Version 3.1.2

• Require winpty-rs 1.0.8, which makes EOF detection work and fixes spawning
  an executable whose path contains a space. Measured through this package's
  own API:

  	3.1.1	3.1.2
  while not pty.iseof(): pty.read()	never terminates	exits via iseof()
  ...time to exit, ConPTY	5.07 s, WinptyError('')	0.20 s, no error
  PTY.iseof(), both backends	always False	True once drained
  spawn from a path containing a space	fails, bad-image error	spawns, argv[0] intact
  errors with an empty message	1	0

• Behavioural change. PTY.iseof() could never return True before, so
  code written around that behaves differently now. A loop of the shape

  while not data and not pty.iseof():
      data = pty.read()

  • stop when there is nothing to read and we are not at EOF - used to end
    on its first empty read and will now run forever; the condition wants
    inverting. There is no API change, so nothing gates the upgrade.

• Spawning from a path containing a space failed differently on each backend.
  The WinPTY backend ran the right executable but split argv, so
  C:\Program Files\Python\python.exe arrived as argv[0] = "C:\Program"
  with the remainder shifted into the caller's arguments. The ConPTY backend
  left argv[0] intact but let CreateProcessW guess which file to run,
  which could fail the spawn outright with a bad-image error whose %1
  placeholder was never substituted - reported in
  #583 as being related to
  path length, though what actually decided it was whether a shorter candidate
  existed on disk.

• Closing a PTY no longer waits out a fixed five seconds. A blocking read
  after the child exits now returns in about 175 ms rather than 5.1 s, and
  teardown no longer cancels I/O that has stopped belonging to it, which could
  abort a read owned by the host application.

• Still not fixed: the output truncation half of
  #565 is
  cpython#86210, where an
  unbuffered _textiowrapper_writeflush ignores a partial write and drops the
  remainder. That is a CPython bug on the producer side and is unaffected by
  this release; a buffered producer, or one that checks the return value of
  write, is not affected. The EOF half of that report is fixed. See also the
  dated correction added to the 3.1.1 notes.

Issues Closed

Fixed in this release:

• Issue 583 - spawn() fails with "is not a valid Win32 application" when the executable path contains a space (PR 592 by @andfoy)
• Issue 565 - ConPTY may fail to reach EOF after child exit (PR 592 by @andfoy)
• Issue 519 - How to properly read until EOF using the WinPTY backend (PR 592 by @andfoy)
• Issue 423 - Freeze on process.iseof() (PR 592 by @andfoy)

Closed during triage after 3.1.1 - already fixed in an earlier release, not
reproducible, or answered:

• Issue 553 - pty.write() panics on strings containing UTF-16 surrogate pairs
• Issue 545 - process read and write not working as expected
• Issue 536 - Resource files are not correctly attached while building
• Issue 529 - unexpected behavior in case of 'command not found'
• Issue 516 - ModuleNotFoundError: No module named 'winpty.winpty'
• Issue 510 - Documented encoding parameter does not work
• Issue 486 - Build fails because of missing project.version in pyproject.toml
• Issue 463 - After the program sends a character, it takes at least 300 ms to read it
• Issue 461 - break in finally swallows exceptions
• Issue 458 - Installing v2.0.14 requires Rust
• Issue 329 - ssh connection establish failed.
• Issue 304 - 'Pty is closed.' BUG
• Issue 269 - Garbled output on Telnet Client
• Issue 197 - test_ptyprocess.py terminating early due to child process
• Issue 183 - Constructing PTY instance kills terminal

In this release 19 issues were closed.

Pull Requests Merged

• PR 592 - Bump winpty-rs to 1.0.8, by @andfoy (583, 565, 519, 423)

In this release 1 pull request was closed.