New Surface Syntax And Parser

[zilinc]

Description

For some reason, we want to rewrite our surface syntax parser, and also take this opportunity to revise our surface syntax.

[gteege]

Just a contribution / proposal for an improved surface syntax: An expression with an "observed variable" <expr> !v is difficult to read, because the variable "banging" is at the end and v is always a valid variable from the context. Since <expr> may be arbitrary large and complex, it may be misleading when you first read <expr> and only realize afterwards that v has been made readonly. Even in the bilby fs code there are examples where <expr> has more than one line.

I propose the syntax let !v in <expr> instead (or additionally, to be upward compatible). Remarks:

• the rationale is that !v can be understood as introducing a new variable which is bound to the readonly value and shadows in its scope <expr> the existing variable with the linear value.
• the usual abbreviation using and could be applied to support several banged variables and also to mix them with normal bindings, such as in let a = 10 and !v and !w and b = v.m in ...
• syntactically let !v in <expr> would be a special form of the normal let-expression and may be used wherever a let-expression is permitted (or is that a problem for the core syntax?). This would even simplify the syntax (and the parser) by removing the special cases now present for banged variables.