17.8.1 Recordisation of bindings

The core language forms that define bindings are:

(library-letrec* ((?lhs ?loc ?rhs) ...) ?body)
(let     ((?lhs ?rhs) ...) ?body)
(letrec  ((?lhs ?rhs) ...) ?body)
(letrec* ((?lhs ?rhs) ...) ?body)
(lambda ?formals ?body)
(case-lambda (?formals ?body) ...)
(annotated-case-lambda ?annotation (?formals ?body) ...)

and as special case:

(set! ?lhs ?rhs)

when present at the top level. In all these expressions ?lhs is a lexical gensym uniquely identifying the binding.

The recordisation process transforms:

• library-letrec* and letrec* forms into rec*bind structs, in which defined lex gensyms are substituted with prelex structs.
• letrec forms into recbind structs, in which defined lex gensyms are substituted with prelex structs.
• let forms into bind structs, in which defined lex gensyms are substituted with prelex structs.
• lambda, case-lambda and annotated-case-lambda forms into clambda structs, in which lex gensyms representing bindings defined by function arguments are substituted with prelex structs.

Undefined bindings and previously defined bindings

In the core language input expression: if a binding is defined by the expression itself, its lex gensym is present as left–hand side in a let, letrec, letrec*, library-letrec* form or as formal argument in a lambda, case-lambda, annotated-case-lambda form. Otherwise the lex gensym represents an undefined binding or a binding defined outside the input form.

Every symbol in the input form not present as left–hand side in a bind, recbind, rec*bind struct or formal argument in a clambda struct is interpreted as a reference to undefined or previously defined binding. This interpretation includes:

• Bindings defined by imported libraries.
• Bindings defined by expressions previously evaluated and added to an interaction environment.
• Bindings that appear for the first time in top level assignment forms:

  (set! ?lhs ?rhs)
  

  like the definitions of syntax transformers in the visit code generated by the expander or standalone expressions evaluated at the REPL; such assignments to undefined bindings cause a new binding to be generated.

for all these cases: the lex gensym acts also as loc gensym.

After the recordisation process:

• All the lex gensyms in reference position representing bindings not defined by the input expression have been converted to:

  (funcall (primref top-level-value) (constant ?loc))
  

  where ?loc is the loc gensym of the binding. If this binding is truly undefined: the call to top-level-value will raise an exception at run–time.

• All the lex gensyms in assignment position representing bindings not defined by the input expression have been converted to:

  (funcall (primref $init-symbol-value!) (constant ?loc) ?rhs)
  

  where ?loc is the loc gensym of the binding; this is the run–time definition of a new binding.

so all the prelex structs introduced by recordize represent bindings defined by the input expression.

NOTE As specified by R6RS, bindings defined by imported libraries must not be mutated; this means core language forms like:

(set! ?loc ?rhs)

in which ?loc is the loc gensym of an imported binding are forbidden. It is the responsibility of the expander not to generate such assignment forms for bindings defined by imported libraries.

NOTE The special handling of core language assignment forms:

(set! ?lex ?rhs)

in which ?lex is both the lex and loc gensym of a binding not defined by the input expression allows:

• The correct handling of visit code generated by the expander.
• The definition of bindings by eval (for example at the REPL) using just set! rather than a proper binding form. This case allows bindings to be added to interaction environment using weird syntaxes; here is an example session at the REPL:

  vicare> (let () (set! a 1) (set! b 2) #f)
  $1 = #f
  
  vicare> a
  $1 = 1
  
  vicare> b
  $1 = 2
  

  It is the responsibility of the expander to forbid such definitions when processing libraries and programs.