Failure to protect read-onlyness of some container types in attributes (regression)

[p6rt]

Migrated from rt.perl.org#127686 (status was 'new')

Searchable as RT127686$

[p6rt]

From @skids

Note that star needs to use EnumMap to test this, but the problem happened well after 2015.09.

(09​:58​:41 AM) skids​: star​: class a { has EnumMap $.f = EnumMap.new() }; my $f = a.new(); $f.f.perl.say; $f.f = :a1, :b2; $f.f.perl.say (09​:58​:41 AM) camelia​: star-m 2015.09​: OUTPUT«EnumMap.new()␤Cannot assign to a readonly variable or a value␤ in block \ at /tmp/lpr91G4eaJ​:1␤␤» (09​:58​:52 AM) skids​: m​: class a { has Map $.f = Map.new() }; my $f = a.new(); $f.f.perl.say; $f.f = :a1, :b2; $f.f.perl.say (09​:58​:53 AM) camelia​: rakudo-moar f93bab​: OUTPUT«Map.new(())␤Map.new((​:a1,​:b2))␤»

psch++ tracked this to the recent attribute inlining optimizations in rakudo fdd37a9 though there were NQP/MoarVM changes involved in that as well.

Note also this, which is not related and also behaves this way on star​:

10​:05​:11 AM) skids​: m​: my %f := Map.new(); %f = (​:a1, :b2); %f.perl.say (10​:05​:11 AM) camelia​: rakudo-moar f93bab​: OUTPUT«Map.new((​:a1,​:b2))␤»

This all fell out from a question asked by JRasPass++ here​:

http://irclog.perlgeek.de/perl6/2016-03-09#i_12160443

[p6rt]

From @zoffixznet

Still present in today's rakudo de5d9e​:

\ m​: my $o = class { has $.a = Map.new​: (​:1a) }.new; say $o.a; $o.a = (​:2b); say $o.a \ rakudo-moar de5d9e​: OUTPUT«Map.new((​:a(1)))␤Map.new((​:b(2)))␤»