surrogate code points in a Char
Hi. The Unicode Standard (version 4.0, section 3.9, D31 - pag 76) says: """Because surrogate code points are not included in the set of Unicode scalar values, UTF-32 code units in the range 0000D800 .. 0000DFFF are ill-formed""" However GHC does not reject this code units: Prelude> print '\x0000D800' '\55296' Is this a correct behaviour? Note that Python, too (2.5.4, UCS4 build, Linux Debian), accept these code units. Thanks Manlio
Enforcing a gap in the middle of the range of Char would be exceedingly
awkward to propagate through all of the libraries. Off the top of my head:
1.) Functions like succ and pred which currently work on Char as an
enumeration would have to jump over the gap, to be truly anal retentive
about the mapping
2.) The toEnum and fromEnum would need to make the gap vanish as well,
ruining the ability to treat toEnum/fromEnum as chr/ord
3.) Every application would take a performance hit
4.) What to do in the presence of an encoding error is even more uncertain.
All you can do is throw an exception that can only be caught in IO.
A couple of less defensible considerations:
5.) It would break alternative encodings like utf-8b which use the invalid
code points in the surrogate pair range to encode ill-formed bytes in the
input stream to allow 'cut and paste'-safe round tripping of
utf-8b->Char->utf-8b even in the presence of invalid binary data/codepoints.
6.) Not all data is properly encoded. Consider, Unicode data you get back
from Oracle, which isn't really encoded in UTF-8, but is instead CESU-8,
which encodes codepoints in the higher plane as a surrogate pair, then utf-8
encodes the surrogate pair.
So, I suppose the answer would be it is functioning as designed, because the
current behavior is the least bad option. =)
-Edward Kmett
On Wed, Nov 18, 2009 at 10:28 AM, Manlio Perillo
Hi.
The Unicode Standard (version 4.0, section 3.9, D31 - pag 76) says:
"""Because surrogate code points are not included in the set of Unicode scalar values, UTF-32 code units in the range 0000D800 .. 0000DFFF are ill-formed"""
However GHC does not reject this code units:
Prelude> print '\x0000D800' '\55296'
Is this a correct behaviour? Note that Python, too (2.5.4, UCS4 build, Linux Debian), accept these code units.
Thanks Manlio _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
On Nov 18, 2009, at 7:28 AM, Manlio Perillo wrote:
The Unicode Standard (version 4.0, section 3.9, D31 - pag 76) says:
"""Because surrogate code points are not included in the set of Unicode scalar values, UTF-32 code units in the range 0000D800 .. 0000DFFF are ill-formed"""
The current version of Unicode is 5.1. This text is now in D90, though otherwise the same. My references below are to the 5.1 documents (freely available on line at: http://www.unicode.org/versions/Unicode5.1.0/ )
However GHC does not reject this code units:
Prelude> print '\x0000D800' '\55296'
Is this a correct behaviour?
I don't think you should consider Char to be UTF-32. Think of Char as representing a Unicode code point. Unicode code points are defined as all in integers in the range \x0 through \x10FFFF, inclusive. Values in the range \xD800 through \xDFFF are all valid code points. (§2.4 in general; §3.4, D9, D10) Not all Unicode code points are "Unicode scalar values". Only Unicode scalar values can be encoded in the standard Unicode encodings. Unicode scalar values are defined a \x0 through \xD7FF and \xE000 through \x10FFFF - All code points except the surrogate pair area. (§3.9, D76) Not all code points are characters. In particular, \xFFFE, \xFFFF are "Noncharacters": They are representable in Unicode encodings, but should not be interchanged. Less well known is the range \xFDD0 though \xFDEF which are also noncharacters. (§2.4, Table 2-3; §3.4, D14, §16.7) In particular, note the stance of Unicode toward application internal forms: "Applications are free to use any of these noncharacter code points. internally but should never attempt to exchange them." - §16.7 ¶3 Accordingly, it is fine for Haskell's Char to support these values, as they are code points. The place to impose any special handling is in Haskell's various Unicode encoding libraries: When decoding, code points \xD800 through \xDFFF cannot be received, and noncharacters can be either retained or silently dropped (Unicode conformance allows this.) When encoding, code points \xD800 through \xDFFF and noncharacters should either error or just be silently dropped. - Mark Mark Lentczner http://www.ozonehouse.com/mark/ mark@glyphic.com
2009/11/25 Mark Lentczner
The current version of Unicode is 5.1. This text is now in D90, though otherwise the same. My references below are to the 5.1 documents (freely available on line at: http://www.unicode.org/versions/Unicode5.1.0/ )
It's been 5.2 for over a month now, I think.
participants (4)
-
Colin Adams -
Edward Kmett -
Manlio Perillo -
Mark Lentczner