: 1 MB for general waste, 100 MB for .git, 50 MB for platform caches, 10 MB for log files

When a waste directory is found, the scan does not recurse into it — the directory is sized as a unit and added to the results.