LTA error: assignment to $_ alised to Array says value is immutable instead of suggesting list assignment

[zoffixznet]

Feels like a case that should Do The Right Thing.

18:34 | Zoffix | m: my @a = <a b c d f b g e>; $_ = 42 with @a;
-- | -- | --
18:34 | camelia | rakudo-moar b22edce40: OUTPUT: «Cannot assign to an immutable value␤ in block <unit> at <tmp> line 1␤␤»
18:34 | Zoffix | Was kinda expecting it to DWIM and .STORE into that array

This works tho:

18:36 | Zoffix | m: my @a = <a b c d f b g e>; .[] = 42 with @a; dd @a
18:36 | camelia | rakudo-moar b22edce40: OUTPUT: «Array @a = [42]␤»

If fixing this, check how much performance is affected for assigning stuff to $_ in such cases (regardless whether it contains an Array or not). If there's a noticeable impact, we should defer the change until a later time IMO, as assigning to aliased $_ is a very common case, while doing the same when it's an array ain't.

[zoffixznet]

http://colabti.org/irclogger/irclogger_log/perl6-dev?date=2018-09-22#l134

nm on assignment not working. It's fine and the same is the case in other instances. However, the error is LTA as the user should instead be told to use list assignment:

18:41 | TimToady | in any case, it's not exactly an immutable value, so the error message is LTA
-- | -- | --
18:42 | TimToady | m: my @a = <a b c d f b g e>; @$_ = 42 with @a;
18:42 | camelia | rakudo-moar b22edce40: ( no output )
18:42 | TimToady | in the design, the scalar vs list assignment is determined statically, so maybe the error should just point them at @$_
18:43 | Zoffix | m: use nqp; my @a = <a b c d f b g e>; nqp::p6store($_,42) with @a; dd @a
18:43 | camelia | rakudo-moar b22edce40: OUTPUT: «Array @a = [42]␤»
18:43 | TimToady | we could make it failover, but too many failovers is a design smell
18:43 | Zoffix | It could use that op, no? Tho I feer it'll impact perf
18:44 | TimToady | well, probably only if it prevents inlining
18:44 | Zoffix | m: my $a := [<a b c d f b g e>]; $a = 42; dd $a
18:44 | camelia | rakudo-moar b22edce40: OUTPUT: «Cannot assign to an immutable value␤ in block <unit> at <tmp> line 1␤␤»
18:45 | Zoffix | hm...
18:45 | TimToady | and it's not like the callsite is gonna be megamorphic
18:45 |   | Zoffix decides to leave the boat unrocked and closes the ticket

18:45 | TimToady | well, it's still LTA
-- | -- | --
18:45 |   | Zoffix renames the ticket instead
18:46 | TimToady | we wanna be able to optimize the heck out of scalar assignment, fershure
18:47 | TimToady | but if we've already gotta check for immutability, a further check in the error handling for list-assignability is probably not going to do much more than spread out cache hits from a longer goto
18:48 | TimToady | and perhaps multiple scalar assignments can share the same checking code out-of-line