Flow-control structures: Difference between revisions
→Elaborations on GO TO: Extend some linkages.
m (→Response) |
(→Elaborations on GO TO: Extend some linkages.) |
||
Line 531:
A compiler may offer the "assigned GO TO" facility, with statements such as <code>ASSIGN 120 TO THENCE</code> scattered about: 120 is a statement label, not an integer, and any statement label may be assigned to variable THENCE (which is an integer variable) as execution proceeds. Then <code>GO TO THENCE</code> will cause a GO TO for the current address held in THENCE... Should you yield to temptations such as <code>THENCE = THENCE - 6</code> (treating it as an ordinary integer), a subsequent <code>GO TO THENCE</code> may end execution with an error message, or something else...
ASSIGN 1101 to WHENCE !Remember my return point.
GO TO 1000 !Dive into a "subroutine"
Line 543:
1000 do something !This has all the context available.
GO TO WHENCE !Return whence I came.</lang>
Since Algol in the 1960s it has been possible to define a routine within a larger routine that has access to all the context of the larger routine and so can be a convenient service routine for it, but Fortran does not allow a subroutine (or function) to be defined within a larger subroutine, except for the arithmetic statement function. One must write separate subroutines and struggle over providing access to context via COMMON and parameters. However, F90 introduced the MODULE arrangement whereby a collection of variables may all be referenced by a group of subroutines in the module without each having COMMON statements in common. Further, it allows a subroutine (or function) to use the CONTAINS feature, after which such a contained routine may be placed. Alas, it may not itself invoke CONTAINS even though Algol allows nesting as desired. And oddly, the contained routine must be at the ''end'' of the containing routine. So much for definition before usage. With such a facility, the possibility arises of perpetrating a GO TO from a contained routine to somewhere in its parent, however the F90 compilers are required to disallow access to outside labels, even those of FORMAT statements - rather a pity for that. Such escapes would have to copy whatever de-allocation steps were needed for a normal exit, which is simple enough on a stack-oriented design such as the B6700. However, its Algol compiler rejected attempts to jump from one routine ''into'' another (!) with the message "Bad GOTO. Too bad." Assembler programmers can do what they want, but for once, Fortran's designers show some restraint.
Once started on this path, many opportunities beckon: perhaps not just action "A" (achieved by "subroutine" 1000) is of use, there may be an action "B", and so on. One can then prepare the equivalent of a "to-do" list via something like<lang Fortran> ASSIGN 2000 TO WHENCE !Deviant "return" from 1000 to invoke 2000.
Line 549:
GO TO 1000
1103 CONTINUE</lang>
So that subroutine" 1000 would be invoked, which then invokes subroutine 2000, which returns via THENCE. Those familiar with LISP or FORTH and similar languages will recognise a struggle to create new "verbs" from existing verbs, and their resulting usage in compound expressions. This is Philip Greenspun's "tenth" rule of programming.
Another such usage would be to implement "coroutines", the classic example being to imagine a system that processes both Fortran statements and Fortran commentary, but each in their own way. After scanning some Fortran source, commentary is found so control flows to resume the commentary processing from where it had left off, then when further Fortran source is found, control flows back whence the Fortran source process left off. This is quite different from having subroutines FCODE and FCOMM which when invoked start at their start each time (as say when a new statement begins) rather than picking up where they left off because the switches occurred in the middle of a statement. Quite aside from questions of mutual recursion.
| |||