[MLton-user] Potential bug in MLton.eq?

[Matthew Fluet]

No bug; you are seeing the expected behavior.

MLton.eq tries very hard to not inhibit other optimizations.  So,
MLton.eq does not imply a use of the components of a data object.
Various other optimizations will eliminate unused components of data
objects.  In your program, the string component of the A variant is
never used, so it can be eliminated.  (Furthermore, the B variant is
never used at all, so can be eliminated from the datatype
declaration.)   What remains is a program equivalent to:

  datatype d = A
  val v1 = A
  val v2 = A

At this point, you can see that common subexpression elimination would
allow us to share the same object for both v1 and v2, and
MLton.eq(v1,v2) would return true.  Actually, MLton will further
optimize the above, accomplish the constant folding of MLton.eq at
compile-time, and your entire program is simply equivalent to:

  val _ = print ("v1 <eq> v2 : " ^ (Bool.toString true) ^ "\n")

You've probably noticed that if you uncomment the line that includes a
call to MLton.equal, then MLton.eq returns false.  This again is the
expected behavior, because MLton.equal is a deep structural equality
test, and it does imply a use of all of the components of a data
object.

On Wed, Apr 28, 2010 at 12:32 PM, Nicholas Kidd <nakidd at gmail.com> wrote:
> Hi,
>
> For the attached program ab.sml the following occurs:
>
> ----------------------------
> $ cat ab.sml
> datatype d = A of string | B of string
>
> val v1 = A("y")
> val v2 = A("x")
>
> val _ = print ("v1 <eq> v2 : " ^ (Bool.toString (MLton.eq(v1,v2))) ^ "\n")
> (*val _ = print ("v1 <equal> v2 : " ^ (Bool.toString
> (MLton.equal(v1,v2))) ^ "\n")*)
> $ mlton
> MLton 20070826 (built Mon Sep 03 18:05:04 2007 on eponym)
> $ mlton ab.sml
> $ ./ab
> v1 <eq> v2 : true
> $ uname -a
> Linux jack 2.6.26-2-686 #1 SMP Wed Feb 10 08:59:21 UTC 2010 i686 GNU/Linux
>
> ----------------------------
>
> I believe the last line should be "v1 <eq> v2 : false" as v1 and v2
> should not hold the same
> address. The output was produced on Debian lenny, and has been
> reproduced on OS 10.{5,6} and with MLton v7387
> and v7452 from svn. Finally, uncommenting the last line outputs:
>
> v1 <eq> v2 : false
> v1 <equal> v2 : false
>
> Any idea what could be going on?
>
> Thanks,
> -Nick
>
> _______________________________________________
> MLton-user mailing list
> MLton-user at mlton.org
> http://mlton.org/mailman/listinfo/mlton-user
>