Variadic addition and multiplication

[hudlow]

Even assuming that all subexpressions are without side effects, sequences of addition (and subtraction) or multiplication (and addition) are order-sensitive and therefore cannot be safely balanced¹. As a consequence, protobuf nesting limits result in a hard cap for the number of terms in a mathematical expressions and this cap also interacts with other structures such as nested parentheticals, function calls, and list/map/message construction. To get more control over #410 (in particular, to cap the number of terms in a mathematical expression consistently, without respect to other structures in the expression), I'd like to see support for variadic addition and multiplication².

Is this feasible? I am a little shaky on what the commitment to forward- and backward-compatibility is for the current AST definition, though I assume forward compatibility of existing evaluators is not a hard requirement, given the in-progress updates to comprehensions.

Footnotes

1. For example, 1e16 + -1e16 + 1.0 does not equal 1e16 + 1.0 + -1e16 and 1 + -1 + 9223372036854775807 does not equal 1 + 9223372036854775807 + -1 (the latter producing an overflow error). ↩

2. Because subtraction and division share precedence with addition and multiplication, respectively, there would need to be a single variadic function for addition and subtraction and a single variadic function for multiplication and division — otherwise, alternating the same-precedent operations would again produce a deep expression that could not be safely rebalanced. ↩