letrec* not evaluating in defined order

[leppie]

Here's a few:

Ikarus Scheme version 0.0.3+ (revision 1447, build 2008-04-13)
Copyright (c) 2006-2008 Abdulaziz Ghuloum

> (letrec* ([b c] [c 1]) b)
1
> (letrec* ([b c] [c b]) c)
> (letrec* ([b c] [c 1]) c)
1

The last 2 should also be signaling an error. Related to similar behavior in letrec?

[Abdulaziz Ghuloum (aghuloum)]

On Apr 13, 2008, at 3:02 AM, leppie wrote:
> The last 2 should also be signaling an error. Related to similar
> behavior in letrec?

Yes they should, and I'm aware that I'm not checking for these
kinds of errors, neither for letrec nor for letrec*. Will fix
at some point but not today. :-)

[leppie (leppie)]

What kind of error should (letrec ((a a)) a) signal?

Petite Chez says:
Error: attempt to reference undefined variable a.

But that is not really correct.

From the R6RS spec it appears it requires it to be of type &assertion.

My feeling is this is a syntax error as it's detectable at compile time.

[Abdulaziz Ghuloum (aghuloum)]

On Apr 13, 2008, at 6:11 AM, leppie wrote:
> What kind of error should (letrec ((a a)) a) signal?
>
> Petite Chez says:
> Error: attempt to reference undefined variable a.
>
> But that is not really correct.

The value of a letrec-bound variable is undefined until after the
body of the letrec is entered. The value of a letrec*-bound variable
is undefined until after the right-hand-side expression is
evaluated. I don't see how Chez's behavior is "not really correct".
Can you elaborate?

>> From the R6RS spec it appears it requires it to be of type
>> &assertion.
>
> My feeling is this is a syntax error as it's detectable at compile
> time.

In general, such violations are not detectable at compile time, but
they can always be detected at run time.

For example, (letrec ((a (begin (foo) a))) a) evaluates correctly for
some values of foo but not for others, so, it cannot be a syntax error.

Aziz,,,

[Abdulaziz Ghuloum (aghuloum)]

BTW, do you have an example of a correct letrec* program that Ikarus evaluates in an incorrect order? The programs on top are all incorrect, so, no implementation is required to evaluate them correctly. :-)

[leppie (leppie)]

Unfortunately not :) But as a test, this confirms correct order.

(letrec* ((a (begin (display 'a) b))(b (begin (display 'b) 1))) a)

> The value of a letrec-bound variable is undefined until after the
> body of the letrec is entered.

Yes the value is undefined, but variable is itself is defined. I guess you could interpret that either way.

> In general, such violations are not detectable at compile time, but
> they can always be detected at run time.

I guess I am lucky I can detect these (in IronScheme) at compile time then :)

> For example, (letrec ((a (begin (foo) a))) a) evaluates correctly for
> some values of foo but not for others, so, it cannot be a syntax error.

Would that be for example an escape procedure? If so, then I guess, back to the drawing board for me!

Cheers

leppie

[Abdulaziz Ghuloum (aghuloum)]

On Apr 13, 2008, at 7:46 AM, leppie wrote:
>
>> For example, (letrec ((a (begin (foo) a))) a) evaluates correctly for
>> some values of foo but not for others, so, it cannot be a syntax
>> error.
>
> Would that be for example an escape procedure?

It could be anything that does not return including calling escape
procedures, raising an exception, calling exit, or going in an
infinite loop.

Aziz,,,

[leppie (leppie)]

That complicates things a bit indeed :)

[Abdulaziz Ghuloum (aghuloum)]

I've marked this as a duplicate of a bug that I just opened since solving that problem solves this automatically.