Comparing 4.1.2...4.1.3 · jline/jline3 (original) (raw)

Commits on May 23, 2026

1. Configuration menu
   Browse the repository at this point in the history

Commits on May 26, 2026

1. fix: clear ISIG in enterRawMode so Ctrl+C reaches raw-mode readers (f…
   …ixes #1900)
   After #1871 set VMIN=1/VTIME=0, Ctrl+C during DefaultPrompter.prompt()
   hangs the JVM because ISIG is still set: the kernel consumes the byte as
   SIGINT instead of delivering 0x03 to the input buffer, and the blocking
   read never returns.
   Clear ISIG in enterRawMode() to match POSIX cfmakeraw(3). Add an
   INTERRUPT widget to LineReaderImpl bound to VINTR so that Ctrl+C typed
   at the terminal still triggers UserInterruptException through the
   keymap rather than through the (now-absent) SIGINT signal.
   
   Configuration menu
   Browse the repository at this point in the history
2. Configuration menu
   Browse the repository at this point in the history
3. Configuration menu
   Browse the repository at this point in the history

Commits on May 27, 2026

1. Configuration menu
   Browse the repository at this point in the history
2. Configuration menu
   Browse the repository at this point in the history
3. Configuration menu
   Browse the repository at this point in the history
4. Configuration menu
   Browse the repository at this point in the history

Commits on May 28, 2026

1. fix: terminal.close() blocks when pump thread is reading stdin (#1911)

• fix: terminal.close() blocks when pump thread is reading stdin (#1909)
  On macOS, a pump thread blocked in a native read() on a tty can prevent
  tcsetattr() from completing, causing terminal.close() to hang until the
  user presses ENTER. Fix by switching to non-canonical mode with
  VMIN=0/VTIME=1 before shutdown, which forces the blocked read() to time
  out within 100 ms so the pump thread can exit cleanly before original
  terminal attributes are restored.
  Also improve NonBlockingInputStreamImpl and NonBlockingReaderImpl
  shutdown: set threadIsReading=false, interrupt the thread, and join
  with a timeout so close() reliably stops the pump thread.
• fix: ensure pump thread shutdown is robust on close

• AbstractUnixSysTerminal: wrap VMIN/VTIME unblock sequence in
  try-finally so input.shutdown() runs even if doGetAttributes()
  or doSetAttributes() throws
• NonBlockingInputStreamImpl/ReaderImpl: close the wrapped stream
  before shutdown/join so the native read() is unblocked before
  we wait for the pump thread to exit

• refactor: extract NonCloseable stream wrappers and add lifecycle tests
  Extract NonCloseableInputStream and NonCloseableOutputStream from
  DumbTerminalProvider into org.jline.utils for reuse. Apply them in
  AbstractUnixSysTerminal to prevent closing shared FileDescriptors
  (stdin/stdout/stderr) when the terminal is closed.
  Add timeout logging to pump thread shutdown in NonBlockingInputStreamImpl
  and NonBlockingReaderImpl. Add comprehensive tests for stream wrapper
  behavior and pump thread lifecycle.
• fix: wrap ExecPty system streams with NonCloseable wrappers
  ExecPty.doGetSlaveInput() and getSlaveOutput() returned raw
  FileInputStream/FileOutputStream on shared FileDescriptors when used
  as a system terminal. Closing the terminal would close the shared
  FDs, breaking System.in/out/err for the rest of the JVM.
  Wrap the system stream cases with NonCloseableInputStream and
  NonCloseableOutputStream, matching AbstractUnixSysTerminal and
  DumbTerminalProvider. Device-file paths are left unwrapped since
  those are non-shared FDs.
• refactor: remove VMIN/VTIME unblock sequence from doClose()
  The VMIN/VTIME trick cannot reliably unblock a native read() that is
  already in progress — tcsetattr only affects the next read call, not
  the current one. The NonCloseable wrappers are the actual fix: they
  prevent FileDescriptor.close0() from being called, which eliminates
  the macOS deadlock. The pump thread exits on its next iteration once
  the current read completes naturally.
• fix: reduce shutdown join timeout and prevent double-wait
  Reduce the thread join timeout from 500ms to 50ms — the thread either
  exits promptly (when waiting) or is stuck in a native read (where no
  timeout helps). Setting thread = null after join prevents the second
  shutdown() call in the close chain from waiting again.
• refactor: extract PumpThread to deduplicate thread lifecycle management
  NonBlockingInputStreamImpl and NonBlockingReaderImpl had identical
  startReadingThreadIfNeeded(), shutdown(), and thread management fields.
  Extract into a shared PumpThread helper that both classes delegate to.
• refactor: move run loop into PumpThread to further reduce duplication
  The run() method was still duplicated between NonBlockingInputStreamImpl
  and NonBlockingReaderImpl. Move the loop into PumpThread.runLoop() with
  IoReader and ResultHandler callbacks. Each Impl class now has a one-liner
  run() that delegates to pump.runLoop().
• fix: address SonarCloud issues in PumpThread
  Store lock as a field instead of passing as method parameter (S2445).
  Use notifyAll instead of notify (S2446). Re-interrupt on
  InterruptedException in runLoop (S2142).
  
  Configuration menu
  Browse the repository at this point in the history

1. Configuration menu
   Browse the repository at this point in the history
2. Configuration menu
   Browse the repository at this point in the history
3. Configuration menu
   Browse the repository at this point in the history
4. fix: fix command argument parsing in DefaultCommandDispatcher (#1924)

• Fix command argument parsing
• Fix potential resource leak by handling all possible unchecked exceptions.
• Fix test as per #1238 (comment)
• fix: resolve rebase conflicts and address review feedback

• Fix leftover conflict markers from rebase onto master
• Remove redundant parser creation in ShellBuilder — pass this.parser
  directly since DefaultCommandDispatcher handles null
• Quote script path in sourceCommand test to handle Windows backslashes

• fix: share single parser instance between dispatcher and reader
  Restore the shared parser creation in ShellBuilder.build() so that
  DefaultCommandDispatcher and LineReader use the same Parser instance
  when none is explicitly configured.

Co-authored-by: Elec332 6115944+Elec332@users.noreply.github.com

Configuration menu
Browse the repository at this point in the history