engine: SUM(pop[coast]) with coast a subrange the target does not range over is refused as "mismatched_dimensions -- pop" -- the diagnostic names the variable, not the subrange or the *:coast spelling that compiles

[bpowers]

Summary

With region = {north, south, east} and coast = {north, east} (a subrange of region), pop[region] a stock:

equation	result (identical on origin/main d04593e6 and branch compiler-unification-v2 HEAD)
x[region] = SUM(pop[coast])	refused: error in model 'main' variable 'x': mismatched_dimensions -- pop (snippet shows the whole SUM(pop[coast]), no span marker)
x[region] = SUM(pop[coast] * 2)	refused, same text
x = SUM(pop[coast]) (scalar equation)	refused: error in model 'main' variable 'x': dimension_in_scalar_context, span under coast
x[region] = SUM(pop[*:coast])	compiles, 400 400 400 (= 100 + 300)
x = SUM(pop[*:coast])	compiles, 400
x[coast] = SUM(pop[coast])	compiles: 100 300 on the branch (by name); 100 200 on origin/main (positional, GH #1029)

Exact text of the first row, from simlin simulate:

    SUM(pop[coast])
    
error in model 'main' variable 'x': mismatched_dimensions -- pop
error compiling model 'main': SimulationError{not_simulatable: failed to compile fragments for variables: x}

Is the refusal right?

Yes, as far as XMILE goes. Section 3.7.1 of docs/reference/xmile-v1.0.html makes a dimension name inside a subscript a placeholder for the enclosing element's index, and says "the array containing the equation MUST itself be sized using those dimension names". coast is not a dimension of x[region], so pop[coast] in that equation is not a valid placeholder, and the spec's slicing (3.7.1.2) is spelled with * and lo:hi ranges. Simlin's named-subrange slice is the star range pop[*:coast], which compiles in both equations, and the MDL importer spells Vensim's pop[coast!] as that StarRange (the writer round-trips it back to coast!, mdl/writer_tests.rs::star_range_subscript_renders_as_bang), so imported models do not reach this refusal.

What is wrong is what the refusal says

• mismatched_dimensions -- pop blames the variable. pop is fine; the offending token is the dimension name coast, which pairs with no axis of the target, and the message neither names it nor marks its span (the scalar-equation arm, dimension_in_scalar_context with the span under coast, already does better).
• Nothing points at the spelling that means what the modeller almost certainly meant (pop[*:coast], one token away), or says that a dimension name in a subscript is only a placeholder for the target's own dimensions.

A modeller coming from Vensim (SUM(pop[coast!])) or from a hand-written XMILE file hits this on the first subrange reduce they write and gets pointed at the wrong identifier.

Where to look

• The sim_err!(MismatchedDimensions, id.as_str().to_string()) arms in src/simlin-engine/src/compiler/context.rs's subscript lowering (the dimension-name index resolves to no active axis of the target); EquationError::details carries the reason, so the fix is to put the dimension name (and the *:coast suggestion when the named dimension is a subdimension of the source axis, DimensionsContext knows) in details and the span on the index.
• src/simlin-engine/src/ast/expr2.rs around the DimensionInScalarContext raise for the scalar-equation arm, which already spans the index and could carry the same suggestion.

Acceptance

• A row per {apply-to-all target over a different dimension, scalar target} x {subrange name, unrelated dimension name} pinning the message text and span through TestProject::error_diagnostics.
• SUM(pop[*:coast]) in both equations and x[coast] = SUM(pop[coast]) keep compiling with the values above.

Related

• engine: subrange dimension reference reads the parent array positionally (out[Sub] = src[Sub] is silently wrong) #1029 -- out[Sub] = src[Sub] reading positionally; the x[coast] = SUM(pop[coast]) row above measures it live on origin/main and by name on the branch since be7adbbf (Phase 6b, design-plan item "A subscript naming a dimension resolves through one rule").
• engine: an iterated-dimension subscript over an undeclared, disjoint-named axis reads by ordinal on the static subscript path and is refused on the dynamic one #1044 -- an iterated-dimension subscript over an undeclared, disjoint-named axis; the sibling refusal family.

Discovery context

Identified during PR #1040 (branch compiler-unification-v2): the V9b report's out-of-scope discovery that SUM(pop[coast]) (probe e10_subdim_reducer) "is refused on both binaries". Reproduced with simlin simulate on origin/main at d04593e6 and branch HEAD; pre-existing and unchanged by the branch.