[ako]

Fixes case 6 of #1508 (mirror of mendixlabs#1598). Stacked on #1528 (fix/flow-graph-geometry, cases 1–5; this builds on its Pinned handling). Base is that branch; retarget to main once #1528 merges.

The bug

create or modify of a nanoflow drawn with @position on every statement, adding an if/else after the declare without @position and editing the calls in the next if's branch:

Modified nanoflow: MyFirstModule.NF_Do (spliced: 1 inserted, 1 replaced, 1 dropped)
@position(40, 200)     if $Flag        was -299  (+339)
@merge(525, 200)       its merge       was 186   (+339)
@position(-124, 100)   call javascript action ... 'b'   (unchanged — left of its own decision)
@position(60, 100)     $Ok = call microflow             (unchanged)
@position(869, 200)    return          was 530   (+339)

Re-running the same script was refused (the connectors of the If at (40, 200) are redrawn). With a plain if (no else) the new if took over the stored decision and the restated if $Flag was drawn on top of it, both merges at (186, 200).

Cause, confirmed from source and reproduced on PedApp:

• makeRoom shifts every top-level node past the cut by the deficit (339);
• the branch Replace states @position, so it is Placed and never shifted;
• movedNode only records a move when stated ≠ stored, so the shifted nodes were never moved back (the merge is not a move candidate at all);
• no-else: lcsStatements weighed ifConditionChanged the same as sameIfShell; the tie went to the earlier pair.

The decision (option a, chosen by the maintainer)

Shifting consistently cannot make the re-run report Unchanged while the script still states the old positions — measured even for a plain one-activity insert into a described flow (+104, then the re-run is refused on @merge). So:

1. Under mdl 1, a node at a stated position is pinned. The executor collects every point the script states a node at — @position, @merge, @start, note position:, parameter @position; loop bodies skipped (relative coordinates) — and the mutator pins whatever is drawn there (MicroflowMutator.Pin). A make-room shift (insert or longer replace) that would move a pinned node is refused, naming the room and the node:

   the inserted statements need 339px more room than there is between the ActionActivity at (-500, 200) and the ExclusiveSplit '$Flag' at (-299, 200), which the script draws where it is; state @position on the new statements, where there is room for them, or leave unstated the positions of what comes after them (@position, @merge, a note's position:) so that it can move along

   Pinning is by point, not $ID, so a node a later Placed fragment puts at its stated position is pinned too.

   mdl 0 is unchanged (ADR-0011). The option as proposed let mdl 0 fall back to the whole-flow rebuild, like every splice refusal. TestFlowModify_InsertIsSpliced (ADR-0012's describe → insert → exec acceptance test, headerless) caught what that means: a headerless describe+insert that used to splice became a rebuild that deletes the Studio Pro merges and $IDs. ADR-0011 says a new refusal applies only under the version that introduces it, so the pins are gated on flowRebuildRefused.Applies; a headerless script makes room exactly as before.

2. LCS weights split: exact 4, position-only / same if shell / editable loop 2, condition changed 1 (all existing ratios kept; only the condition-changed pair drops). The restated if $Flag now pairs with the stored one; the new if is an insert, refused like the else variant instead of stacking.

alter microflow is unaffected (it states no kept positions, so nothing is pinned).

Evidence

• TestSplice_MakingRoomRefusesToMoveAPinnedNode (mfmutator): insert / longer replace / earlier placed fragment refused; controls (nothing pinned → b moves 420→570; pins only before the gap → accepted).
• TestLCSStatements_TheSameConditionWinsOverAChangedOne.
• TestCreateOrModify_InsertDoesNotMoveStatedNodes1508 (PedApp, modelsdk backend): both reported variants refused and nothing written; control = the same change with positions after the insert left unstated → spliced, both calls between if $Flag and its merge, re-run Unchanged nanoflow.

Control runs (fix stubbed, tests run):

