Re: GOTO considered necessary (reworked)

[bobduff@world.std.com (Robert A Duff)]

In article <dewar.866578231@merv>, Robert Dewar <dewar@merv.cs.nyu.edu> wrote:
>Another form of this I find annoying is "a return is basically a goto,
>so don't use returns" (for example SPARK forbids the use of gotos in
>procedures).

But a return is like a goto in a lot of ways.  E.g. when reading some
code from the top, you don't know whether a given loop might terminate
prematurely due to a return.  And, as I've said in another posting, I've
inserted code at the end of a procedure, thinking it will always be
executed just before the procedure returns, only to find that this is
one of those unusual procedures that contains a "return;" statement.

I'd like to have a language where there was a statement "some returns
coming up soon" at the front, so I can see that easily.

Of course, for functions, Ada doesn't allow you to set the result
without doing a return.  Which I've sometimes found annoying.

>I personally find returns very handy, even though I know perfectly well 
>they are gotos:
>
>  procedure ... is
>  begin
>     if condition then
>        do something simple
>        return;
>
>     else
>       LOTS MORE CODE
>
>it is very handy to read the procedure and immediately know that you
>have the whole story if condition is true, without having to see if
>there is some code following the end if.

I'm curious: Why don't you code this as:

 procedure ... is
 begin
    if condition then
       do something simple
       return;
    end if;

    LOTS MORE CODE

?  I mean, the "else" is redundant nonsense, isn't it?

>I occasionally use gotos in a similar way, almost like an exception (and
>some fanatics would insist on turning them into exceptions) -- you are
>buried deep in complex logic, and you want to say
>
>"get out, abandon all this stuff, this is not what we are looking for, go
>and do something else"
>
>occasionally a goto conveys this intention as clearly as anything :-)

I agree -- if a goto works, then it's usually better than an exception,
in a given case.  Exceptions should be reserved for the case where the
"do something else" needs to be dynamically determined (by the caller,
presumably).

>\x1adp
>
>(a little quiz, why do my messages occasionally end with ^Zdp?
> I know the answer, I am just wondering if it is obvious :-) :-)

I always assumed it's because Robert Dewar uses some weird editor, and
he's sloppy about typing.  No?

- Bob

P.S. "Weird editor" means "editor I'm not used to".  I'm not interesting
in starting a flame war about editors.