v0.8.0a5 — screenshooter forest/menu accept .scriptreeforest

Patch release on top of v0.8.0a4.

What's new

The screenshooter's forest and menu kinds now accept a
full .scriptreeforest file (the workspace container) as input,
not just a single .scriptree / .scriptreetree catalog.

forest kind on a .scriptreeforest

Each item in the forest file becomes a cell at a honeycomb
neighbour slot around the hub (inner ring first, outer ring on
overflow). The composite shows your REAL workspace layout as a
single PNG — same composition the live forest launcher produces
at startup, just rendered off-screen.

menu kind on a .scriptreeforest

Every item gets added to the forest's _members, so the merged
tree-popup populates with all of them. The captured menu matches
what the live ScripTreeForest pops up on a forest double-click
(header + search bar + sub-menu per item, nested ring sub-menus
included).

Usage

python screenshooter.py forest my-workspace.scriptreeforest --out workspace.png --cell-size 96
python screenshooter.py menu   my-workspace.scriptreeforest --out workspace-menu.png --cell-size 96

Or via the new run_screenshooter.bat launcher (also v0.8.0a4).

Tests

2 new in tests/test_screenshooter.py write a small on-disk
.scriptreeforest fixture (tmp_path) and assert both kinds
round-trip successfully. Suite: 18/18 screenshooter tests pass.

Install

Download the zip, extract, and double-click run_scriptreeforest.bat
(forest workspace), run_scriptreering.bat (bare ring shell),
run_scriptree.bat (V1 editor), or run_screenshooter.bat
(headless screenshot tool — no args opens a GUI form).

v0.8.0a4 — launcher table fix + run_screenshooter.bat

Patch release on top of v0.8.0a3.

Fixes

• README launcher table — previously claimed "two launchers" but
  the project actually ships three. The primary entry point
  (run_scriptreeforest.bat — the forest workspace) was missing.
  Now lists all four entries: forest workspace, bare ring shell, V1
  editor, and the new screenshooter launcher.

• run_screenshooter.bat (new). Double-click launcher for the
  headless screenshot tool. With no arguments, opens
  screenshooter.scriptree in the V1 editor so the user gets a
  labeled form (dropdown of the 8 kinds — cell / form / tree /
  editor / tabs / forest / menu — plus width / height / cell-size
  spinners and a file picker). With arguments, CLI passthrough to
  screenshooter.py. Same Python search as the other launchers,
  so a portable install (with lib/python/ vendored) works without
  a system Python install.

• Vendored-pypi prelude in screenshooter.py so the script
  runs on portable installs where PySide6 lives under lib/pypi/
  rather than being installed system-wide.

• One stray Unicode arrow in argparse help that broke cmd.exe's
  cp1252 console on --help.

Install

Download the zip, extract, and double-click one of:

• run_scriptreeforest.bat — the forest workspace (most users
  want this).
• run_scriptreering.bat — bare cell + ring shell.
• run_scriptree.bat — V1 editor directly.
• run_screenshooter.bat — headless screenshot tool (GUI form
  with no args; CLI passthrough with args).

v0.8.0a3 — screenshooter editor/tabs/forest/menu

Patch release on top of v0.8.0a2.

Screenshooter — four new kinds

The headless screenshot tool can now capture the views a tool author
actually wants for documentation:

• editor — full editor MainWindow with tree (left) + form
  (centre) + extra-args / command-line / output (right & bottom).
  When given a .scriptreetree it auto-selects the first leaf so
  every dock populates — no more "tree-only" capture with an empty
  right side.
• tabs — StandaloneWindow.from_tree capture with one tab per
  leaf tool. This is what a cell shows when single-clicked on a
  .scriptreetree.
• forest — composite PNG of the workspace forest hub with the
  catalog's cell docked beside it. Shows the user "what this looks
  like attached to the desktop forest".
• menu — composite PNG of the forest hub with its merged tree-
  popup menu rendered below. Shows the menu the user sees when
  they double-click the forest.

