fix(plotting): ngspice plotting module - data extraction rewrite + correctness fixes (follow-up to #416)

[VaradhaCodes]

fix(plotting): ngspice plotting module - data extraction rewrite + correctness fixes (follow-up to #416)

Follow-up to #416. After that merged we ran the new plotter against all 11 example circuits across all three analysis types and did extended manual testing on real designs. This PR has everything that came out of that. All 5 files are PyQt6 ready and compatible with #497.

Files changed: plot_window.py, NgspiceWidget.py, data_extraction.py, plotting_widgets.py, TerminalUi.py

---

What is in this PR

Area	Fixes
Data Extraction	Full rewrite, tested on all 11 circuits across 3 analysis types
Digital Timing View	9 fixes found during real circuit testing
Cursor System	4 fixes, rebuilt with drag support
General Plotter	12 correctness fixes
Simulation Console	3 fixes in NgspiceWidget and TerminalUi

---

Data Extraction - Full Rewrite

This is the main change in the PR. We rewrote the entire data parser from scratch.

The old parser was reading each output file multiple times and using line-index arithmetic to find data. This was working fine for simple circuits but was breaking silently on several ngspice output formats that show up in real projects. We replaced it with a single-pass state machine in _parse_plot_file that handles everything in one read.

What we fixed in the process:

1. Multi column group files - ngspice splits wide node tables across multiple sections with * headers when there are too many nodes. The old parser was only reading the first group and silently dropping the rest.

2. Page break headers - ngspice repeats the Index header row every 55 or so data rows. The old parser was treating each of these as the start of a new data group which was messing up the data.

3. AC complex format - AC analysis rows have two tab columns per node, real and imaginary parts. The real part also has a trailing comma that ngspice puts there. We now strip that and discard the imaginary column, matching the original behaviour.

4. Duplicate node names - ngspice truncates long node names and two different nodes can end up with the same string. The old dict based approach was merging them into one channel with double the data. Fixed with positional lists so each column gets its own slot.

5. Pre existing truncation bug - there was a 2 row truncation in the original parser that we also fixed here. BasicGates was showing 164 points, correct count is 166.

Tested against all 11 example circuits - Transient, DC, AC dec and AC lin, subcircuits, and circuits with up to 76 nodes. Parses Monostable555 (205k data points, 22 nodes) in 0.89s. Output verified to be identical to the original parser across all circuits.

The public interface is unchanged so nothing breaks. openFile(), computeAxes(), numVals() all work exactly as before.

---

Digital Timing View

Tested the timing view on real circuits and found several things that needed fixing.

1. Threshold logic - updated to per-trace midpoint (VDD/2). This is how every scope and logic analyzer auto thresholds and it works correctly on mixed voltage circuits. Manual spinbox override still works.

2. Trace height - updated to use uniform normalised heights regardless of voltage domain. On a mixed voltage circuit like 3.3V and 5V signals together the traces were overlapping because height was proportional to actual voltage swing.

3. Threshold lines - fixed the threshold line position. Each trace is stacked at rank times spacing in y-space and the line was not accounting for that so it was in the wrong position for every trace except the bottom one.

4. Disabled for AC and DC - the timing view was enabled for all analysis types. AC data is frequency domain so showing it as a digital timing diagram gives a meaningless plot. Disabled it with a tooltip for non-transient analysis and added a guard inside the function as a second layer.

5. Constant DC signals - when a signal had zero swing the threshold comparison always came out False and showed the signal as all-low regardless of actual voltage. Now detected separately and shown as DC: X.XX V.

6. Transient offset handling - fixed the transient start time offset. It was trying to slice obj_dataext.y in-place but calling .keys() on a Python list which throws an AttributeError. The exception was being swallowed silently so nothing was working. We now build a local y_data dict per render and never touch obj_dataext.

7. Per trace time slice - fixed a variable leak in the trace loop. When one trace had a shorter y-array than the time axis the code was trimming the outer time_data variable which was then affecting all the traces after it. Now uses a local trace_time per iteration.

8. Timing to normal view switch - fixed an issue where switching from timing view back to normal view with autoscale off was restoring the timing y-limits into a voltage plot. Signals spanning 0-5V were showing as a flat line near the x-axis. Added a _last_was_timing flag to block this.

