Motivation

Encoding::UTF_32 doesn't actually encodes strings correctly and we were calculating code points with a wrong encoding. It's UTF_32LE that encodes the string and provides us the right locations back.

Implementation

Started using Encoding::UTF_32LE, which applies the correct encoding to the string. It's easy to verify that it was previously wrong by doing this:

# The length should be based on codepoints (42)

# Using the wrong encoding
"class Foo; end\n\"🙂\"; Foo\n".encode(Encoding::UTF_32).length
# => 100

# Using the right encoding
"class Foo; end\n\"🙂\"; Foo\n".encode(Encoding::UTF_32LE).length
# => 24

Automated Tests

Added a test.