bug#16464: + folding differs between compiler and interpreter

[Mark H Weaver]

tags 16464 notabug
thanks

Zefram <address@hidden> writes:

> The + procedure left-folds its arguments in interpreted code and
> right-folds its arguments in compiled code.  This may or may not be a
> bug.

[...]

> R5RS and the Guile documentation are both silent about the order of
> operations in cases like this.  I do not regard either left-folding or
> right-folding per se as a bug.  A portable Scheme program obviously can't
> rely on a particular behaviour.  My concern here is that the compiler
> and interpreter don't match, making program behaviour inconsistent on
> what is notionally a single implementation.  That mismatch may be a bug.

As you suggest, this is not a bug.  The compiler is free to apply a
different ordering to each '+' and '*' within a program, which can be
used to good effect by an optimizer.

> Obviously, with exact numbers the direction of folding makes no
> difference.  But the difference is easily seen with flonums, as flonum
> addition is necessarily non-associative.

Indeed, and if you need to explicitly order the operations (and this
_is_ often needed when working with inexacts to prevent things like
catastrophic annihilation), then you can do so using extra parentheses.

BTW, there are _many_ other cases like this, where certain aspects
behavior are left unspecified, and might not only differ between the
compiler and interpreter, but might even differ between two identical
subexpressions.  Most notably, the order in which the operand and
operands of a procedure call are evaluated is left unspecified, and
again, this can be used to good effect in a compiler for optimization
purposes.  Another is that the initializers in a 'let' or 'letrec' (but
not 'letrec*') can be evaluated in any order.

However, there _are_ some related bugs that you missed.  R5RS specifies
that '-' and '/' are left-associative, which implies that that (- x y z)
must evaluate as (- (- x y) z), and similiarly for '/', but in 2.0.9
(and afaik, in all existing 2.0.x releases), we transform (- x y z) to
(- x (+ y z)), which is incorrect.

I discovered and fixed these bugs in August in the git repository, and
they will be part of Guile 2.0.10.  I also changed the order of
operations of '+' and '*' to be left-associative while I was at it, but
I later realized that this is not required, and we reserve the right to
change the order in the future.

http://git.savannah.gnu.org/gitweb/?p=guile.git;a=commitdiff;h=71673fba930d735c09184d5ca115882239edabb3

I'm closing this bug now.

     Thanks,
       Mark