9. Annotations - each trace now shows VOH and VOL from the actual simulation data, for example H: 4.97 V and L: 0.02 V. Periodic signals also show detected frequency like 1.00 MHz. Constant DC signals show DC: 3.30 V. Current traces use proper units like mA and uA. Frequency detection uses rising-edge timing with a guard to filter out non-periodic signals.

---

Cursor System

1. Right click conflict - right click was triggering both a Qt context menu and placing cursor 2 at the same time. The context menu items were already in the toolbar and side panel so we removed it from the canvas entirely. Left click is C1, right click is C2, no conflict.

2. Drag to reposition - added drag support. _find_nearest_cursor() checks if the click is within 8px of an existing cursor. If yes, dragging moves it in-place instead of dropping a new cursor on every click.

3. Cursors disappearing after redraw - refresh_plot calls fig.clear() which wipes all matplotlib artists including the cursor lines. The positions were stored but the lines were never recreated after a redraw so toggling a trace or changing the grid was making cursors vanish visually. Added _restore_cursors() to recreate them after every refresh.

4. Unified time scale - the time unit selection logic was copy pasted in three places and two of them had a bug with zero-span simulations. Pulled it into one method _get_time_scale_and_unit() that everyone now calls.

---

General Plotter

1. Waveform clicks not registering - item.isSelected() always returns False when list items have custom child widgets via setItemWidget() because clicks land on the child widget and Qt never updates the selection model. Replaced with an explicit t.visible toggle.

2. Hide and left-click out of sync - right click Hide was calling line_object.set_visible(False) directly without updating t.visible. Any redraw after that would re-read t.visible = True and silently bring the trace back. Both paths now go through t.visible and refresh_plot().

3. Plot Function stacking traces - clicking Plot Function multiple times was stacking traces on top of each other with no way to clear. We now remove the previous result before adding a new one, tracked in self._func_line.

4. eval() on user input - the expression input was going directly to Python's eval(). Replaced with _safe_eval(), an AST based evaluator that only allows trace names and arithmetic operators.

5. vs matching node names - if 'vs' in function_text was matching any node name that contains vs like v(vss), avss, vsupply and so on. These are very common supply rail names in circuit simulation. Changed to check for vs with spaces on both sides.

6. steps-post rendering as steps-pre - axes.step() was being called without a where argument which defaults to 'pre'. Transitions were drawing one sample too early. Now passes where='post'.

7. Multimeter widget getting garbage collected - MultimeterWidgetClass had no parent and no persistent reference outside the loop it was created in. Python could GC it and close the window right after creation. Now kept in self._meters for the lifetime of the plot window.

8. Figure Options wiping the title - self.fig.suptitle('').get_text() was being used to read the current title. Calling suptitle() always writes so the title was being cleared every time the dialog opened. Fixed by reading self.fig._suptitle directly.

9. Autoscale checkbox doing nothing - stateChanged was never connected and refresh_plot was never reading the checkbox state. Now properly wired up. When autoscale is off, zoom limits are preserved across redraws.

10. Zoom In and Out menu actions - Zoom In was toggling the toolbar rubber band selector mode. Zoom Out was going back in the view history. Neither was actually zooming. Both now do a proper scale step from the centre of the current view.

11. Escape key stuck in zoom or pan mode - Escape was always calling clear_cursors() regardless of what toolbar mode was active. Now checks the toolbar mode first and exits zoom or pan if active.

12. format shadowing Python built-in - a local variable named format was shadowing the built-in format() function in that scope. Renamed to fmt.

---

Simulation Console

1. Stderr filter was never running - QProcess.MergedChannels merges stderr into stdout at the OS level so readAllStandardError() always returned empty bytes. The noise filter for NGSpice batch mode messages was dead code from the beginning. Changed to SeparateChannels with separate slots for stdout and stderr so the filter actually runs now.

2. Shell commands breaking on project paths with spaces - GAW and xterm commands were built with plain f-strings so any project name or path with a space was breaking the shell command without any error. All paths now go through shlex.quote(). Windows mintty was also switched to QProcess.start(program, args) so Qt handles quoting.

3. Dead property read removed - finish_simulation was reading a "plotFlag" property from the process that nothing in the codebase was ever writing. Removed. Also renamed "plotFlag2" to "redoPlotFlag" in both files for clarity.

---

The complete root cause analysis and notes for each fix are in BUGS_AND_FIXES.md in the repo root for anyone who wants to dig deeper.