wrong output for 1000.fmt("%e") and 1_000_000.fmt("%e")

[p6rt]

Migrated from rt.perl.org#126968 (status was 'open')

Searchable as RT126968$

[p6rt]

From @grondilu

$ perl6 --version
This is Rakudo version 2015.11-686-g09aaf93 built on MoarVM version
2015.11-106-g7545aa9
implementing Perl 6.b.

$ perl6 -e 'printf "%e\n", $_ for 1, 10, 100 ... 1_000_000;'
1.000000e+00
1.000000e+01
1.000000e+02
10.000000e+02
1.000000e+04
1.000000e+05
10.000000e+05

Outputs for 1_000 and 1_000_000 are not what they should.

[p6rt]

From @lizmat

On 19 Dec 2015, at 22​:19, grondilu@​yahoo.fr (via RT) <perl6-bugs-followup@​perl.org> wrote​:

# New Ticket Created by grondilu@​yahoo.fr
# Please include the string​: [perl #​126968]
# in the subject line of all future correspondence about this issue.
# <URL​: https://rt-archive.perl.org/perl6/Ticket/Display.html?id=126968 >

$ perl6 --version
This is Rakudo version 2015.11-686-g09aaf93 built on MoarVM version
2015.11-106-g7545aa9
implementing Perl 6.b.

$ perl6 -e 'printf "%e\n", $_ for 1, 10, 100 ... 1_000_000;'
1.000000e+00
1.000000e+01
1.000000e+02
10.000000e+02
1.000000e+04
1.000000e+05
10.000000e+05

Outputs for 1_000 and 1_000_000 are not what they should.

I assume this is some kind of roundoff error when using floating point calculations to determine to go to the next power​:

$ 6 'printf "%e\n",$_ for 11, 110, 1100 ... 1_100_000'
1.100000e+01
1.100000e+02
1.100000e+03
1.100000e+04
1.100000e+05
1.100000e+06

seems pretty ok, but then the value is not near the boundary.

This seems to be happening inside nqp​::sprintf , and thus is really an NQP bug, or an underlying math lib bug.

This reminds me of a bug my navigator has​: when the distance gets below 10 km, it will show one digit of accuracy (like in 9.9). However, the cutoff point is wrong, because if the value is still 9.95 km or above, it will round to 10, and thus show 10.0 for a little while. Not sure whether this anything to do with it.

Liz

[p6rt]

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