When launched via pythonw.exe (no console), sys.stdout/stderr are None.
Accessing .encoding on None raised AttributeError, caught by the GUI's
pipeline import guard and shown as 'Failed to load pipeline'. Added
None check before the encoding check in fix_console_encoding(), added
a test, and documented the pitfall in CLAUDE.md.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Three issues caused the Logs view to appear blank during a real pipeline run:
1. `from run import step_fetch, step_link` was outside the worker's
try/except/finally. An import failure silently killed the thread,
leaving _pipeline_done uncalled and the Run button stuck disabled
forever. Now wrapped in its own try/except that posts the error to
the log and resets the UI.
2. `build_server_index` makes N sequential HTTP requests (one per mod
folder's meta.cpp) with no output during the scan. Added an optional
`progress_fn(current, total, name)` callback; step_fetch wires it to
print progress every 25 folders so the log never goes silent.
3. No immediate feedback after clicking Start — the log was blank until
the worker thread started printing. Now posts a "Pipeline started"
banner from the main thread before the worker launches.
GUI log batching (_poll_log now drains queue into a single CTkTextbox.insert
call per 80 ms tick instead of N calls, each with see("end") scroll).
_QueueWriter strips ANSI/CSI escape codes and bare \r before enqueuing so
tqdm progress output is legible in the log textbox. OSC sequences terminated
by both BEL (\x07) and ST (\x1b\) are handled.
Wizard "Test Connection" moved off the main thread: requests.get runs in a
daemon thread; result posted back via after(0, ...). Widget refs captured
before thread launch to prevent stale updates if user navigates away. Bare
except narrowed to TclError (destroyed-widget guard only).
Code quality: import os moved to module level in app.py; _read_raw_config()
helper extracted to deduplicate dual raw config.json reads; return type
annotations added to _get_view_class, _get_dashboard, and cfg property.
Tests: 11 new unit tests for _QueueWriter (RED -> GREEN on OSC-ST fix).
Introduces a two-language (EN/VI) i18n system with hot-swap support.
All ~160 user-facing strings are centralised in gui/locales.py; views
retranslate in-place on language switch without restarting the app.
- gui/locales.py: new file — _EN/_VI dicts, t() lookup, set_language(),
get_language(); assert guard ensures EN/VI key parity
- gui/app.py: switch_language(), _apply_startup_language(),
_save_language_pref(), _rebuild_nav_labels(); language stored in
config.json under ui.language; pipeline step headers and run_tool
status lines translated
- gui/views/settings.py: Language dropdown card (English / Tiếng Việt)
- gui/views/dashboard.py: all strings via t(); static header widgets
stored and retranslated in refresh()
- gui/views/mods.py: all strings via t(); _STATUS dict built at call
time so server status labels update on language switch
- gui/views/tools.py: all strings via _translatable registry; tab names
and segmented-button values kept in English (CTkTabview constraint)
- gui/views/logs.py: title + Copy/Clear buttons stored, retranslated
- gui/wizard.py: all 3 pages fully translated
- docs/huong-dan-su-dung.md: full Vietnamese user guide
- CLAUDE.md: documents localization architecture and constraints
Covers: common commands, package vs CLI architecture, data flow,
group naming convention, server index structure, junction/symlink
rules (critical: never shutil.rmtree), check_names two-pass
classification, Python 3.9 from __future__ requirement, and
test suite structure.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
