Skip to content

ExtUtilsSlimMigration.md

Latest commit

 

History

History
128 lines (111 loc) · 8.2 KB

ExtUtilsSlimMigration.md

File metadata and controls

128 lines (111 loc) · 8.2 KB
Raw
status landed
summary Move the toolkit's tools off the full ExtUtils clone (NoNode, its 11 exec DATs, Keyboard In, announcer) onto the slim ExtUtilsMinimal clone, each command owner announcing itself. NoNode is wired in none of them. The full master stays for QuickExt users.
since 2026-09-14
skill fns-command-registration

Moving tools to ExtUtilsMinimal

Owner's decision, 2026-09-14: tools get the slim clone (FNS_CustomParTools/QuickExt/ExtUtilsMinimal) and announce their own commands, the shape FNS_BeatMod and FNS_Remote already have.

What was measured before deciding

  • 57 full ExtUtils instances, not "a few". 46 are toolkit tools or their FNS_About boxes; the rest are /project1 test components, TDFam, and Private Investigator's release manager.
  • NoNode is wired in none of the 33 toolkit roots that carry it: no CHOP, DAT, parameter or keyboard exec is targeted, and no tool source calls its API. It is dead weight inside the toolkit.
  • The full clone's two live duties are the announcer (auto-registers a tool's commands 60 frames after init) and extAutoInit (touches .extensions so a COMP nothing pulls on still constructs its extension, and with it its CustomParHelper callbacks). The slim master has neither.
  • Only one tool reaches into a full-only child by path: FNS_Hub/HubExt.py calls ExtUtils/FNSCommandAnnouncer.
  • Baseline before any change: 151 commands from 47 owners (op.FNS_COMMANDREGISTRY.Commands()).

Not in scope

  • The full master QuickExt/ExtUtils stays exactly as it is: QuickExt ships it as the template for users' own extensions, and NoNode is user-facing API there (CoreToolsBacklog.md item 2).
  • /project1/* test components, /TDFam_create, /private_investigator1.
  • FNS_BackupCleaner (a non-clone copy with no FNSCommand) is listed separately below; it is handled after the fleet, on its own.

The plan

  1. Give the slim master extAutoInit. Copy the full master's Execute DAT into ExtUtilsMinimal, so every slim clone keeps the lazy-init guarantee. Only the slim master's clones inherit it (five today: FNS_BeatMod's and FNS_Remote's).
  2. Pilot one tool end to end (FNS_QuickCollapse), and record here anything it teaches before touching the rest.
  3. Per command owner: the extension announces itself from a deferred onInitTD through FNSCommand.announce(self.ownerComp) (merged into an existing onInitTD where there is one). Then the tool's ExtUtils is pointed at the slim master, children re-synced, docks repaired.
  4. Per FNS_About box: clone swap only (no commands).
  5. FNS_Hub: replace its ExtUtils/FNSCommandAnnouncer call with the extension's own announce.
  6. Verify per tool: both error kinds (get_op_errors and scriptErrors(recurse=True)), NoNode gone, the tool's command ids equal to the baseline, a forced reinit still announces.
  7. Save: PI Save() every owning suspect, nested suspects first (HideTimeline, BorderlessWindow, ClearPars, QuickParent, QuickParCustom, QuickExt, TDTypings, ScriptSyncFile, MIDIResetPLS), then their parents, /FNSTools root last. Build the list from owners, not PI's Dirty column (clone-driven changes do not trip it).
  8. Fleet check: distributor survey(), total commands back to 151 from 47 owners.

Traps to respect (from fns-command-registration)

  • Batch text surgery: AST-parse every edited extension and diff its def names against git HEAD.
  • Cloning does not carry docks: repair them after every re-sync.
  • Extensions that are not file-backed (FNS_Misc, mapTables, FNS_Output, FNS_ResetPLS, FNS_QuickTime, FNS_HydroHomie, FNS_OpMenuMods, FNS_VSCodeTools, FNS_GlobalVolControl) are edited in TD and saved through their tox.