Composition uses a new _capture_composite helper that
grab() -s each unshown widget and pastes them onto one canvas —
the underlying "no flash, no focus theft" guarantee is preserved.

Front-end screenshooter.scriptree dropdown updated with the
four new choices and descriptive labels.

Refactor

tree_popup.show_tree_popup_for split into
build_tree_popup_menu (builds + returns the QMenu) +
show_tree_popup_for (calls build then positions / exec()'s).
This is what lets the screenshooter capture the menu without
showing it.

Tests

6 new in tests/test_screenshooter.py covering each new kind +
the auto-leaf-select branch + the tabs-rejects-.scriptree
error path. Full suite: 1816 passed, 5 skipped.

Install

Download the zip, extract, and double-click run_scriptreering.bat
(the cell shell) or run_scriptree.bat (the V1 editor).

Usage

python screenshooter.py editor my-tree.scriptreetree --out editor.png --width 1200 --height 780
python screenshooter.py tabs   my-tree.scriptreetree --out tabs.png
python screenshooter.py forest my-tool.scriptree     --out forest.png --cell-size 96
python screenshooter.py menu   my-tree.scriptreetree --out menu.png   --cell-size 96

v0.8.0a2 — hover tooltip stays on-screen

Patch release on top of v0.8.0a1.

Fix

Hover tooltip no longer falls off the right edge of the screen.
Previously the tip was offset down-right of the cell's bottom-centre
with a fixed +12/+18 px offset; for cells parked near the right edge
of the screen most of the tip ended up off-display, and even on the
left side it sat noticeably far below the cell.

v0.8.0a2 centres the tip horizontally on the cell and parks it 6 px
below the cell's bottom edge. Both axes are clamped to the
containing screen's available rect — if the below-cell position
would run off the bottom edge the tip flips above the cell
instead.

Other

• Bundled combridge.exe runtime is unchanged from v0.8.0a1.
• Test suite: 1810 passed, 5 skipped.

Install

Download the zip, extract, and double-click run_scriptreering.bat
(the cell shell) or run_scriptree.bat (the V1 editor).

v0.8.0a1 — link/dock split + bundled combridge

First alpha of the v0.8.0 line. Bundled COM-bridge runtime
(lib/combridge/) plus a major rework of the cell-graph model
that powers the hexagonal-cell shell.

⚠️ Pre-release — tested in single-user desktop usage but the
link/dock split touches a lot of code. Back up your
~/Documents/ScripTree/ (or wherever your .scriptreering
files live) before trying it on important layouts.

What's new

Cell graph: link vs dock split

The old _group_master_id field conflated two different
relationships. v0.8.0 splits them:

• LINK — tree-shaped membership (forest → rings → cells). Drives
  the merged menu, save / close behaviour, autoload.
• DOCK — spatial edge-adjacency. Drives the drag cascade and the
  snap-commit dock writes.

Rings stay link-attached to the forest even when dragged out of the
forest cluster; only dock-attached siblings move along during a forest
drag. Cells dragged out of a ring stay put when the ring later
moves (was: they followed the ring around).

Auto-named rings + cleaner merged menu

Freshly spawned rings get a session-unique name (Ring 1,
Ring 2 …) shown in the hover tooltip and merged-menu header. The
forest menu now recurses into nested rings so a brand-new ring shows
up as a sub-menu (empty rings get a (empty) hint instead of
disappearing).

Shake-to-close rings (replaces auto-close on quorum loss)

Rings no longer disappear automatically when their member count drops
below 2. Shake a ring to get a Save / Close / Cancel prompt
instead. Forest is excluded (it never closes by shake).

Bundled combridge runtime

lib/combridge/ ships with the release zip. Includes the
combridge.exe host, the ComBridge.Core engine, and adapter
plugins for SolidWorks, Excel, Word, PowerPoint, and Outlook. Catalogs
that target a vendor app can run straight from the portable copy.

