Bug #21528: SyntaxError#message may have broken encoding with multibyte source under Prism

• Status: Open
• Assignee: -
• Target version: -
• Backport: 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: UNKNOWN

[alpaca-tc (Hiroyuki Ishii)]

Since the introduction of Prism, when parsing Ruby source code that contains multibyte characters, SyntaxError#message can sometimes have invalid encoding.
Here is a reproducible example:

begin
RubyVM::InstructionSequence.compile(<<~CODE, nil, nil, 1)
          if a
          # 0000000000000ああああああ
          #
CODE
rescue SyntaxError => e
$e = e
  puts e.message # string contains a multibyte character that is cut off mid-byte. \xE3
  # <compiled>:3: syntax errors found
  #   1 | if a
  # > 2 | # 0000000000000あああああ\xE3 ...
  #     | ^ expected an `end` to close the conditional clause
  # > 3 | #
  #     |  ^ unexpected end-of-input, assuming it is closing the parent top level context

  puts e.message.valid_encoding? #=> expected true, but got false
end

This appears to be caused by a truncation process in prism's error message generating that does not consider multibyte character boundaries.
See: The truncation logic around prism_compile.c L10696-L10709
I'm not sure how to correctly fix it due to lack of knowledge about safe byte truncation.

I discovered this issue through irb, which attempts to display source code even when it contains syntax errors. Because irb uses SyntaxError#message, it raised an ArgumentError: invalid byte sequence in UTF-8. See: https://github.com/ruby/irb/blob/f60dfa8549f746f69e9a6d160604a7a4974ffac1/lib/irb/ruby-lex.rb#L255-L256

If this is considered an irb issue, I already have a patch for IRB that handles it.