#!/bin/sh # beadprompt.sh — emit the swarm prompt for a bead and copy it to the clipboard. # # ./beadprompt.sh wwwdb-i7e one bead # ./beadprompt.sh next HANDABLE bead in BATCH, and show the queue # ./beadprompt.sh --all print every BATCH prompt, copy nothing # # Edit BATCH to change the queue. Requires bd on PATH; pbcopy is optional. # # NOTE: this file is in .gitignore — git will not bring it back for you. It has # been lost once already. If you are about to delete it, copy it somewhere first. # # The scheduling lists below came from beadprompt-next.sh, which was a # throwaway for one batch and is now folded back in — a batch almost always has # at least one bead that cannot run beside its neighbour, and hand-remembering # which is how two agents end up in one file. # # SOLO one bead per line. Never handed out while anything else in BATCH is # claimed, and claiming it makes everything else unhandable. # SEQ "child:parent". The child is not handed out until the parent CLOSES. # EXCL "A:B". Symmetric, no ordering implied — shared files. A is held while # B is claimed and B while A is. # LAST held until every other bead in BATCH is CLOSED. For a bead that # changes a GATE everyone else has to pass. BATCH="wwwdb-epj.9 wwwdb-epj.7 wwwdb-3jq.1 wwwdb-yoj wwwdb-epj.10 wwwdb-epj.8 wwwdb-8cx.5 wwwdb-epj.1 wwwdb-e5tk" SOLO="" SEQ="wwwdb-epj.1:wwwdb-8cx.5" EXCL="wwwdb-epj.9:wwwdb-epj.7 wwwdb-3jq.1:wwwdb-yoj wwwdb-epj.10:wwwdb-epj.8" LAST="wwwdb-e5tk" # # THE REACHABILITY ROUND. # # The previous batch closed nine beads and did NOT close wwwdb-epj, which it # was framed as closing: it filed four new children on the way through and the # epic stands at 14/20. Read why before picking anything up, because it is the # shape of this batch. Three separate agents ported code that is CORRECT and # that nothing can reach: # # * wwwdb-qhn built sortVisualSides but not FitTexturesForm, so Ctrl+F fires # once at fixed defaults where UDB always shows a live dialog -> epj.7. # * wwwdb-960 built the rotation and scale actions with no key binding at # all -> epj.9. # * wwwdb-epj.2 built brightenExtraCeiling, which nothing can aim at because # pick.js cannot pick a 3D floor -> 3jq.1. # # That is not three coincidences, it is a standing hazard of a port organised # by UDB class: the class gets ported, the thing that CALLS it in UDB is a # different class in a different file, and it does not. If you finish your # bead and your code has no caller, you have not finished -- file the wiring # and say so in the close reason, the way 960 did. # # epj.9 P2, epic epj. Wire rotation/scale into visualpane.js and view.html. # Files: visualpane.js, view.html. # epj.7 P2, epic epj. FitTexturesForm itself, live-updating over qhn's # sortVisualSides. Files: dialogs.js, visualpane.js, and a DERIVED # table (DIALOG_LAYOUT) -- re-dump it, never hand-merge it. # 3jq.1 P2, epic 3jq. Filed 2026-09-14 off epj.2's close. pick.js emits two # flat candidates per sector where UDB enumerates FixedGeometry. # Files: pick.js, visualpane.js (the pickObject call at :1523). # yoj P3, epic epj. The VERTEX and SLOPE-HANDLE selection categories. # Geometry FIRST, selection second -- nothing builds a # VisualVertexHandle or VisualSlopeHandle at all. Files: # visualgeometry.js, visualselect.js, pick.js. # epj.10 P2, epic epj. Thing insert and delete. The one REMOVAL bead this # round. Files: visualthing.js, topology.js. # epj.8 P2, epic epj. BaseVisualThing.OnChangeScale. Files: visualthing.js. # 8cx.5 P2, epic 8cx. Filed 2026-09-14. Thread a parsed MapInfo into the # data layer -- the WIRING, not the parser and not the rest of # DataManager. Files: resources.js, mapinfo.js, visualpane.js. # epj.1 P2, epic epj. SEQ behind 8cx.5. UpdateLightFogFlag. Held for two # batches because it was pinned behind wwwdb-9w3 (P4); re-pointed # 2026-09-14 at 8cx.5, which is the only thing it actually needed. # Files: visualedit.js. # e5tk P2. LAST. Rolls citecheck's PORT_CLAIMS across the rest of the # tree. Files: citecheck.js. # # The conflict map: # # epj.9 and epj.7 are an EXCL pair. Both rework the SAME region of # visualpane.js -- epj.9 adds key bindings beside the Ctrl+F handler, epj.7 # replaces what that handler does. visualpane.js is a declared shared APPEND # point and that exemption does NOT cover two agents editing one region. # # 3jq.1 and yoj are an EXCL pair, and the reason is precise: the slope-handle # pick branches yoj needs (VisualMode.cs:919-926) are inside the SAME # candidate loop 3jq.1 is rewriting (:913-918). Same function, not just the # same file. # # epj.10 and epj.8 are an EXCL pair -- both are BaseVisualThing, both land in # visualthing.js. # # Everything else runs concurrently. 