Re^4: Problem with a sort result

[rdfield]

Looks to me like it's an integer/bigint problem: 9 digits:

perl -E'
$h={foo=>{bar=>10000000000},baz=>{bar=>70000000},qux=>{bar=>2000000000
+0}};
say "$_: $h->{$_}->{bar}" for sort { eval { say $h->{$b}->{bar}. " - "
+.$h->{$a}->{bar}." = ".($h->{$b}->{bar} - $h->{$a}->{bar}); $h->{$b}-
+>{bar} - $h->{$a}->{bar} } } keys %$h;
'
10000000000 - 70000000 = 9930000000
20000000000 - 10000000000 = 10000000000
qux: 20000000000
foo: 10000000000
baz: 70000000

the expected output is correct. 12 digits:

perl -E'
$h={foo=>{bar=>10000000000000},baz=>{bar=>70000000},qux=>{bar=>2000000
+0000000}};
say "$_: $h->{$_}->{bar}" for sort { eval { say $h->{$b}->{bar}. " - "
+.$h->{$a}->{bar}." = ".($h->{$b}->{bar} - $h->{$a}->{bar}); $h->{$b}-
+>{bar} - $h->{$a}->{bar} } } keys %$h;
'
20000000000000 - 70000000 = 19999930000000
10000000000000 - 70000000 = 9999930000000
foo: 10000000000000
baz: 70000000
qux: 20000000000000

incorrect ordering.

[Discipulus]

Infact it is something strange: while in motorbike this morning i thought i was wrong (in my previous post): the results of subtraction must be valid too when sorting: big minus less is greater than zero; equal minus equal is 0 and little minus big is less than zero:


perl -e "print join qq(\n),sort {$a - $b} @ARGV" 33 1 2 99
1
2
33
99
# same result using <=> or -

using big nums give me incorrect ordering using subtraction when one number is ten digits or more:

perl -e "print join qq(\n),sort {$a - $b} @ARGV" 10000000000 3 2
10000000000
2
3



L*

[kzwix]

Discipulus, rdfield, you are both correct: This is a problem with long integers.

What I don't understand is: Why are the integers correctly remembered, correctly displayed, but incorrectly compared to 0 ? Because, let's face it, it's the core of the problem:

When it's a "small" number, it is correctly evaluated as positive, negative, or null.

When it is a "big" number, then there may be problems in evaluating the sign. Much like if some bit from the big number was used as a sign bit, or whatever, where the real good practice would merely be to check the sign at the start of the number string...

I mean, if it starts with a "+" or a digit, then it's positive. If it starts with a "-", it's negative. We keep that "sign", unless all digits are 0 -> in which case it is merely null.

Did I miss something, there ?

[choroba]

The problem is that the code that implements the behaviour is in C. There's no number string, there are different integer types with different byte sizes.