Byron/dua-cli dua-core-v4.2.0

dua-core v4.2.0

on GitHub

New Features

• make walk result buffering configurable per worker

  Replace the fixed QUEUED_BATCHES_PER_WORKER constant with
  Options::queued_batches_per_worker and a chainable builder method. Use
  two queued batches per worker by default in both production and tests,
  preserving the original bounded buffering behavior.

  Consumers that pause to process entries can opt into a larger queue to
  keep traversal workers busy, trading memory for less idle time. A value
  of zero selects an unbuffered channel. Check the worker-count
  multiplication so excessive capacities cannot silently overflow.

  Update existing Options literals to inherit the new default. Verify the
  actual queue capacity with one and two workers, and exercise larger and
  unbuffered queues through the public single-root and multi-root walkers.

Bug Fixes

• pass strict workspace Clippy on Rust 1.99

  Replace empty-container assertions with equality comparisons and empty
  string checks with inequality comparisons. This satisfies Clippy's
  assert_is_empty lint while showing actual values when assertions fail.

  Locally allow the deprecated fetch_update alias in directory identifier
  allocation. Its replacement, try_update, requires Rust 1.95, while
  dua-core supports Rust 1.88. Preserve the existing atomic update and
  overflow handling without raising the minimum supported Rust version.

• keep verbatim paths verbatim when resolving relative paths on Windows
  When the current directory is verbatim (e.g. after dua "\\?\D:\dir"
  chdirs into a verbatim input path, or when dua runs from a verbatim cwd),
  absolute_verbatim_path resolved relative paths like "." into an
  already-verbatim absolute path, then fed it into the generic \\ (UNC)
  branch and produced \\?\UNC\?\D:\dir. CreateFileW on that path fails
  with ERROR_BAD_NETPATH (os error 53), so dua/dua aggregate could not
  enumerate \\?\ directories at all.

  An absolute path that already carries a \\?\ or \??\ prefix is now
  passed through untouched, matching Win32 verbatim semantics; \\.\,
  UNC and ordinary local paths are unchanged.

  Fixes #406

Performance

• let walk workers queue more batches ahead of the iterator
  The result channel held threads * 2 batches. When the iterator's
  consumer pauses, each worker that finishes a batch then blocks until
  the consumer drains the channel, so the walk stops whenever the
  consumer is busy. A consumer that does real work per entry, such as
  parsing some of the files it's handed, spends much of its time waiting
  on workers that sat idle while it worked.

  Allow 128 batches per worker instead. The channel stays bounded, and
  dua aggregate is no slower. The unit tests that block a worker on a
  full channel fill two batches to do it, so they keep that bound.

Commit Statistics

• 6 commits contributed to the release.
• 31 days passed between releases.
• 4 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details