3jq.1 and yoj both touch pick.js while # epj.9/epj.7 touch visualpane.js in a different region; that is deliberate, # not an oversight. Four agents at once is the intent. # # e5tk is LAST because it changes a GATE every other bead in this batch has # to pass, in a file every one of them appends to. The slot has been empty # for two batches and this is what it was built for. # # Standing hazard carried forward from the 2026-09-14 review: # # The previous round's commits were made in one pass at the end and TWO went # wrong. wwwdb-epj.6's tests were committed while its implementation was not # (HEAD ran 1182/1187 red until it was fixed up), and wwwdb-1sh got no commit # of its own at all -- its whole port rode in under epj.2's message. Neither # is a porting defect and both were invisible to a green working tree, # because the working tree HAD everything. If you are the one committing this # round, commit per bead and run the gates from a CLEAN CHECKOUT of what you # committed, not from the tree you were working in. DIR="$HOME/Desktop/Work/WWWDB/udmf-editor-foundation" extra() { case "$1" in wwwdb-epj.9) cat <<'EOF' SPECIFIC TO THIS BEAD wwwdb-960 ported the actions and could not reach them. You are the caller, and the bindings are DATA, not a design problem — Actions.cfg has them and this project already derives that file: scaleup 105 (Num 9) scaledown 103 (Num 7) scaleupx 102 scaledownx 100 scaleupy 104 scaledowny 101 rotateclockwise 262138 (Ctrl + Shift + ScrollUp) rotatecounterclockwise 262139 (Ctrl + Shift + ScrollDown) Read those out of Actions.cfg yourself rather than trusting this list, and if they belong in a derived table, DUMP them — do not hand-type eight rows. 960's close reason worried that "this pane's wheel is already claimed by FOV zoom". Check that before you design around it: rotate is Ctrl+Shift+Scroll, a MODIFIED wheel, and visualpane.js:274's FOV handler takes the PLAIN wheel. If that is right, there is no conflict to resolve and nothing to invent. If the handler swallows modified wheel events too, that is the bug to fix — not a reason to pick a different key than UDB's. The six scale actions all dispatch through ONE private ScaleTexture(incrementx, incrementy) (BaseVisualMode.cs:3486-3498), which opens PreAction(UndoGroup.TextureScaleChange) and then loops OnChangeScale over the selected objects. Port that dispatcher shape; six independent handlers that each open their own undo is drift, and the undo GROUP is the part that shows. The fixture trap: a binding test that asserts "the handler was called" passes whether or not the right increment reached it. Assert on the (incrementX, incrementY) pair each action produces — six actions, six distinct pairs, and the x/y-only ones are where a transposition hides. You share visualpane.js with wwwdb-epj.7, which is held while you are claimed. EOF ;; wwwdb-epj.7) cat <<'EOF' SPECIFIC TO THIS BEAD The engine is done and you must not rebuild it. wwwdb-qhn landed sortVisualSides/sidedefVertexBounds/fitSortedSide in visualedit.js and wired Ctrl+F to run them at DEFAULT_FIT_OPTIONS' fixed values. You are adding the DIALOG that UDB always shows for visualfittextures (BaseVisualMode.cs :3659-3687) and that live-updates on every control change (UpdateChanges, FitTexturesForm.cs:120-148). If you find yourself computing a strip or a bound, stop — that is qhn's code and you should be calling it. Do not invent a layout. This is a two-column dialog and the project has a derived DIALOG_LAYOUT with a checker: re-dump it from the designer geometry, never hand-merge it, and run dialogparity.js --layout and --dialog. The ui-conventions memory and the dialogs already ported are the pattern. Two behaviours that are easy to miss and are the whole point of the dialog: "Fit across connected surfaces" is ENABLED only when strips.Count > 1 (FitTexturesForm.cs:107) and ships CHECKED (:50-54) — qhn already flipped DEFAULT_FIT_OPTIONS.fitAcrossSurfaces to true to match, so do not flip it back. And UpdateChanges re-runs the fit on EVERY control change against the ORIGINAL offsets, not cumulatively against the last preview. A dialog that applies its delta on top of its own previous preview drifts every keystroke, and a fixture that changes one control once cannot tell the two apart — change a control THREE times and assert the result equals the one-shot answer. Modals must not leak events to the map: modalOpen() and typingInField() gate every mouse and key handler. Any new global handler needs both. You share visualpane.js with wwwdb-epj.9, which is held while you are claimed. EOF ;; wwwdb-3jq.1) cat <<'EOF' SPECIFIC TO THIS BEAD Filed 2026-09-14 out of wwwdb-epj.2's close, and it is the reason three rounds of 3D-floor work are still unreachable: pick.js pushes exactly TWO flat candidates per sector (:346-347 for the camera's own sector, :362-363 for every sector the ray touches) where UDB enumerates vs.FixedGeometry — every entry with Triangles > 0 (VisualMode.cs:913-918). BaseVisualSector.Rebuild says exactly what is in that list, and it is more than two: the plain floor (:217), the plain ceiling (:221), then PER EXTRA FLOOR a VisualFloor (:256) and, when RenderInside, its backside (:262); per extra ceiling a VisualCeiling (:294) and its backside (:300). All through the same base.AddGeometry. wwwdb-phs built the stack and wwwdb-4p1x built the backside pass, so the geometry ALREADY EXISTS in visualgeometry.js — you are not building surfaces, you are enumerating the ones that are there. Port the ENUMERATION, not "two plus extra floors". If your candidate list reads as the old pair with a loop bolted beside it, you have written the special case again — the plain floor and ceiling are the degenerate SectorLevel stack, which is exactly the framing wwwdb-phs was given and met. Not yours, and in the same loop: the pickingmode slope-handle branches (VisualMode.cs:919-926) belong to wwwdb-yoj, which is held while you are claimed for precisely that reason. Leave them alone; do not "while I am here". The fixture trap: a sector with no 3D floor produces the same two candidates either way and cannot tell a correct enumeration from today's hardcode. Build a sector with at least one extra floor AND one extra ceiling, aim a ray that crosses the extra floor's plane BEFORE the real ceiling, and assert on WHICH surface comes back — never on the candidate count. Prove the payoff: wwwdb-epj.2's brightenExtraCeiling is currently exercised only by direct call because nothing can pick an extra ceiling. If it is reachable when you are done, say so in the close reason. EOF ;; wwwdb-yoj) cat <<'EOF' SPECIFIC TO THIS BEAD Read the bead body before you plan: "both categories have no visual geometry here at all — nothing builds a VisualVertexHandle or a VisualSlopeHandle". This is GEOMETRY first and selection second, and the selection half really is small because visualselect.js already carries both through its filter and clear arguments. If you scope this as "adding a case", you will have written the selection half of a feature with nothing to select. The pick side is VisualMode.cs:919-926 — the pickingmode branches (sidedefslopehandles / vertexslopehandles) inside the candidate loop. That loop is being rewritten by wwwdb-3jq.1, which is held while you are claimed and which you are held behind: SAME function, not just the same file. Whichever of you goes second re-reads the loop on disk before patching it. Do not extend this into slope EDITING. The handles' selection and geometry are the bead; dragging one to change a plane is not, and if you find yourself in planes.js, file it. EOF ;; wwwdb-epj.10) cat <<'EOF' SPECIFIC TO THIS BEAD This is the REMOVAL bead this round, and removal in this document model RENUMBERS — UDMF references are