StringIO#ungetbyte stores incorrectly

[Nakilon]

$ rbenv shell 2.3.8
$ ruby -e "s = StringIO.new; s.ungetbyte(255); p [s.string, s.string.encoding, s.string.bytes, s.getc]"
["\xFF", #<Encoding:UTF-8>, [255], "\xFF"]

$ rbenv shell 3.0.1
$ ruby -rstringio \
       -e "s = StringIO.new; s.ungetbyte(255); p [s.string, s.string.encoding, s.string.bytes, s.getc]"
["\xFF", #<Encoding:UTF-8>, [255], "\xFF"]

$ rbenv shell truffleruby-21.1.0
$ ruby -e "s = StringIO.new; s.ungetbyte(255); p [s.string, s.string.encoding, s.string.bytes, s.getc]"
["ÿ", #<Encoding:UTF-8>, [195, 191], "ÿ"]

This results in several tests failing in my project.

[Nakilon]

And this one:

$ ruby -rstringio -e "Encoding::default_external = 'ASCII-8BIT'; s = StringIO.new; 1.times{ s.ungetbyte 255 }; puts :OK"
OK
$ ruby -rstringio -e "Encoding::default_external = 'ASCII-8BIT'; s = StringIO.new; 2.times{ s.ungetbyte 255 }; puts :OK"
.../truffleruby-21.1.0/lib/truffle/stringio.rb:615:in `ungetbyte': incompatible character encodings: UTF-8 and ASCII-8BIT (Encoding::CompatibilityError)
	from -e:1:in `block in <main>'
	from <internal:core> core/integer.rb:148:in `times'
	from -e:1:in `<main>'

[aardvark179]

I see the problem. The StringIO#ungetbyte is treating the byte as a character rather than raw bytes. Since we're commonly working in UTF8 this single byte is converted into a multibyte UTF8 encoding, and appended to the front of the string.

To illustrate this consider the string "\u01A9". This is encoded into the byte sequence 0xC60xA9. When a byte is read we get 0xc6, but if we try to unget that byte we append 0xC30x86 to the start of the string, because that's the UTF8 encoding of \u00C6.

This particular piece of our library looks like it can be simplified considerably as ungetbyte will only accept a single number, and will mask it to be a single byte.

[eregon]

This was fixed by @aardvark179 in a52bd0f