Map comprehension difference between compiler & interpreter

[michalmuskala]

Describe the bug
Given the following module:

-module(test).

-export([test/0]).

test() ->
  #{A => B || X <- [1, 5], {A, B} <- [{X, X+1}, {X, X+3}]}.

Running in a shell:

1> c(test).
{ok,test}
2> test:test().
#{1 => 4,5 => 8}
3> #{A => B || X <- [1, 5], {A, B} <- [{X, X+1}, {X, X+3}]}.
#{1 => 2,5 => 6}

Expected behavior
I'm not actually sure which result is correct - I expect the one from compiled code. Either way, the two should agree

Affected versions
Tested on latest master

[hajduakos]

I think you can further simplify the example:

-module(test).

-export([test/0]).

test() ->
  #{A => B || {A, B} <- [{1, 2}, {1, 3}]}.

And then:

1> c(test).
{ok,test}
2> test:test().
#{1 => 3}
3> #{A => B || {A, B} <- [{1, 2}, {1, 3}]}.
#{1 => 2}

Edit: https://www.erlang.org/eeps/eep-0058#semantics says that it should be equivalent to maps:from_list/1, and that seems to agree with the compiled version:

4> maps:from_list([{1, 2}, {1, 3}]).
#{1 => 3}