positional. Everything that deletes goes through topology.js's disposal cascade and syncIndices, never a raw block splice. HANDOFF's "Two things to know before touching removal" is required reading before your first edit. Nothing else in this batch is near topology.js and that is deliberate; keep it that way. Note carefully that this is NOT shaped like wwwdb-1sh, which found OnInsert dead and OnDelete to be texture bookkeeping. On BaseVisualThing both halves are real, and they live in different places: * INSERT is not on the class at all — BaseVisualThing.OnInsert() is an empty body (BaseVisualThing.cs:672). The real action is BaseVisualMode. InsertThing (:3885-3916), bound to "insertitem": GetHitPosition(), a finite-check that DISPLAYS A WARNING and returns, ClearSelection, PreActionNoChange, ClearAllRedos, CreateUndo("Insert thing"), CreateThing, and a WithdrawUndo when CreateThing answers null. Port the withdraw — an undo level left open on a failed insert is exactly the kind of "cleaner" omission section 0 is about. * DELETE is BaseVisualThing.OnDelete (:699-709) and it is genuine disposal: Thing.Fields.BeforeFieldsChange(), Thing.Dispose(), this.Dispose(), then ThingsFilter.Update(). wwwdb-ahg.1 already made the filter a live consumer, so that last line has a real caller here. The fixture trap follows from the renumbering: a map where the deleted thing is the LAST one cannot tell a correct renumber from no renumber at all, because nothing shifts. Delete from the MIDDLE and assert on the references that had to move, not on the count. You share visualthing.js with wwwdb-epj.8, which is held while you are claimed. EOF ;; wwwdb-epj.8) cat <<'EOF' SPECIFIC TO THIS BEAD You are porting the SIBLING of something that landed last round, so read it first: wwwdb-960's changeWallTextureScale in visualedit.js is the same pixel-preserving formula BaseVisualThing.OnChangeScale (:819-850) uses — pix = round(dimension * scale) + increment newscale = round(pix / dimension, 3) newscale == 0 ? scale * -1 : newscale including the zero-to-sign-flip. Match that shape. A second convention for the same arithmetic in a second file is drift even when both are correct. Three things that are NOT in the wall version and are yours: * The gate is `!General.Map.UDMF || sprites == null || !sprites[0].IsImageLoaded` (:821). It needs a LOADED SPRITE, because the formula divides by the sprite's pixel dimensions. If sprite loading is not there yet, that is a dependency to FILE, not to guess a dimension around — a fallback of 1 would silently produce a wrong scale rather than no scale. * Undo COALESCING: `NextUndo == null || NextUndo.TicketID != undoticket` (:823-824) opens one undo level across a held repeat instead of one per increment. A fixture that calls it once cannot see this. Call it three times and assert ONE undo level. * It writes through Thing.SetScale and sets this.Changed, where the wall version writes UDMF fields directly. You share visualthing.js with wwwdb-epj.10, which is held while you are claimed. EOF ;; wwwdb-8cx.5) cat <<'EOF' SPECIFIC TO THIS BEAD Filed 2026-09-14. You are unblocking wwwdb-epj.1, which has been parked for two batches behind wwwdb-9w3 (P4, two thirds of DataManager's 3,882 lines) when all it ever needed was this. epj.1 is SEQ'd behind you in this batch, so a clean close here hands a second bead to the next agent. The scope is the WIRING and nothing else. The parser exists — wwwdb-aiu ported MapinfoParser + MapInfo into mapinfo.js from MapinfoParser.cs / MapInfo.cs, and wwwdb-kdv.7 already consumes its DoomEdNum/SpawnID tables. What does not exist is any path by which the editor HOLDS a parsed MapInfo for the open map. UDB's is General.Map.Data.MapInfo, set when the map loads. Find the MAPINFO/ZMAPINFO lump through the resource layer, parse it with the parser that is already there, hang the result where visual mode can read it, and answer with MapInfo's OWN documented defaults when the map has no such lump. Do not start consuming what you thread. The sky/fog/lightmode fields and the fade/outsidefog colours (mapinfo.js:95-97, :233-238) are parsed but unconsumed on purpose; consuming them