duct

Changes since 0.13.7:

• The API is stabilized as-is, and most callers won't need to make any
  changes.
• Fixed a bug where Handle::try_wait could block indefinitely if a
  child process had exited but a longer-lived grandchild process was
  keeping its IO pipes open and its IO threads running (i.e.
  stdin_bytes, stdout_capture, or stderr_capture). Commit 04c2c8c.
• Handle now does lazy cleanup of leaked zombie children. If a child
  has not exited by the time its Handle is dropped, that child gets
  added to a global list. Whenever any Duct function is about to spawn a
  new child, it polls each child in the leaked list and cleans up any
  that have exited. This strategy is copied from Python's subprocess
  module. ReaderHandle contains a Handle internally and does the
  same cleanup.
• With that new cleanup mechanism in place, Handle::kill and
  ReaderHandle::kill no longer wait on child processes to exit. This
  avoids obscure edge cases (FUSE filesystems, debugger pauses) where
  waiting after SIGKILL could still block indefinitely.
• Duct no longer kills your child processes in any circumstance other
  than when you call .kill(). Previously Duct would kill child
  processes when dropping a ReaderHandle, or in rare pipe spawning
  error cases.

Changes since 0.13.6:

• Avoid unnecessarily dup'ing standard file descriptors. This unbreaks
  some use cases where those descriptors might not exist at all, like
  Windows services. See #99.

Changes since 0.13.5:

• Dependency updates only.

Changes since 0.13.4:

• Doc changes only.

Changes since 0.13.3:

• The duct_sh crate is now versioned in lockstep with duct.
• Doc changes.

Changes since 0.13.2:

• Added Handle::pids and ReaderHandle::pids.

Changes since 0.13.1:

• Added ReaderHandle::try_wait.
• Debug output for Expression is more concise.

Changes since 0.13.0:

• ReaderHandle now supports kill().
• Kill methods no longer wait on IO threads to complete. This avoids
  blocking on unkilled grandchildren.