Control-flow stack for i, j, leave, unloop

[ruv]

When the compilation semantics for the words i, j, leave, unloop are performed, at least one do-sys must be present on the control-flow stack (and two do-sys in the case of j).

But currently a system cannot check this fact and cannot throw an exception if these words are used not within do...loop, because the control-flow stack diagram is not specified for compilation semantics. A system may only check in run-time if loop-sys is present on the top of the return stack and throw an exception if not — since loop-sys is specified in the stack diagrams.

We should specify compilation semantics for these words, with control-flow stack diagrams like: ( C: do-sys i*x -- do-sys i*x ), or ( C: do-sys2 i*x do-sys1 j*x -- do-sys2 i*x do-sys1 j*x ) for the word j.

It will allow a system to formally check do-sys at compile-time and throw an exception if these words are used incorrectly in compile-time (according to 3.1, which says: "An ambiguous condition exists if an incorrectly typed data object is encountered").

See also

• Discussion: Stack effect of LEAVE during compilation
• Implementation of DO ... LOOP in the high level — via BEGIN ... UNTIL

[ruv]

A similar issue exists for the of word.
Currently the spec says:

Compilation: ( C: -- of-sys )

But it should say:

Compilation: ( C: case-sys -- case-sys of-sys )

[ruv]

An argument against ( C: do-sys i*x -- do-sys i*x ) for the compilation semantics of i is that the execution semantics of the following definition will be noncompliant without any ground:

: build-foo
  :noname
  [: postpone do ;] execute-balance n>r
  \ ...
  postpone i
  \ ...
  nr> drop postpone loop
  postpone ;
;

In the same time, it is desirable to make the following definition grammatically noncompliant:

: bar
  \ ...
  i
  \ ...
;