is wwwdb-1n5's rendering work. Thread the OBJECT. The defaults are the whole risk, and the safety argument recorded on epj.1 on 2026-09-10 is why: a map with no MAPINFO must come back with MapInfo's real defaults, not with every term false. A downstream flag decision that reads a false it should not have would CLEAR a correct flag on a map authored in real UDB — worse than not touching it. So the discriminating fixture is a map with NO MAPINFO lump: assert the defaults, not merely that nothing threw. EOF ;; wwwdb-epj.1) cat <<'EOF' SPECIFIC TO THIS BEAD You are SEQ'd behind wwwdb-8cx.5 and must not start until it CLOSES. Read what it landed before you read your own bead body. This bead's description carried a factual error until 2026-09-13 and the correction is the reason it is P2 and not a footnote. It claimed the UpdateLightFogFlag call only fires for a plain `light` key, exempting the distinct-PART case. It does not: Tools.cs:2437's early-out fires only when NONE of light / light_top / light_mid / light_bottom is present, so the call fires for all four, and BaseVisualGeometrySidedef.cs:1595-1623 is one branch in which fieldname becomes light_. The DISTINCT-PART case is the most affected one. Build the fixture on a distinct part, not on a plain `light`. Of the three blockers once recorded, two are already gone: hasSkyCeiling exists (visualpane.js:93 = Sector.cs:109) and `lightfog` is in the field model (defaults.js:110). What is left is a sidedef SetFlag equivalent — and note wwwdb-vse added exactly such a helper for the unpegged toggles last round (setLineFlag, which omits a false rather than storing it, matching UniversalStreamWriter.cs:235). Read it before you write a second one. The safety rule still governs: do NOT run this against MapInfo terms you had to assume. If 8cx.5's defaults are not actually reaching you, stop and say so — clearing a correct lightfog flag on a map authored in real UDB is worse than leaving the flag alone. EOF ;; wwwdb-e5tk) cat <<'EOF' SPECIFIC TO THIS BEAD You are LAST: every other bead in this batch is CLOSED before you are handed out. That is not politeness, it is the point — you are changing a GATE all of them had to pass, in a file every one of them appended a claim to. Run citecheck BEFORE your first edit and write the number down, so a count that moves later is attributable. wwwdb-o0c landed Check 3 as a PILOT, deliberately scoped to three known instances. This is the ROLLOUT, and there are two classes: Class A — PORT drift. PORT_CLAIMS has exactly ONE entry (stitch.js's second SplitLinesByVertices call, MapSet.cs:2223). Every ported call site that passes an argument, flag or condition that CLAIMS already pins on the REFERENCE side is a candidate that has never been checked against the port. Class B — unregistered inline claim: prose that quotes what a cited line says but was never promoted into CLAIMS, so only the cheap in-range check ever ran. Audit every MODULE in citecheck.js's MODULES list, not just selftest.js. And audit the MODULES list itself first: a module that is not in it is not partly checked, it is UNCHECKED, and the suite stays green either way. That has hidden a bad citation three times in this project, and wwwdb-1ze was filed because the list was missing selftest.js and citecheck.js themselves. The last round is fresh material and the most likely place to find Class A: nine beads landed ports across visualedit.js, visualgeometry.js, sectorlevels.js, visualautoalign.js, visualselect.js and visualpane.js, several of them widening code an EARLIER bead wrote against a collapsed assumption (epj.3's controlSide === sidedef is the worked example). Start there. A found instance is a RESULT, not a failure — report the count either way. If the sweep finds nothing in a module, say which modules you cleared, so the next person does not redo it. EOF ;; esac } emit() { case " $SOLO " in *" $1 "*) _co="You are running SOLO: no other agent should be working this tree while you are. The shared-file cautions below still apply to what you LEAVE behind." ;; *) _co="Other agents are working other beads in this same tree at the same time." ;; esac cat <