Drag-end behaviour

• Snap-shift on a cell propagates to every cell docked to it.
• When a ring docks to a cluster and some members would overlap other
  cells, the master stays at its dock slot and only the overlapping
  members get re-slotted individually (no more "ring bounces away
  from its dock").
• Snap-preview overlay now shows when dragging a ring (previously only
  cells got the dock-slot outline).
• Closing the whole forest prompts exactly once per unsaved ring
  inside (was: prompted for the forest itself too, producing two
  identical-looking dialogs).

Tiling math centralised

New scriptree/shell/tiling.py module owns honeycomb slot-position
and polygon-collision math; layout.py, snap_engine.py,
group_layout.py, and the layout_sim developer tool all
delegate to it.

Test suite

1799 passed, 5 skipped (intentional spec-change skips on the now-
disabled quorum-loss auto-close path).

Install

Download the zip, extract, and double-click run_scriptreering.bat
(the cell shell) or run_scriptree.bat (the V1 editor). The
bundled combridge means catalogs that drive vendor COM apps work
without a separate install.

v0.6.38 — membership self-heal + slot-aware drag cascade

What the v0.6.37 trace exposed

Cells with parent=forest, slot=set, NOT in forest._positioned — they were correctly linked but got "left behind" when you dragged the forest. Also phantom ids in _positioned (leftover from closed pair-masters) and one cell with parent=None, slot=set (orphan).

Root cause: the snap-commit pair-master spawn path mutates membership in multiple steps; failure at any step leaves inconsistent state. Fixing each of the 60+ mutation sites would be a multi-session refactor.

What this release does

_audit_membership() — defensive self-heal on every master drag

On every mousePressEvent for a master, walk _members + _positioned, repair four kinds of corruption in place:

• Phantom ids in _positioned (no registry entry) → removed
• Cells in _members but registry has nothing → removed
• Cells linked-with-slot but not in _positioned → added
• Orphan cells (parent=None, slot set, but in _members) → reparented

Every fix is logged via the layout_trace AUDIT_* events so the next diagnostic session can see what was healed.

Drag cascade uses link+slot as ground truth

GROUP_MOVE now iterates the union of _positioned AND cells whose parent + slot independently point to this master. A stale _positioned can no longer strand a docked cell.

Tests

1745 passed (4 new audit cases).

v0.6.37 — two bugs nailed by the v0.6.36 trace

Bugs fixed (both diagnosed from the v0.6.36 diagnostic log)

1. Cells stacked at spawn instead of laid out

The trace showed every new cell spawning at the master's left-edge position (e.g. 92,1311) with NO LAYOUT_RUN between cells. Only after some unrelated event triggered the next repack did all the stacked cells snap into honeycomb slots at once — what you saw as "started off spaced out".

Fix: forest_controller._spawn_item now calls the master's layout function after each cell is added, so every new cell lands on its slot as it appears.

2. Master snapping to its own descendants

25+ SNAP_COMMIT events in the trace had the forest snapping to one of its own children. When the forest is dragged, the children move with it (the GROUP_MOVE cascade), and the snap engine saw them as nearby snap targets — offering the forest a slot relative to its own children. That's the "highlighted docking is misaligned with the forest" symptom.

Fix: SnapEngine._tick now excludes the dragged source's descendants (recursive through any nested masters) from candidate snap targets.

3. Icons / overlay oversized — instrumentation added

No root cause from the v0.6.36 trace (SNAPSHOT showed all cells at size_px=56). Extended the SNAPSHOT to log widget_w/h + icon_scale per cell, so the next time it happens the trace will show whether logical size and rendered size diverge.

Tests

1741 passed.

v0.6.36 — diagnostic tracing for layout issues

What's new

Verbose diagnostic tracing of every layout-affecting event. Per user request: "tie up cpu cycles if you have to just to do the tracking and logging. get all the data you think you need."

The log is written to %TEMP%/scriptree-layout-trace-<stamp>-<pid>.log (path printed to stderr on startup). Every event is one grep-friendly line.

