Two independent root causes for '左右能拉，上下不能拉':
1. Event delivery: the top 6px edge band belongs to QMenuBar, which
actively consumes mouse events (menu hover) — nothing ever propagated
to QMainWindow, so T/TL/TR (and maximize-restore from top edge) were
dead on every platform. Install an event filter on menuBar()/statusBar():
within the edge band it handles hover cursor, press (grabMouse), drag
and release; outside the band everything passes through (menus open
normally, native QSizeGrip corner drag preserved).
2. Layout minimum: MonitorPanel's minimumSizeHint (~1015px of stacked
tables/group boxes) propagated through QTabWidget → central widget →
QMainWindow, making the window minimum size 898x1121. Any vertical
shrink was clamped dead. Set explicit per-tab-page minimumSize
(320x200) + window minimumSize (760x480).
Tests: test_resize.py now drives edges through real event delivery
(childAt → child widget, propagation/filter path) for L/R/T/B/TL/TR/BL,
verifies BR stays on QSizeGrip, maximize-restore via menubar filter,
and minimum-size regression. All 8 suites pass.
- ◀ button in sidebar header + '视图→隐藏侧栏 (Ctrl+B)' menu item
- collapsed state persisted to ~/.sshclient/settings.json, restored on startup
- floating '主机' button (theme-aware dark/light) to re-show sidebar
- fix: remove duplicate QShortcut — QShortcut + QAction shortcut on same key
both fired, toggling twice per Ctrl+B press (net zero effect)
- fix: floating button positioned with parent-local coords instead of
mapToGlobal — button no longer drifts off the window edge when the
window is not at screen origin; resizeEvent/moveEvent overrides removed
- remove dead terminal_input_set_enabled (referenced nonexistent cmd_input)
- test_sidebar.py: 7-step coverage incl. single-fire shortcut regression,
splitter width restore, cross-restart persistence; settings isolated to tmpdir
User asked: '窗口应该可以前后左右拉'
Added custom edge drag handling in MainWindow (QMainWindow's default
resize support exists but ignored + we want full control for
maximized state too):
New methods in ui/main_window.py:
- _hit_test_edge(pos) -> 'L'/'R'/'T'/'B'/'TL'/'TR'/'BL'/'BR'/''
6px margin around window. Returns which edge (or corner) the
cursor is on. Returns '' when maximized/fullscreen.
- _edge_to_cursor(edge): sets Qt.SizeHor/Ver/FDiag/BDiag cursor
- _do_resize(global_pos): computes new geometry based on drag
delta. Respects self.minimumWidth/Height (QMainWindow's auto
min size from menu/toolbar/statusbar) and screen bounds.
- _restore_from_max(global_pos): when maximized, click+drag on
TOP edge → showNormal() + set geometry to mouse-relative size
+ continue drag from 'TL' corner (like Windows native behavior)
- mousePressEvent/Move/Release: hook into resize state
- leaveEvent: clear cursor when leaving window
- changeEvent: clear resize state on window state change
State fields added in __init__:
- self._resize_edge, self._resize_start_geo, self._resize_start_pos
- self._resize_margin = 6
- self._dragging_from_max = False
Fix: 'QTabWidget.RightSide' was wrong API; used 'QTabBar.RightSide'
in the multi-tab commit (unrelated, but I noticed while testing).
New test_resize.py (5 tests, all pass):
- 8-region hit-test (4 edges + 4 corners)
- Right-edge drag → width grows
- Left-edge drag → x moves right (width limited by minW)
- Top-left corner drag
- Maximized + click top edge → restored to normal
Existing tests (core 6/6, UI 3/3) still pass.
Build: 57MB exe.
User asked: '标签使用鼠标右键应该有复制标签的功能'
In ui/terminal_tab_widget.py _on_tab_context_menu:
- Added '📋 复制标签' action between '重命名' and '关闭标签'
- New method _duplicate_tab(idx): creates a new terminal tab for
the same host (reuses SSH connection, opens independent shell
channel via paramiko invoke_shell)
- New tab title = '<src_title> (副本)' with ⚪ state dot
- If host not connected: duplicate creates a placeholder tab
- If host connected: duplicate immediately opens a new shell
Verified: 3 tabs after duplicating 'test-server' (欢迎 + ⚪ test-server
+ ⚪ test-server (副本))
All tests still pass.
Root cause: _current_line_start was set after \r\n but BEFORE the prompt
was rendered, so Backspace could erase prompt characters after reaching
the input line start.
Fix: send \x7f (DEL) to shell on Backspace and \x1b[3~ on Delete instead
of locally deleting characters. The shell's readline handles backspace
correctly - it ignores DEL at the input line start, so the prompt is
never touched. The existing \b handler in _on_data already correctly
processes the shell's \b-space-\b echo sequence.
Verified: typing 'abc' + 3 Backspace deletes input, 5 more Backspace
leaves prompt 'root@host:~# ' intact.
