Fix GUI window freeze on Windows during operations

[RenanGBarreto]

🤔 Why?

On Windows, starting any operation (pack folder, pack file, verify) made the entire GUI window freeze — "Not responding" in the title bar until the command finished. The window couldn't be moved or interacted with.

The root cause: multiprocessing.Pool uses spawn on Windows, which re-imports __main__ in each child process. When this happens from a non-main thread inside a --windowed PyInstaller binary, the spawn handshake blocks the Windows message pump. macOS uses fork and was never affected.

🔧 What changed

• Replaced the in-process cli_mkpfs_main() call in BasePanel._run_mkpfs with a subprocess.Popen child that streams output back to the log pane
• Added a --gui-subprocess router in mkpfs/gui/__main__.py so the frozen binary runs the CLI instead of the GUI when the marker is present
• Stored the Popen handle on self._proc for a future stop/cancel button
• Suppressed Windows console-window flash with CREATE_NO_WINDOW (essential for dev mode, harmless on frozen exe)
• Auto-confirmed overwrite prompts by piping "y\n" to stdin (replacing the old builtins.input patch)

🧪 How to test

# Verify the subprocess routing works (dev mode):
python -m mkpfs.gui --gui-subprocess -V

# Run full test suite:
bash run-tests.sh

💬 Notes

• The progress bar becomes indeterminate across the process boundary (detectable phase progress is not available), but the log pane still streams all output in real time
• The change is minimal and focused on the threading fix; no other panel code or behavior is affected
• Closes [BUG] GUI window becomes unresponsive (Not responding) on Windows after starting any operation #120

[github-code-quality]

Code Coverage Overview

Languages: Python

Python / code-coverage/pytest

The overall line coverage in commit 2bbf86a in the fix/gui-windows-free... branch remains at 72%, unchanged from commit 2abb9cd in the main branch.

---

_{Updated August 31, 2026 02:47 UTC}

[RenanGBarreto]

Code Review

Overview

The subprocess approach is sound — it avoids multiprocessing.Pool's spawn re-importing __main__ and blocking the Windows message pump. The freeze_support() → --gui-subprocess routing is correctly placed. However, there are three confirmed bugs (two reported in #122) plus dead code from the progress listener.

---

🔴 Bug 1 — Progress bar spam: text=True universal newlines split \r-delimited progress into separate lines

base.py:375 — text=True enables universal newline mode, which translates \r → \n. pbar.py writes \r-delimited progress updates (e.g. \r[####--] 50% compress). With universal newlines, each progress tick becomes its own line in the parent — the \r-stripping logic at lines 405-406 is dead code because \r is already consumed.

Verified empirically: a child writing \r[----] 0%\r[#---] 20%...done\n yields 6 separate lines to the parent.

Fix: Use binary pipes + io.TextIOWrapper(proc.stdout, encoding="utf-8", errors="replace", newline=""). newline="" preserves \r so the existing rsplit("\r", 1) logic works.

---

🔴 Bug 2 — UnicodeEncodeError on Windows (#122)

Two compounding problems:

1. cli.py:1640 hardcodes 🎉 directly, bypassing the icon()/supports_utf8() ASCII fallback. The sibling call site at cli.py:1281 does it correctly: info("Image created successfully!", icon_name="success").

2. logging.py:83 — print() has no UnicodeEncodeError guard. When the pipe encoding is cp1252, any non-ASCII output crashes.

Fix: Use icon_name="success" at cli.py:1640. Harden logging.log() with a try/except UnicodeEncodeError fallback. Reconfigure stdout/stderr to UTF-8 in the --gui-subprocess branch of __main__.py.

---

🔴 Bug 3 — Dev-mode subprocess invocation is broken

base.py:365 — cmd = [sys.executable, "--gui-subprocess", *args] passes --gui-subprocess directly to the Python interpreter, which rejects it:

$ python3 --gui-subprocess -V
unknown option --gui-subprocess

This only works in a frozen PyInstaller binary where sys.executable is the exe itself. In dev mode, sys.executable is python3 and --gui-subprocess is an unknown Python option.

Fix: Branch on getattr(sys, "frozen", False) — in dev mode, use [sys.executable, "-m", "mkpfs.gui", "--gui-subprocess", *args].

---

🟡 Dead code — QueuedProgress, _progress_queue, _queued_progress, _drain_progress_events

With the listener never set in the child process, nothing pushes to _progress_queue. _drain_progress_events() polls an empty queue every 80ms — these are now dead.

Progress can be tracked by parsing the \r-delimited progress lines from the subprocess output (the [bar] N% phase format from pbar.py) to drive the determinate progress bar. This eliminates the spam and restores real progress tracking, so the dead code can be removed entirely.

---

🟢 Minor — self._proc thread safety

self._proc is written from the worker thread and would be read from the UI thread by a future stop button. A threading.Lock would prevent a TOCTOU race. Not a blocker since the stop button isn't wired up yet, but worth noting before it is.