search

Raku / old-issue-tracker

Tickets from RT
https://github.com/Raku/old-issue-tracker/issues
2 stars 1 forks source link

1**NaN is 1, but any numeric operation with NaN should be NaN (say 1**NaN) #5752

Closed p6rt closed 8 years ago

p6rt commented 8 years ago

Migrated from rt.perl.org#129894 (status was 'rejected')

Searchable as RT129894$

p6rt commented 8 years ago

From @AlexDaniel

*Code​:* say 0**NaN; say 1**NaN; say 2**NaN;

*Output​:* NaN 1 NaN

See also IRC log​: https://irclog.perlgeek.de/perl6-dev/2016-10-16#i_13412758

p6rt commented 8 years ago

From @zoffixznet

More, at nqp level​:

\ m​: use nqp; dd nqp​::pow_n(NaN, 0e0) \ rakudo-moar 2b5226​: OUTPUT«1e0␤» \ m​: use nqp; dd nqp​::pow_n(1e0, NaN) \ rakudo-moar 2b5226​: OUTPUT«1e0␤»

p6rt commented 8 years ago

The RT System itself - Status changed from 'new' to 'open'

p6rt commented 8 years ago

From @zoffixznet

This was uncovered by coverage work and the num case is fudged in https://github.com/perl6/roast/commit/47aba2d2ea

Once fixed, more tests will be needed for this ticket, but please unfudge that test as well.

p6rt commented 8 years ago

From @AlexDaniel

Just for the record, another case​:

Code​: say NaN**0

Output​: 1

p6rt commented 8 years ago

From @usev6

On Sun Oct 16 11​:07​:16 2016, alex.jakimenko@​gmail.com wrote​:

Just for the record, another case​:

Code​: say NaN**0

Output​: 1

There was an earlier RT for this one (coming to the conclusion that 1 is a reasonable answer here)​: https://rt-archive.perl.org/perl6/Ticket/Display.html?id=124450. It has a link to a discussion from stackoverflow -- to me the following comment was interesting​: http://stackoverflow.com/a/17865226

p6rt commented 8 years ago

From @zoffixznet

So that leaves nqp​::pow_n(1e0, NaN) == 1e0. Is that also part of IEEE?

I see that Perl 5 gives a NaN for NaN**0, but 1 for 1**NaN​:

zoffix@​VirtualBox​:\~$ perl -wlE 'say -sin(9**9**9)**0'
NaN zoffix@​VirtualBox​:\~$ perl -wlE 'say 1**-sin(9**9**9)'
1

On Mon Oct 17 00​:25​:15 2016, bartolin@​gmx.de wrote​:

On Sun Oct 16 11​:07​:16 2016, alex.jakimenko@​gmail.com wrote​:

Just for the record, another case​:

Code​: say NaN**0

Output​: 1

There was an earlier RT for this one (coming to the conclusion that 1 is a reasonable answer here)​: https://rt-archive.perl.org/perl6/Ticket/Display.html?id=124450. It has a link to a discussion from stackoverflow -- to me the following comment was interesting​: http://stackoverflow.com/a/17865226

p6rt commented 8 years ago

From @zoffixznet

I took a gander at 2008 IEEE 754 standard and section "9.2.1 Special values" has this to say​:

  pow (x, ±0) is 1 for any x (even a zero, quiet NaN, or infinity)   pow (+1, y) is 1 for any y (even a quiet NaN)

So our behaviour is correct. Rejecting.

On Mon Oct 17 03​:05​:33 2016, cpan@​zoffix.com wrote​:

So that leaves nqp​::pow_n(1e0, NaN) == 1e0. Is that also part of IEEE?

I see that Perl 5 gives a NaN for NaN**0, but 1 for 1**NaN​:

zoffix@​VirtualBox​:\~$ perl -wlE 'say -sin(9**9**9)**0' NaN zoffix@​VirtualBox​:\~$ perl -wlE 'say 1**-sin(9**9**9)' 1

On Mon Oct 17 00​:25​:15 2016, bartolin@​gmx.de wrote​:

On Sun Oct 16 11​:07​:16 2016, alex.jakimenko@​gmail.com wrote​:

Just for the record, another case​:

Code​: say NaN**0

Output​: 1

There was an earlier RT for this one (coming to the conclusion that 1 is a reasonable answer here)​: https://rt-archive.perl.org/perl6/Ticket/Display.html?id=124450. It has a link to a discussion from stackoverflow -- to me the following comment was interesting​: http://stackoverflow.com/a/17865226

p6rt commented 8 years ago

@zoffixznet - Status changed from 'open' to 'rejected'