Destructuring assignment

[jbreeden]

Noticing some strange behavior when mruby is performing destructuring assingment inside of an interpolated string, and the assigned value comes from a local variable. Not sure if this belongs in the "not implemented" list or not. If so, let me know and I'll happy to close this and make a comment there instead.

Showing the issue below, copied from my terminal (comments added after). I'm running on Windows, for what it's worth.

C:\projects\mruby-bindings\mrbgems\mruby-nanomsg>..\..\mruby-1.1.0\bin\mirb.exe --version
mruby 1.1.0 (2014-11-19)

C:\projects\mruby-bindings\mrbgems\mruby-nanomsg>..\..\mruby-1.1.0\bin\mirb.exe
mirb - Embeddable Interactive Ruby Shell

# Assigning from array literal, everything looks good
> puts "#{@one, @two = [1, 2]}"  
[1, 2]
 => nil

# Making a variable to hold the array
> a = [1, 2]  
 => [1, 2]

# Perform the same `puts` as before.
 # With the variable, there seems to be some confusion and the current `this` is returned.
> puts "#{@one, @two = a}" 
main 
 => [main]

# Now things are really wacky, `a` seems to have lost it's value
> @one, @two = a  
 => nil

# Yep, that `a` is nil
> a  
 => nil

# Fixing `a`
> a = [1, 2]  
 => [1, 2]

# OK, sure enough, the assignment works outside of the string once `a` is restored
> @one, @two = a  
 => [1, 2]

Just for reference, MRI seems to handle this fine. I know MRuby isn't MRI (and I love that about it :), but I just wanted to make sure I wasn't doing something syntactically weird...

C:\projects\mruby-bindings\mrbgems\mruby-nanomsg>irb

# With an array literal...
irb(main):001:0> puts "#{@one, @two = [1, 2]}"
[1, 2]
=> nil
irb(main):002:0> a = [1, 2]
=> [1, 2]

# with an array variable
irb(main):003:0> puts "#{@one, @two = a}"
[1, 2]
=> nil

[jbreeden]

This may be a separate issue, but it seems closely related. Another test shows that being in an instance method is enough to throw off the destructuring.

Here we can see that the array_variable_test struggles with the deconstruction, returning only the last value in the array (instead of the entire array). This was also confirmed to work on MRI.

C:\projects\mruby-bindings\mrbgems\mruby-nanomsg>..\..\mruby-1.1.0\bin\mirb.exe
mirb - Embeddable Interactive Ruby Shell

> class Test
*   attr_accessor :a, :b
*
*   def array_lit_test
*     @a, @b = [1, 2]
*   end
*
*   def array_variable_test
*     ary = [1, 2]
*     @a, @b = ary
*   end
* end
 => :array_variable_test
> t = Test.new
 => #<Test:0xbcd970>
> t.array_lit_test
 => [1, 2]

# Why isn't this `[1, 2]` as in the previous test?
> t.array_variable_test
 => 2

The same test works outside of any methods on MRuby

C:\projects\mruby-bindings\mrbgems\mruby-nanomsg>..\..\mruby-1.1.0\bin\mirb.exe
mirb - Embeddable Interactive Ruby Shell

> ary = [1, 2]
 => [1, 2]
> one, two = ary
 => [1, 2]

[cremno]

Your first bug seems to be fixed on master. I can reproduce the second one. It's caused by the codegen's optimization (workaround: set mrbc_context::no_optimize)

[jbreeden]

Thanks, @cremno. Next time I'll check master before posting an issue.

By the way, does anybody know when another release might get tagged?