• Ben Hinkle (15/15) Aug 10 2004 All the recent talk of removing opEquals and opCmp and possibly messing ...
• Matthew (16/31) Aug 13 2004 If you're having another go, then I will. :)

Ben Hinkle <bhinkle4 juno.com> writes:

All the recent talk of removing opEquals and opCmp and possibly messing with
TypeInfos got me thinking again about the fact that running
 Object x;
 x == null;
will seg-v. Writing that generically the code
 T x;
 x == T.init
will work fine for T=int, T=double, T=char[] but for T=Object it will seg-v.
That seems evil. Is there any hope of changing that behavior? It would also
make porting Java (and the equivalent C/C++) simpler since code fragments
like
 if (x != null) { ... }
currently will seg-v if x actually turns out to be null. It's pretty
annoying to uncover those typos one by one during run-time testing.

-Ben

"Matthew" <admin stlsoft.dot.dot.dot.dot.org> writes:

If you're having another go, then I will. :)

    x == y

should be automatically rewritten, for object types, as

    (null === x) ? (null === y) ? true : false : (x == y)

except where x and/or y can be deduced (by the optimiser, not by the
programmer) to be non-null, in which case the
rewritten form may be one of

    (null === x) ? false : (x == y)

    (null === y) ? false : (x == y)

    (x == y)

I continue to reject all the counter arguments that were espoused at the time
this was raised - about 15 months' ago, if
memory serves - and continue to maintain that this will dramatically increase
robustness, reduce surprise, and, given
Ben's arguments, also help enormously with generics. Furthermore, I state that
this will have NO discernible performance
cost, since most times that the automatic rewriting results in a test against
null, that test would have been performed
somewhere manually anyway.


But do I expect this to fly? No.


"Ben Hinkle" <bhinkle4 juno.com> wrote in message
news:cfatup$31ip$1 digitaldaemon.com...
 All the recent talk of removing opEquals and opCmp and possibly messing with
 TypeInfos got me thinking again about the fact that running
  Object x;
  x == null;
 will seg-v. Writing that generically the code
  T x;
  x == T.init
 will work fine for T=int, T=double, T=char[] but for T=Object it will seg-v.
 That seems evil. Is there any hope of changing that behavior? It would also
 make porting Java (and the equivalent C/C++) simpler since code fragments
 like
  if (x != null) { ... }
 currently will seg-v if x actually turns out to be null. It's pretty
 annoying to uncover those typos one by one during run-time testing.

 -Ben