a bare or double-quoted String literal in a CHANGE / FILTER expression passes `check --references`, writes, and fails `mx check` with CE0117

[marclehane-creator]

Eval row 17 (Change AND Filter, two errors) and row 33 (Change, fixed in-loop). Isolated on
probe/p3a-bare.mdl (Status = Open, Status = Closed) and probe/p3b-dquote.mdl
(Status = "Open", Status = "Closed"):

CREATE MICROFLOW Bench.Probe_BareLiteral ($Customer: Bench.Customer) RETURNS void BEGIN
  RETRIEVE $AllOrders FROM $Customer/Bench.Order_Customer;
  $OpenOrders = FILTER($AllOrders, Status = Open);
  LOOP $Order IN $OpenOrders BEGIN
    CHANGE $Order (Status = Closed);
    COMMIT $Order;
  END LOOP;
END;

Both pass rung 2 (0 errors, 0 warnings, 1 info — the MDL067 commit note) and both exec. Row 17's
compiler output for the same shapes: [CE0117] "Error(s) in expression." at Change object activity 'Change 'Order' (Status)' and … at List operation activity 'Filter'. In a Mendix expression
Open is an unresolvable identifier and "Open" is not a string literal; check already parses the
expression (Issue 80 proves it re-parses) and knows Status is a String attribute, so both are
detectable at rung 2 — or exec could quote a bare identifier that matches no variable/attribute
when the target attribute is a String, as Studio Pro's editor would refuse to save it.

Issues 81 and 82 — RETRACTION RETRACTED 2026-09-04 19:50Z: both REPRODUCE on a fresh copy; the earlier "NOT REPRODUCED" came from a BLIND probe copy

The 17:14Z/17:18Z notes above said Probe_LogObjectParam, Probe_BareLiteral and Probe_DoubleQuotedLiteral
compiled clean. They were compiled on the isolation probe, a fixture copy that had been exec'd about
fifteen times by then. At 19:48Z a control flow containing RETURN $X + $DoesNotExist; ALSO compiled clean
on that copy — and produced [CE0109] "Undefined variable 'DoesNotExist'." on a fresh copy. The copy had
stopped reporting MICROFLOW errors while still reporting PAGE errors (the CE1613s above are real). Cause
unknown; every microflow verdict from that copy is void.

Re-measured on a fresh copy (the isolation probe), four probes exec'd, ONE compile:

probe	statement	mx check
p2	LOG WARNING NODE 'Orders' 'Customer {1} has no open orders' WITH ({1} = $Customer) (object)	[CE0117] "Error(s) in expression." at Log message activity 'Log message (warning)'
p9	LOG INFO NODE 'Archive' 'Archived {1} orders.' WITH ({1} = $Count) (Integer)	[CE0117] … at Log message activity 'Log message (info)'
p3a	FILTER($AllOrders, Status = Open) / CHANGE $Order (Status = Closed)	[CE0117] … at List operation activity 'Filter' + … at Change object activity 'Change 'Order' (Status)'
p3b	the same with "Open" / "Closed" (normalised to bare by the writer)	the same two CE0117s

So Issue 81 is broader than written: the WITH ({1} = …) template-parameter form fails the compiler for
EVERY parameter type at 0.20.0 — the writer's template/parameter emission is wrong, not the caller's type.
Real app row 62 hit it live with an Integer ({1} = $Count) and, steered by a decision line that blamed
the type, DROPPED the count from the message to get to 0 errors. Issue 82 stands as written (bare and
double-quoted literals both pass check --references, both fail the compiler). Both are ready to file.

Method note for this register: a probe copy that has been written many times is not the fixture any
more. Every probe session from here starts on a fresh copy with a deliberately broken control of the
same unit type; exec counts on a copy are recorded with its verdicts.

---

mxcli 0.20.0, Mendix 11.12.x, Windows 11. Found while evaluating a self-hosted agent loop that drives check → check --references → exec → mx check; each item above was reproduced on a fresh copy of the fixture or app named.