From @masak

<tadzik> perl6​: my %a = grammar => 5;
<p6eval> pugs​: ( no output )
<p6eval> ..rakudo 7a5174​: OUTPUT«===SORRY!===â�¤Malformed $*PKGDECL at
line 1, near "=> 5;"�»
<p6eval> ..niecza v9-32-g380d891​: OUTPUT«Potential difficulties​:â�¤ %a
is declared but not used at /tmp/o02WrVecNi line 1​:â�¤------> �[32mmy
�[33m��[31m%a = grammar => 5;�[0m��»
<tadzik> b​: my %a = grammar => 5;
<p6eval> b 1b7dd1​: ( no output )
<tadzik> *sniff sniff* smells like a nom regression
<masak> argh.
* masak submits rakudobug
<jnthn> nom​: my %a = class => 42
<p6eval> nom 7a5174​: OUTPUT«===SORRY!===â�¤Malformed $*PKGDECL at line
1, near "=> 42"�»
<jnthn> nom​: class => 42
<p6eval> nom 7a5174​: OUTPUT«===SORRY!===â�¤Malformed $*PKGDECL at line
1, near "=> 42"�»
<jnthn> hm
<jnthn> No clue what changed.
<jnthn> Oh
<jnthn> Maybe something can't backtrack now that used to be able to.
* tadzik writes a spectest
<jnthn> std​: class => 42
<p6eval> std c22b2ee​: OUTPUT«ok 00​:01 118mâ�¤Â»
<tadzik> yeah, seems so
<jnthn> OK, no idea how STD handles it.
<jnthn> As in STD, the first thing is
<jnthn> { $*SCOPE ||= 'our'; }
<jnthn> Which would terminate LTM before it even started, so it can't be that.
<jnthn> oh, unless it's LTM, but on 'grammar => ' as the declarative prefix.

From @tadzik

Test added in Raku/roast@762eb26315

From @moritz

fixed, test passes.

@moritz - Status changed from 'new' to 'resolved'