tcltk-perl

RE: [tcltk-perl] return of 1=>1.0 transformation error

[Konovalov, V. <vko...@sp...>]

> There are no integer literals in Perl.  All numbers and arithmetics
> are/are done with doubles.  If you want to force integers you need to
> declare that intention with the 'use integer' pragma.  perl might try
> to use integers when it can get away with it and newer version are
> more aggressively doing so, but there are no guarantees.

'use integer;' will probably bring problems with having this pragma turned
on/off when user do not expect it to be changed somehow.
Much better to early stringification early.

Vadim.

RE: [tcltk-perl] return of 1=>1.0 transformation error

[Konovalov, V. <vko...@sp...>]

> 	if (SvIOK(sv) && ((double)(ival = SvIV(sv)) == dval)) {
> 
> returns different results for 0 as 1.0, but a quick C check
> with this:
> 
>     int ival = 1;
>     double dval = 1.0;
>     printf("%d\n", (((double)(ival = (int)dval)) == dval));
> 
> Says it should convert back and forth OK (that shows '1').  I
> know you will have issues with incorrect double equality on
> some corner cases.
> 
> > You can investigate this using Devel::Peek.  It is not part 
> 
> I only have 5.8 installed, so it always works for me.  Perhaps
> Vadim would like to hunt this down?

I will hunt this down with a great pleasure, but will not have possibility
to do this until nearest Saturday.
For sure will do, but probably someone will do this faster.

Vadim.

Re: [tcltk-perl] return of 1=>1.0 transformation error

[Gisle A. <gi...@Ac...>]

"Konovalov, Vadim" <vko...@sp...> writes:

> > There are no integer literals in Perl.  All numbers and arithmetics
> > are/are done with doubles.  If you want to force integers you need to
> > declare that intention with the 'use integer' pragma.  perl might try
> > to use integers when it can get away with it and newer version are
> > more aggressively doing so, but there are no guarantees.
> 
> 'use integer;' will probably bring problems with having this pragma turned
> on/off when user do not expect it to be changed somehow.

The user code would have to use this pragma.  It is lexical and it is
the user's numbers we want to be ints.  Because of this it is not a
solution to this problem.  I just mentioned it to describe how perl works.

> Much better to early stringification early.

Or just add more checks to the code so that it creates int objects
whenever that would be an exact translation of the SV.  A few more
conditionals and it should be all ok again.

--Gisle

RE: [tcltk-perl] return of 1=>1.0 transformation error

[Jeff H. <je...@ac...>]

> > Much better to early stringification early.
> 
> Or just add more checks to the code so that it creates int 
> objects whenever that would be an exact translation of the 
> SV.  A few more conditionals and it should be all ok again.

Vadim - please verify that just removing the "SvIOK(sv) &&"
condition is all that is necessary.  IOW:

    else if (SvNOK(sv)) {
	double dval = SvNV(sv);
	int ival;
	/*
	 * Perl does math with doubles by default, so 0 + 1 == 1.0.
	 * Check for int-equiv doubles and make those ints.
	 */
	if ((double)(ival = SvIV(sv)) == dval) {
	    objPtr = Tcl_NewIntObj(ival);
	} else {
	    objPtr = Tcl_NewDoubleObj(dval);
	}
    }

we could possibly save a cycle or two with:

	if ((double)(ival = (int)dval) == dval) {

but I'm not sure avoiding the SvIV is a good idea.  ???

Jeff

Re: [tcltk-perl] return of 1=>1.0 transformation error

[Vadim K. <va...@ar...>]

> > Or just add more checks to the code so that it creates int 
> > objects whenever that would be an exact translation of the 
> > SV.  A few more conditionals and it should be all ok again.
> 
> Vadim - please verify that just removing the "SvIOK(sv) &&"
> condition is all that is necessary.  IOW:

This cures the problem, no more fails in perl-5.6.1.

Thank you for providing a fix!


> 
>     else if (SvNOK(sv)) {
> double dval = SvNV(sv);
> int ival;
> /*
> * Perl does math with doubles by default, so 0 + 1 == 1.0.
> * Check for int-equiv doubles and make those ints.
> */
> if ((double)(ival = SvIV(sv)) == dval) {
>     objPtr = Tcl_NewIntObj(ival);
> } else {
>     objPtr = Tcl_NewDoubleObj(dval);
> }
>     }
> 
> we could possibly save a cycle or two with:
> 
> if ((double)(ival = (int)dval) == dval) {
> 
> but I'm not sure avoiding the SvIV is a good idea.  ???

I may be wrong, but probably your fix could be #ifdef-ed for perl-5.6.1? 
But in this case there will be unnecessary complications, IMHO

Vadim.