From @masak

<sorear> rakudo​: my $r1 = role { method foo() { 5 } }; my $r2 = role {
method foo() { 7 } }; my %hash{Any}; %hash{"quux" but $r1} = 9;
%hash{"quux" but $r2} = 11; say %hash.keys>>.foo
<p6eval> rakudo d4dc7d​: OUTPUT«5␤»
<moritz> ouch
<masak> huh.
< moritz> rakudo​: my $r1 = role { method foo() { 5 } }; my $r2 = role
{ method foo() { 7 } }; say ("quux" but $r1).WHICH; say ("quux" but
$r2).WHICH
<p6eval> rakudo d4dc7d​: OUTPUT«Str+{<anon>}|quux␤Str+{<anon>}|quux␤»
<moritz> seems we are insufficiently paranoid in ClassHOW.gist, or we
should be using something else in .WHICH
<masak> those look more like .WHAT than .WHICH to me.
<moritz> masak​: feel free to rakudobug
* masak submits rakudobug
<jnthn> Er, we shoudln't be using the typename. :)
<jnthn> Just use nqp​::where($the_type.HOW) to get a unique identifier
for the type.
<moritz> ok

From @usev6

Current behaviour​:

$ perl6-m -e 'my $r1 = role { method foo() { 5 } }; my $r2 = role { method foo() { 7 } }; my %hash{Any}; %hash{"quux" but $r1} = 9; %hash{"quux" but $r2} = 11; say %hash.keys>>.foo'
(5 7)

Is that the expected answer?

@usev6 - Status changed from 'new' to 'open'

@diakopter - Status changed from 'open' to 'resolved'

From @usev6

I added a test to S09-hashes/objecthash.t with commit Raku/roast@92543e196c