Inventory

Command owners (root or nested tool): FNS_Hub, FNS_Misc, mapTables, FNS_Output, QuickMarks, FNS_AutoRes, FNS_ColorUI, FNS_OpenExt, FNS_SwapOps, FNS_Updater, FNS_ResetPLS, FNS_AltSelect, FNS_ParOPDrop, FNS_QuickPane, FNS_QuickTime, FNS_SwitchOPs, FNS_HydroHomie, FNS_OpMenuMods, FNS_AutoCombine, FNS_OpTemplates, FNS_VSCodeTools, FNS_BorderlessTD/HideTimeline, FNS_BorderlessTD/BorderlessWindow, FNS_PreviewPanel, FNS_HotkeyManager, FNS_OpToClipboard, FNS_ParRandomizer, FNS_QuickCollapse, FNS_SetSmoothness, FNS_CustomParTools, FNS_CustomParTools/ClearPars, FNS_CustomParTools/QuickParent, FNS_CustomParTools/QuickParCustom, FNS_ExprHotStrings, PasteFromClipboard, FNS_GlobalVolControl.

No commands (clone swap only): FNS_CommandPalette, midiMapper/MIDIResetPLS, and 10 FNS_About boxes (VSCodeTools/TDTypings, VSCodeTools/ScriptSyncFile, BorderlessTD/HideTimeline, BorderlessTD/BorderlessWindow, CustomParTools/QuickExt, ClearPars, QuickParent, iopPromoter, QuickParCustom), plus FNS_BackupCleaner's pair (not clones; last).

Progress

Step State
1 slim master gets extAutoInit done: copied into ExtUtilsMinimal; its five existing clones picked it up only after an explicit enablecloningpulse
2 pilot FNS_QuickCollapse done: it already announced itself from onInitTD, so only the clone swap was new; the swap removed every full-only child, kept the host dock, and after a reinit its two commands came back from its own announce
3 self-announce code done: 25 file-backed extensions and 9 in-tox extensions got onInitTD + _announceCommands through one checked transform (AST parse, def-name set equal to HEAD plus exactly the two new methods); FNS_Hub's ExtUtils/FNSCommandAnnouncer call now calls FNSCommand.announce
4-5 clone swap done: 46 instances (35 tools and nested tools, 10 FNS_About boxes, FNS_CommandPalette) plus the pilot
6 verify done: every command owner unregistered, extension reinitialised, and the registry compared to the baseline: 151 commands from 47 owners, none missing, none extra; no script errors under /FNSTools
7 save done: 46 owning suspect toxes, nested first, /FNSTools root, then SaveDirty() empty
8 fleet check survey: full targets 65 -> 18 (only /project1, TDFam, PI and FNS_BackupCleaner remain); still unhealthy for the same four non-clone copies it flagged before this work
cold boot done 2026-09-14: project saved, TouchDesigner restarted. Every migrated tool announced itself at boot and all extensions were ready, with no script errors; the registry held 148 commands from 46 owners. The three missing were the Installer's (available, install, installed), unrelated to this work: the dev rail /FNSTools/FNS_Installer reloads from a tox saved 2026-09-10 that embeds an InstallerExt older than its command registration (55 KB against the 92 KB source), and the baseline only listed them because the v3.2.11 release's rail rebuild had re-embedded the source live. EnsureDevRails() refreshed it, it registered (151 / 47), and its tox was saved
FNS_BackupCleaner done (owner's call): both non-clone full copies (tool and FNS_About) now clone the slim master; extStubser (a legacy stubs COMP nothing referenced) and NoNode went with the swap; no commands, so no announce code; extension reinitialised clean, CustomParHelper properties resolve; tox and root saved. Survey now flags only TDFam's two copies, outside FNSTools

What the run taught

  • Adding a child to a clone master does not reach existing clones until a clone pulse (par.enablecloningpulse.pulse()).
  • Clone-immune children survive a clone swap. FNS_HydroHomie's full-only exec DATs were cloneImmune (left from its cook-diet work) and stayed behind, erroring once NoNode was gone. Check every swapped instance against the master's child names, not just for errors, and destroy the leftovers.
  • Destroying a host destroys the DATs docked to it. Removing extKeyboardIn took extKeyboardin_callbacks with it mid-loop; iterate over names and re-resolve each op.
  • 46 PI saves in one call outlast the 30 s MCP timeout. The call keeps running on the main thread; wait for the tox count in git status to settle instead of re-running it.
  • extutils_distributor.rollout(apply=True) clones anything still carrying NoNode to the FULL master. Every toolkit instance is slim now, so a rollout leaves them alone; a new tool built from QuickExt starts on the full clone and must be swapped the same way.