ALTER, ASSIGN and labels-as-values (was: Unicode in strings)

"Stephen Fuld" <SFuld@alumni.cmu.edu.invalid> writes:

That feature (Alter GOTO) was also in Fortran, as the, long since
deprecated, assigned GOTO statement.  I believe they were there to
support some older computers that didn't have indexed jump/branch
instructions, so achieved the effect by modifying the branch
destination in the instruction itself.

Fortran's assigned goto directly corresponds to indirect branches on
modern machines, so in 1954 when Fortran was designed, the target
machines for Fortran had such instructions.

The ALTER statement looks somewhat stranger, judging from the example
given at
<https://www.microfocus.com/documentation/cobol-it/4-11/user-guide/compiler-suite/procedure-division-statements-alter.html>:

           ...
              ALTER ALTER-1-EXIT TO PROCEED TO ALTER-2-EXIT
           END-IF.

           DISPLAY "GO TO ALTER-1-EXIT...." LINE 4 COL 10.
           GOTO ALTER-1-EXIT.

       ALTER-1-EXIT.
           GOTO LABEL-1.

       LABEL-1.
           DISPLAY "ALTER-1 EXIT!" LINE 10 COL 10.
           ACCEPT DUMMY LINE 10 COL 30.
           STOP RUN.

       ALTER-2-EXIT.
           DISPLAY "ALTER-2 EXIT!" LINE 10 COL 10.
           ACCEPT DUMMY LINE 10 COL 30.
           STOP RUN.

What confuses me is that there is a GOTO right in the ALTER-1-EXIT
section (or is it a procedure?).  So does the ALTER change the "GOTO
ALTER-1-EXIT" and every other occurence of "GOTO ALTER-1-EXIT" that
occurs in the program (as I would expect from the general rule 1. on
the web page linked above; if so, why have the "GOTO LABEL-1" which is
followed right by the label "LABEL-1."?  Of is it required that the
given section or procedure starts with a GOTO, and the ALTER statement
changes that GOTO?

Let's compare this to FORTRAN's assigned goto.
<https://docs.oracle.com/cd/E19957-01/805-4939/6j4m0vn9j/index.html>
gives the following example:

       ASSIGN 10 TO N
       ...
       GO TO N ( 10, 20, 30, 40 )
       ...
10     CONTINUE
       ...
40     STOP

Here N could be a register, the ASSIGN statment loads a constant (the
code address of the label 10) into that register, and the GOTO N
performs a branch to that code address.  The labels given in
parentheses limits the possible targets, allowing the compiler to
perform more accurate data-flow analysis and especially limit the
variables which are considered to be alive at the GOTO (I wonder when
the parenthesis part was added).

Why the strange syntax.  Why not just write

N = 10

Because the ASSIGN syntax tells the compiler that it is dealing with a
label and not with an integer here, and it can perform the conversion
from label number to code address at compile time instead of having to
look the code address up in a sparse table at run-time.  Another
option would have been to introduce a new type for labels, and after
declaring N to be a label the assignment above could easily have
produced code for loading the code address.  But I guess the ASSIGN
feature was introduced before programming language designers tried to
attack all kinds of problems with types.

People love to point out that the assigned GOTO has been removed from
Fortran, as a damnation of the concept.  But it just tells us that
Fortran programmers don't program the kind of code where this feature
is useful.  Every assembly language still has this feature, and GNU C
has added it to C in the labels-as-values extension.  And from there
it has spread to other C compilers: Intel's icc, Bellard's tcc, and
clang are among them.  In GNU C the Fortran example looks as follows:

void *n;
n = &&l10;
...
goto *n;
...
l10:  ...
...
l40:
...

Again, an existing type is used, and syntax rather than types are used
to tell the compiler what is going on.  There is no way to specify the
possible targets.

I use this feature for implementing threaded-code interpreters, but
also for implementing code-copying systems (in-between things that
share some characteristics with interpreters and some with compilers).
I find it highly useful.

- anton
--
'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'
  Mitch Alsup, <c17fcd89-f024-40e7-a594-88a85ac10d20o@googlegroups.com>