refactor implementation of END indexing in interpreter (bug #58953) * oct-lvalue.h, oct-lvalue.cc (octave_lvalue::eval_for_numel): New function. (octave_lvalue::m_nel): Delete data member. (octave_lvalue::numel (octave_idx_type)): Delete. (octave_lvalue::numel (void) const): Compute result instead of returning cached value. * pt-eval.h, pt-eval.cc (tree_evaluator::convert_to_const_vector): Simplify. (Fend): Call tree_evaluator::evaluate_end_expression to do the real work. (tree_evaluator::make_value_list): Don't check for magic_end here. (end_value, tree_evaluator::evaluate_end_expression): New functions. (tree_evaluator::m_index_list, tree_evaluator::m_index_type): New data members. (tree_evaluator::set_indexed_object, tree_evaluator::index_list, tree_evaluator::set_index_list, tree_evaluator::clear_index_list, tree_evaluator::append_index_list, tree_evaluator::index_type): New functions. * pt-idx.cc (make_value_list): Delete. (tree_index_expression::lvalue, tree_index_expression::evaluate_n): Refactor to avoid evaluating partial expressions to obtain values for the END function. Instead, cache info necessary to do that job as needed and inside the END function itself.

--- a/test/struct.tst	Sun Sep 27 12:24:24 2020 -0400
+++ b/test/struct.tst	Mon Sep 21 10:45:18 2020 -0400
@@ -155,9 +155,13 @@
 %! assert ({b.name}, {1, 2, "b", "e", 3, 4});
 %! assert ({b.value}, {100, "5", 100, "6", 100, "7"});
 
+%!error <deal: nargin>
+%! [a(1:3).x] = deal ([1, 5], [3, 7], [8, 9]);
+%! [a(2:3).x(2)] = deal (10, 11);
+
 %!error <a cs-list cannot be further indexed>
 %! [a(1:3).x] = deal ([1, 5], [3, 7], [8, 9]);
-%! [a(2:3).x(2)] = deal (10, 11);
+%! [a(2:3).x(2)] = 1;
 
 %!error <a cs-list cannot be further indexed>
 %! [a(1:3).x] = deal ([1, 5], [3, 7], [8, 9]);