# pin guard stubbed (pinnedPast returns nil):
--- FAIL: TestSplice_MakingRoomRefusesToMoveAPinnedNode (0.00s)
    --- FAIL: TestSplice_MakingRoomRefusesToMoveAPinnedNode/an_insert,_the_node_after_the_gap_pinned (0.00s)
        splice_pin_test.go:41: got <nil>, want a refusal naming the pinned activity at (420, 200) and the 150px lacking
    --- FAIL: TestSplice_MakingRoomRefusesToMoveAPinnedNode/a_replace_longer_than_what_it_replaces,_the_end_pinned (0.00s)
        splice_pin_test.go:74: got <nil>, want a refusal naming the pinned end event
    --- FAIL: TestSplice_MakingRoomRefusesToMoveAPinnedNode/an_earlier_fragment_drawn_at_a_stated_point (0.00s)
        splice_pin_test.go:92: got <nil>, want a refusal naming the placed activity at (600, 400)
FAIL
FAIL	github.com/mendixlabs/mxcli/mdl/backend/mfmutator	0.004s
--- FAIL: TestCreateOrModify_InsertDoesNotMoveStatedNodes1508 (3.11s)
    --- FAIL: TestCreateOrModify_InsertDoesNotMoveStatedNodes1508/the_reported_if/else (2.15s)
        flow_modify_stated_positions_1508_test.go:114: want a refusal naming the decision the script draws at (-299, 200), got <nil>
        flow_modify_stated_positions_1508_test.go:117: the refused statement wrote the nanoflow:
              @position(40, 200)
              @merge(525, 200)
                @position(-124, 100)
              @position(869, 200)
    --- FAIL: TestCreateOrModify_InsertDoesNotMoveStatedNodes1508/a_plain_if (0.50s)
        flow_modify_stated_positions_1508_test.go:114: want a refusal naming the decision the script draws at (-299, 200), got <nil>
        flow_modify_stated_positions_1508_test.go:117: the refused statement wrote the nanoflow:
              @position(40, 200)
              @merge(525, 200)
                @position(-124, 100)
              @position(869, 200)
FAIL
FAIL	github.com/mendixlabs/mxcli/mdl/executor	3.150s
FAIL

# LCS split reverted (condition changed weighs 2 again):
--- FAIL: TestLCSStatements_TheSameConditionWinsOverAChangedOne (0.08s)
    flow_declared_match_test.go:199: pairs [[0 0] [1 1]], want [[0 0] [2 1]]: the restated if paired with the stored one, the new if inserted
--- FAIL: TestCreateOrModify_InsertDoesNotMoveStatedNodes1508 (7.28s)
    --- FAIL: TestCreateOrModify_InsertDoesNotMoveStatedNodes1508/a_plain_if (2.12s)
        flow_modify_stated_positions_1508_test.go:114: want a refusal naming the decision the script draws at (-299, 200), got <nil>
        flow_modify_stated_positions_1508_test.go:117: the refused statement wrote the nanoflow:
              @position(-500, 200)
              @position(-299, 200)
              @merge(186, 200)
              if $Out then
                @position(-124, 100)
              @position(-299, 200)
              @merge(186, 200)
              if $Flag then
                @position(-124, 100)
                @position(60, 100)
              @position(530, 200)
FAIL
FAIL	github.com/mendixlabs/mxcli/mdl/executor	7.434s
FAIL

Real binary on a PedApp copy (mdl 1): both variants refused with the message above; the free variant → Modified … (spliced: 1 inserted, 1 replaced, 1 dropped), describe shows decision 40 / calls 215.. / merge 685 / return 1029, re-run → Unchanged nanoflow: MyFirstModule.NF_Do. mxcli docker check on the result: no error on NF_Do / JS_Do / SUB_Check (the 1078 errors are PedApp's baseline: CE6083/CE0462/CE0535 from its stripped theme and widgets). Headerless: spliced as before (test case "mdl 0 splices as before").

Gates

• make build, make test, make lint, make check-mdl, make check-findings: pass. (A first make test failed only on no space left on device: /tmp is a 12 GB tmpfs shared with other sessions; re-run with TMPDIR on the overlay disk, green.)
• Integration (-tags integration ./mdl/roundtrip): TestFlowRerunProperty (+ controls), TestPedAppFlowSpliceParity, TestPedAppSpliceParity_*, every TestFlowModify_* (incl. TestFlowModify_InsertIsSpliced), TestSpliceRerun_*, TestFlowVerdictAgreement_*: pass. Skipped here (no TestApp fixture in this checkout): TestTestAppFlowSpliceParity, TestFlowModify_TestAppHeaderAndPositions.

Review notes (/mxcli-dev:review)

No blockers. Two moderate points left open on purpose:

• No mdl 0 warning. A headerless exec of a statement mdl 1 would refuse splices silently (ADR-0011 asks headerless scripts to warn where meaning differs). check -p / fmt --upgrade -p under mdl 1 do report it, through the same decideFlowModify. A warning on plain mdl 0 exec would need a "spliced, but warn" verdict that exec, diff and check do not have yet.
• statedPoints names the position-bearing annotation fields by hand (Position, Merge, Start, note positions, parameters). A new one would have to be added there.

Also

• Bug-test: mdl-examples/bug-tests/flow-geometry-1508-splice-keeps-stated-positions.mdl
• Finding: .claude/skills/fix-issue/findings/mdl-backend/2026-10-21-splice-shift-leaves-placed-fragment-behind.json
• docs-wiki/bug-patterns/flow-graph-geometry.md: "making room is a move the script did not ask for"
• write-microflows (pitfall 13) / write-nanoflows skills: the new refusal and its two ways out

Not touched, noticed: the make-room shift also moves the parameters (y 53) along with the row ($Flag 200 → 539/699 here), so describe starts printing a parameter @position after any insert. Pre-existing; separate follow-up.

🤖 Generated with Claude Code