Ruby String.encode still gives “invalid byte seque

Question

In IRB, I'm trying the following:

1.9.3p194 :001 > foo = "\xBF".encode("utf-8", :invalid => :replace, :undef => :replace)
 => "\xBF" 
1.9.3p194 :002 > foo.match /foo/
ArgumentError: invalid byte sequence in UTF-8
from (irb):2:in `match'

Any ideas what's going wrong?

Answer 1

If you're only working with ascii characters you can use

>> "Hello \xBF World!".encode('utf-8', 'binary', :invalid => :replace, :undef => :replace)
=> "Hello � World!"

But what happens if we use the same approach with valid UTF8 characters that are invalid in ascii

>> "¡Hace \xBF mucho frío!".encode('utf-8', 'binary', :invalid => :replace, :undef => :replace)
=> "��Hace � mucho fr��o!"

Uh oh! We want frío to remain with the accent. Here's an option that keeps the valid UTF8 characters

>> "¡Hace \xBF mucho frío!".chars.select{|i| i.valid_encoding?}.join
=> "¡Hace  mucho frío!"

Also in Ruby 2.1 there is a new method called scrub that solves this problem

>> "¡Hace \xBF mucho frío!".scrub
=> "¡Hace � mucho frío!"
>> "¡Hace \xBF mucho frío!".scrub('')
=> "¡Hace  mucho frío!"

Answer 2

I'd guess that "\xBF" already thinks it is encoded in UTF-8 so when you call encode, it thinks you're trying to encode a UTF-8 string in UTF-8 and does nothing:

>> s = "\xBF"
=> "\xBF"
>> s.encoding
=> #<Encoding:UTF-8>

\xBF isn't valid UTF-8 so this is, of course, nonsense. But if you use the three argument form of encode:

encode(dst_encoding, src_encoding [, options] ) → str

[...] The second form returns a copy of str transcoded from src_encoding to dst_encoding.

You can force the issue by telling encode to ignore what the string thinks its encoding is and treat it as binary data:

>> foo = s.encode('utf-8', 'binary', :invalid => :replace, :undef => :replace)
=> "�"

Where s is the "\xBF" that thinks it is UTF-8 from above.

You could also use force_encoding on s to force it to be binary and then use the two-argument encode:

>> s.encoding
=> #<Encoding:UTF-8>
>> s.force_encoding('binary')
=> "\xBF"
>> s.encoding
=> #<Encoding:ASCII-8BIT>
>> foo = s.encode('utf-8', :invalid => :replace, :undef => :replace)
=> "�"

Answer 3

This is fixed if you read the source text file in using an explicit code page:

File.open( 'thefile.txt', 'r:iso8859-1' )