[GuntherRademacher]

The bindings of a closure were kept in a HashMap<Var, Expr> where Var does not override Object.hashCode(), so the order of the captured variables (and the variable numbering) in the query plan depended on identity hash codes and changed from run to run. Plans for the same query could therefore differ spuriously (in tests, bug reports, or when comparing parser implementations).

Example: run the following a few times with basex -x:

    let $a := <a/>, $b := <b/>, $c := <c/>, $d := <d/>
    return fn() { ($a, $b, $c, $d) }

The Closure lists its variables as $c $a $b $d in one run and $d $b $c $a in the next.

The maps are now LinkedHashMaps (insertion order), in the parser (LocalVars), in Closure (constructor, copy, inlining of nested closures) and in PartFunc. This preserves query semantics while making capture order deterministic. A test in FuncItemTest compiles five such queries 100 times each and compares the plans; it fails if any single one of these changes is reverted.

The declared types (HashMap<Var, Expr>) are left as they are to keep this change minimal. I can follow up with a refactoring to Map<Var, Expr> if desired.