Captured events

• CREATE — cell construction
• MOVE — every moveEvent with from/to/delta/drag-state/slot/parent
• PRESS / RELEASE — drag boundaries with full world SNAPSHOTs
• GROUP_MOVE — master drag cascade with delta and member list
• LAYOUT_RUN — every _compute_layout call
• SLOT_ASSIGN — slot assignments inside _compute_layout
• LAYOUT_MOVE — every position write from _compute_layout
• AUTO_HIDE / AUTO_SHOW — visibility flips
• COLLAPSE_START / EXPAND_START — animation kickoffs with snapshots
• SNAP_ATTACH / SNAP_DETACH / SNAP_COMMIT — snap-engine activity
• SNAPSHOT — full world dump at key transitions

CPU cost

Tracing is event-driven — no continuous timer. When idle, no events fire, no CPU spent. The idle-CPU contract is preserved. Disable with SCRIPTREE_LAYOUT_TRACE=0.

Other changes

• __build_date__ is now local time (was UTC). About dialog reflects this immediately.

Tests

1741 passed.

v0.6.35 — slot-based layout algorithm wired into startup

Algorithm-first refactor: startup path

This version ports the proven slot-based layout algorithm (designed + tested in a pure-Python simulator first, see tools/layout_sim.py) into the live cell shell. The startup path now uses it; everything else (drag, snap, collapse) still rides the older code paths until the next session ports them too.

What you should see

• No more startup jumble. When the forest opens, every cell snaps onto a free honeycomb slot around its master via the algorithm — same code the simulator proves correct across 14 scenarios.
• No more "cells at the forest's centre." The algorithm reserves the back-toward-parent slot for any docked master, so a child of a ring can never land on the forest's centre point.
• No more cross-cluster aliasing. The slot allocator runs a global collision check against every other placed cell in the workspace; two cells in different clusters can no longer hash to the same world coordinate.

Algorithm proof

tests/test_layout_algorithm.py runs all 14 simulator scenarios as pytest cases — including the user-reported nightmare cases (forest into corner, 20 random forest positions, nested collapse during outer drag, forest resize, floating cell survives collapse). Any algorithm regression fails CI.

What didn't change yet

• Drag-snap, collapse animation, live-edge-reflow, and master-drag cascade are still on the older code paths. The new model + the older mutations co-exist; the canonical layout function overrides positions during repack. Porting the rest is tracked in help/LLM/scenegraph_layout_plan.md and is the next focused session.
• No new periodic timers. The idle-CPU contract is preserved.

Tests

Full suite: 1741 passed.

v0.6.34 — revert layout races, instant startup, collapse safety belt

Why this revert ships

The v0.6.32 additions (overlap watchdog, drag cascade, sibling redock) introduced races that caused random cell movement, disappearing cells, and "click forest does nothing" — the layout system has too many independent helpers fighting for the same data. This release rolls those back to a known-working baseline and adds two narrow fixes so the app is usable while the proper scene-graph refactor is planned.

What changed

Reverts (v0.6.32 helpers removed)

• Overlap watchdog (_resettle_overlapping_neighbours) — was settling cells during collapse animations
• Cascade translate (_cascade_translate_positioned) — replaced with an inline self-contained shift inside the existing drag handler
• Sibling redock (_try_redock_to_sibling) — the re-link path was the cause of cells disappearing when the forest moved to a corner

Targeted fixes

• Instant startup layout — _repack_members(instant=True) at forest open so members snap onto canonical slots immediately; no 260 ms eased animation from stale-saved positions (which looked like jumble).
• Collapse safety watchdog — if a collapse / expand animation's finished signal is lost, a 1 s QTimer.singleShot force-clears the state so the next click isn't ignored by the animation-in-flight guard.

Looking forward

The underlying problem is the data model: cell positions are stored as absolute screen coordinates and seven independent helpers mutate them. This release stops the bleeding; the real fix is a slot-based scene graph (each cell stores a slot index relative to its parent; one layout tick replaces all the scattered helpers). Full plan in help/LLM/scenegraph_layout_plan.md.

Tests

1724 passed.