_on_delete() was missing _current_line_start check that _on_backspace
already had. Clicking into the prompt area (root@hostname:~#) and pressing
Delete would erase prompt characters. Now:
- Cursor before input line: moved to input start, no deletion
- Selection crossing prompt boundary: blocked
- ui/theme.py: light/dark QSS stylesheets, persisted to ~/.sshclient/settings.json
- main_window.py: View menu with dark theme checkbox, applies on startup
- ui/widgets.py: SparklineChart widget for mini trend visualization
- MonitorPanel: CPU and memory sparkline charts showing last 60 samples
- Charts cleared on stop/switch host
- core/monitor.py: add fast metrics script (0.3s vs 1.4s sleep),
collect() now defaults to fast=True for responsive UI
- ui/workers.py: MonitorWorker changed from loop mode to single-shot
(QTimer re-arms next cycle on finished signal, avoids paramiko
blocking in infinite loop)
- ui/widgets.py: remove debug print from _kick_one_sample
- test_monitor_nonblock.py: fix sample counting by wrapping _on_sample
instead of disconnecting signals (old approach missed new workers)
- test_process_monitor.py: assertions account for MAX_RENDER_ROWS=200,
fix MEM% column index (4 not 3) in sort test
User feedback: '进程比较关键，以下显示太少了，需要继续优化下监控UI'
Process monitor panel was rebuilt:
1. core/monitor.py: 200 -> 500 processes, added PPID / PRI / NICE
New ps fields: pid ppid user pcpu pmem vsz rss stat pri nice etimes times args
Python parser: new fields ppid, pri, nice
2. ui/widgets.py MonitorPanel:
a) Toolbar: search box + sort dropdown (CPU/MEM/PID/start/user/comm)
+ view toggle (Flat / Tree) + CPU% min filter (SpinBox 0-100%)
+ batch kill button
b) Table: 12 columns (Flat) / 13 cols (Tree + indent col)
PID PPID USER CPU% MEM% RSS STAT PRI NI '启动' 'CPU时间' 命令
- Compact row height 20px (was default 30)
- Extended selection (Ctrl/Shift multi)
- Color highlights:
CPU% >=80 red bg+fg bold, >=50 red fg bold, >=20 orange
MEM% >=15 orange bg+fg bold, >=5 orange
STAT: Z orange bg + orange-red fg bold, R green, D purple
NI: negative green, positive orange
- Red bg on EXTREME rows for instant visibility
c) Tree view: builds PPID tree with ├─/└─/│ connectors;
root processes (PID 1) at depth 0, sorted by CPU within siblings.
Useful for 'which process is spawning these things?'
d) Right side panel:
- 进程详情 (QTextEdit dark theme): full PID/PPID/USER/STAT/PRI/NI/
CPU/MEM/RSS/VSZ/etime/cputime, command, parent process name,
child processes list (top 10)
- 🔥 CPU TOP 5 (QListWidget dark): click to jump+select in main table
- 💾 内存 TOP 5 (QListWidget dark): same
e) Right-click menu enhanced:
kill (SIGTERM) | kill -9 (SIGKILL) | batch kill (if multi-select)
| copy PID | copy command | filter by command | view details
f) Batch kill: collects all selected PIDs, runs 'kill p1 p2 p3 ...'
in a single SSH call, then loops 'kill -0' to report survivors.
One confirmation dialog lists top 10 + '还有 N 个'.
g) Status row at bottom: 'Ctrl+A 全选, Shift+点击多选, 右键批量操作'
3. test_process_monitor.py: 5/5 tests pass with 500 rows, tree view,
TOP 5, CPU min filter, etc.
All tests pass: core 6/6 + UI 3/3 + E2E 6/6 + terminal 5/5 + proc 5/5
Build: 57 MB single-file exe.
User asked for process monitoring. Added:
1. core/monitor.py:
- Extended _LINUX_METRICS_SCRIPT with ===PROC=== section
- ps -e -o stat= | awk counts total/running/sleep/disk-sleep/zombie
- ps -eo pid,user,pcpu,pmem,vsz,rss,stat,etimes,times,args
Uses etimes/times (numeric) instead of start/time (string with
spaces) to avoid column-shift bugs. awk joins 10th col onwards
with spaces so comm retains its full command line.
- New empty-dict fields: proc_total, proc_running, proc_sleep,
proc_disk, proc_zombie, processes
- New processes[] array with full per-process fields
2. ui/widgets.py MonitorPanel:
- New _build_proc_summary() - 4-stat overview card (total/running/
sleep/zombie, color-coded)
- New _build_proc_table() - 9-column process list with:
* Search box (multi-keyword AND match across pid/user/stat/comm)
* Sort dropdown (CPU / MEM / PID / start time / user)
* Color-coded CPU% (red >=50%, orange >=20%)
* Color-coded MEM% (red >=10%, orange >=5%)
* Color-coded STAT (red for zombie)
* RSS formatted with format_bytes
* ETIME / TIME formatted as 5d3h / 2h15m / 45s
- _apply_proc_filter() - filter + sort + render in one pass
- _proc_context_menu() - right-click menu:
* kill (SIGTERM) | kill -9 (SIGKILL) | copy PID | copy cmd |
filter by this command
- _proc_kill_selected() / _proc_kill() - sends kill over SSH
with confirmation dialog, shows remote exit code in result
- Splitter: 4 metric cards + proc summary + disk + net on top,
process list on bottom. User-draggable.
