Strange behavior of »--

[frithnanth]

I was toying with the code shown in this blog post Assuming optionality and I realized that running the following one-liner repeatedly:

$ raku -e'my BagHash $a=‘Perl Weekly Challenge’.comb.BagHash»--;$a.keys.sort.say'

got me different results, such as:

(  e l)
(  W e l)
(  C e l)

OTOH, the following:

$ raku -e'my BagHash $a=‘Perl Weekly Challenge’.comb.BagHash;$a.keys.sort.say'

always returns the same result, so I guess that the problem lays with the '»--'.

I'm running Rakudo v2022.07 on Debian sid.

[lizmat]

Good catch! This is indeed weird.

But the >>-- appears to only be an issue when used when creating the BagHash. This appears to always produce the same results (at least for me):

$ raku -e 'my $a="Perl Weekly Challenge".comb.BagHash; say $a; $a>>--; say $a'
BagHash( (2) C P W a e(5) g h k l(4) n r y)
BagHash(  e(4) l(3))

[frithnanth]

Yes, your version works fine here too!

[lizmat]

$ raku -e '(my $a=‘Perl Weekly Challenge’.comb.BagHash)»--; say $a'
BagHash(  e(4) l(3))

also does the right thing.

[lizmat]

Some progress: It turns out that in the case of the error, it is always the first key that is being decremented, that loses its decrement. If that is one of the higher valued keys, then you won't notice it unless you actually look at the value. If it is one the keys that has a value of 1, it won't get removed.

[lizmat]

FWIW, this is the golf I'm working on ATM:

$ raku -e 'my $a:=‘Perl Weekly Challenge’.comb.BagHash.deepmap(&postfix:<-->); say $a'

as the >>-- effectively codegens to a call to deepmap.

[lizmat]

Another datapoint: .MixHash has the same issue.

[lizmat]

$ raku -e 'my $a:=‘Perl Weekly Challenge’.comb.MixHash.deepmap(&postfix:<-->); say $a'

[lizmat]

Another Datapoint:

$ raku -e 'my $a:=‘Perl Weekly Challenge’.comb.BagHash.deepmap(&postfix:<++>); say $a'
BagHash( (2) C P W a e(5) g h k l(4) n r y)

Note that the increments somehow got lost :-(

[lizmat]

Ok, finally got to the bottom of this: 63d03eb899

[lizmat]

d0ec99a861 changed the example:

$ raku -e 'my $a=‘Perl Weekly Challenge’.comb.BagHash»--;$a.keys.sort.say'
(  C P W a e g h k l n r y)

and this is correct, as the increments happen on the original ‘Perl Weekly Challenge’.comb.BagHash and the result of >>-- on that is stored in $a. Since this has been specced as an object of the same type, but built on the return value of the operator, you effectively get the original BagHash and the decrements were done on the original BagHash object, which is not saved anywhere and therefore lost.

Adding parentheses to make sure that $a is bound before running >>--, makes the example run reliably now with the expected result:

$ raku -e '(my $a=‘Perl Weekly Challenge’.comb.BagHash)»--;$a.keys.sort.say'
(  e l)

So I'm now closing this issue, now that we hopefully have full clarity on the issue at hand, and its resolution.