Referencing result of earlier segment when pattern matching generates invalid Erlang

[richard-viney]

The following code references the result of the pattern's first segment in its second segment:

pub fn main() {
  let assert <<length, _:size(length)-unit(8)>> = <<3, 1, 2, 3>>
}

Building for Erlang gives:

/bit_array_test/src/bit_array_test.gleam:6:16: variable 'Length' is unbound
error: Shell command failure

Erlang can handle this kind of self-referencing pattern so Gleam probably should too.

The following Erlang works as expected:

test_pattern() ->
    _assert_subject = <<3, 1, 2, 3>>,
    case _assert_subject of
        <<Length, _:Length/unit:8>> -> _assert_subject;
        _assert_fail -> erlang:error()
    end.

Gleam version: 1.6.3 and 1.7.0-rc3.

[lpil]

What Erlang does it compile to today?

[richard-viney]

main() ->
    _assert_subject = <<3, 1, 2, 3>>,
    <<Length@1, _:Length@1/unit:8>> = case _assert_subject of
        <<_, _:Length/unit:8>> -> _assert_subject;
        _assert_fail ->
            erlang:error(#{gleam_error => let_assert,
                        message => <<"Pattern match failed, no pattern matched the value."/utf8>>,
                        value => _assert_fail,
                        module => <<"bit_array_test"/utf8>>,
                        function => <<"main"/utf8>>,
                        line => 17})
    end.

If the match inside the case captured the first segment into a Length variable similar to the assignment on the previous line then that would likely fix it.

Is this doing the pattern match twice? Or does Erlang optimise that away?

[lpil]

Ah, thank you.

I believe it does not get optimised.