3. New test_process_monitor.py (5 tests, all pass):
- Real SSH connection, collects process data
- Renders MonitorPanel with 200 rows
- Tests search filter / sort change / clear
- Tests killing a real sleep process: starts sleep 60, finds it
in the table, calls _proc_kill (with QMessageBox monkey-patched
to auto-yes), verifies remote kill -0 returns 'No such process'
- Also fixed passwords in test_e2e.py + test_terminal.py to match
the working local sshd credentials
All tests pass: core 6/6 + UI 3/3 + E2E 6/6 + terminal 5/5 + proc 5/5
Build: 57 MB single-file exe.
User reported: each prompt line shows a leading '[?]' box.
Root cause: Ubuntu's default PS1 uses \\e]0;user@host:dir\\a
(OSC 0, set window title) which our ANSI parser did not recognize.
It only handled SGR (\\e[...m), so the raw bytes were drawn as tofu boxes.
Fix in ui/terminal_panel.py:
- Added _OSC_RE for ESC ] ... BEL|ESC\\ sequences (set title, set cwd)
- Added _DCS_RE for ESC P/X/^/_ ... ESC\\ (DCS/SOS/PM/APC)
- _on_data() branches ESC prefix by next byte: '['->CSI, '] '->OSC,
'P/X/^/_'->DCS, others->skip 2 bytes
- Captured OSC 0/1/2 content into self._window_title (so future
versions can display it)
- _strip_ansi() now also strips OSC/DCS (for clean history recall)
Added [4b/5] test: injects a literal Ubuntu PS1 byte stream
(OSC 0 + SGR colors) and verifies the rendering is just the plain
'root@ubuntu2204:~# ' with no tofu / no leaked escape chars.
All 21 tests pass (core 6 + UI 3 + E2E 6 + terminal 6).
User feedback: 'your terminal can only send commands, not type
freely. Useless.'
Replaced the old 'single command + output' terminal with a true
interactive shell panel:
NEW: ui/terminal_panel.py
- Uses paramiko invoke_shell() to open a long-lived shell session
- Every keystroke is forwarded to the remote shell (not a textbox)
- Remote output (incl. ANSI color codes) is parsed and rendered
inline in a QPlainTextEdit (16-color SGR, bold, underline, invert)
- Cursor kept on the current input line; user can't edit past output
- Local command history with up/down navigation (stores last 200)
- Standard terminal shortcuts: Ctrl+C (SIGINT), Ctrl+D (EOF),
Ctrl+L (clear), Ctrl+A/E (line start/end), Ctrl+U/K/W
- Terminal resize forwarded to remote PTY (vim/top work correctly)
- Tab key forwarded to remote for completion
CHANGED: core/ssh_client.py
- Added open_shell(term_type, cols, rows) -> Channel
- Added resize_shell(chan, cols, rows)
CHANGED: ui/main_window.py
- Replaced old _build_terminal_tab (now removed) with TerminalPanel
- On host connect -> auto-open shell, switch to Terminal tab
- On host disconnect / switch -> close shell cleanly
NEW: test_terminal.py (5 tests, all pass)
- Real SSH connection, opens shell
- Sends 'ls /tmp' via simulated keypresses, verifies 162-line output
- Sends ANSI color command, verifies colored text appears
- Tests up-arrow history recall
- Verifies prompt returns after command completes
All existing tests still pass (core 6/6 + UI 3/3 + E2E 6/6 + terminal 5/5)
User reported: '已连接, 但终端啥也不显示'
Likely cause: command ran but stdout was empty / scrolled to wrong place
/ output widget didn't refresh before worker finished.
Changes:
- Eagerly show the typed command in the output area (\$ <cmd>) so user
sees the input was received even if the worker hangs/fails
- Show '(stdout 为空)' placeholder when worker returns no output, so it's
clear the worker actually ran
- Force output.repaint() and ensureCursorVisible() after append
- Add 2 diagnostic buttons in terminal bottom bar:
* [🔍 诊断] prints current SSH connection / worker state
* [▶ 测试命令] pre-fills a known-good test command
- Make CommandWorker explicitly check conn.connected before exec
- Surface full traceback in stderr on exception instead of bare str(e)
All 15 existing tests still pass (core 6/6 + UI 3/3 + E2E 6/6).
- Remove all non-ASCII text (cmd default GBK mangles UTF-8)
- Replace 'if errorlevel 1 (' multi-line blocks with 'goto :label'
- Replace 'echo(' with 'echo.'
- Use 'for %%I' to get file size (avoid UTF-8 console findstr issues)
- All error paths now end with explicit exit /b 1
Fixes#1: '涓嬭浇鍦板潃' (mojibake) and '('") parse errors.
Your Name