It is also possible to emit bulk code into either the generated header file or generated body file.

Header and body insertions are said to be floating because they're not emitted where they're written, instead they float up to a fixed place at the top of the generated file.

Floating insertions are only emitted if they're required.

Floating insertions can be named:

  body mystuff = '''
    struct mytype { 
      int data; 
      mytype(int x) : data(x) {} 
    };
  ''';
  type mytype = "mytype" requires mystuff;
  ctor mytype : int = "mytype($1)";
  ctor int : mytype = "$1.data";
  println$ int (mytype 42);

42

If a floating insertion is given a tag name like this, it is not emitted unless it is required. The println command here must be executed, and that requires the constructor for mytype which in turn requires the type mytype which in turn requires the header mystuff due to the requires clause. Therefore the body of the generated Felix program will contain {#include "mystuff.h"} which presumably defines the functions make and toint.

Requires clauses can be added to C bindings (including header and body statements).

Felix will only ever add a (monomorphic) floating insertion once. The string value of the insertion is checked to ensure this. For this reason cyclic requirements are allowed, and will not lead to infinite insertions. Instead, the transitive closure of requirements is subject to insertion. Note that the closure may be empty and this is not an error: requiring a tag name which is not defined it permitted and follows from the definition, although the behaviour may seem non-intuitive.