Re: some proposed changes

[Owen Taylor <otaylor redhat com>]

Huw Rogers <count0 building2 co jp> writes:

> > 
> > Huw Rogers <count0 building2 co jp> writes:
> > > * if a sequence is undef, it is marshalled as a zero length sequence.
> > > Right now if it is not a true reference to an array it dies.
> > 
> > I don't agree with this at all.
> > 
> >  $a = []; print scalar (@$a), "\n";
> >  $a = undef; print scalar (@$a), "\n";
> > 
> > Are very different; the second one is a warning. 
> > 
> > Allowing this would be similar to allowing "undef" to be marshalled as
> > an empty string without any warning, which is similar to thinking
> > in C that "" is the same as NULL. <shudder>
> 
> The sequence marshalling of undef is the most dubious of my proposals,
> and I guess I should retract it.
> 
> However I note that attempting to marshal undef as a string currently
> results in a core dump. 

:-(. well, that's not right, of course. 

> In perl, undef is a scalar, interchangeable as
> such with a string, and in CORBA "" is a null string (actual NULL is
> invalid).

The equivalence is, as I as I see it is:

 C      NULL   ""        
 Perl   undef  ""
 CORBA  ----   ""

CORBA doesn't have such a thing as an NULL string, though it does have
 the nil object for object references. Now, Perl has a long tradition
 of coercing undefined variables to the the empty string in string
 contexts:

 "" . undef 

gives, but it's basically there just for compatibiltiy
with Perl 4 and gives a warning when -w is used. (And
everybody uses -w when developing, I hope...)

 So although I'd retract my sequence marshalling proposal, I
> would say that undef should indeed be marshalled as a null string ("") 
> (which would certainly be better than the current SEGV behaviour).

Well, my opinion is it should be marshalled as a *empty* string,
but with a warning when -w is used. I believe "" and a warning is
generated if a C Perl extension calls call SvPV() on an undefined Sv.

Regards,
                                        Owen