wiki:rfc23_ogr_unicode

Version 2 (modified by warmerdam, 16 years ago) ( diff )

mark as proposed.

RFC 23: Unicode support in OGR

Authors: Frank Warmerdam
Contact: warmerdam@…
Status: Proposed

Summary

This document proposes preliminary steps towards GDAL/OGR handling strings internally in UTF-8, and supporting conversion between different encodings.

Main concepts

GDAL should be modified in a way to support three following main ideas:

  1. The CPLString class will be upgraded to support a variety of encoding conversions, including conversion between representations (ie. UTF-8 to UCS-16/wchar_t).
  2. Character encodings will be identified by iconv() style strings.
  3. OFTString/OFTStringList feature attributes in OGR will be treated as being in UTF-8.

This RFC specifically does not attempt to address issues of using non-ascii filenames. It also does not attempt to make definitions about the encoding of other strings used in GDAL/OGR (such as field names, metadata, etc). These would presumably be addressed in a later RFC building on this one.

CPLString

The CPLString class will now be assumed to be a potentially multi-byte encoded string, but with no effect within the CPLString to keep track of the encoding it is in. This is left to the higher level code for now.

However, the CPLString is extended with some convenient mechanisms for recoding, and for conversion of UTF-8 to/from "wchar_t" (aka UCS-2).

It is stressed that normal initialization of a CPLString from "const char *" does not attempt to do any conversions to/from UTF-8. This rule is kept, in part to minimize string processing costs for the common case. When encoding is believed to be an issue the calling code must keep track. 

// Convert the internal string to a new encoding.
char* CPLString::recode( const char *pszSrcEncoding, const char *pszDstEncoding );

// Set value based on input encoded string with CPLString set to UTF-8. 
// This is equivelent to normal setting, and then a recode() with a destination
// encoding of "UTF-8" and thus is just for convenience.

CPLString &CPLString::SetAsUTF8( const char *pszInput, const char *pszEncoding = "" );

// Set value based on input encoded string with CPLString set to UTF-8. 
// This is equivelent to normal setting, and then a recode() with a destination
// encoding of "UTF-8" and thus is just for convenience.

CPLString &CPLString::SetAsUTF8( const wchar_t *pszInput, const char *pszEncoding = "UCS-2" );

// Construct UTF-8 string object from array of wchar_t elements. 
CPLString::CPLString( const wchar_t*pszInput, const char *pszEncoding = "UCS-2" );

// Returns a wchar_t string which becomes the ownership responsibility of
// the caller (free with CPLFree()).  It is assumed the CPLString is UTF-8. 
wchar_t *CPLString::GetAsWChar( const char *pszDstEncoding = "UCS-2" );

I have specifically avoided additional constructors or casting operators to to avoid any possible overloading ambiguities or complication in maintaining extra state in the CPLString. Such services can be added in the future based on the above methods if desired.

Encoding Names

It is proposed that the encoding names will be the same sorts of names used by iconv(). So stuff like "UTF-8", "LATIN5", "CP850" and "ISO_8859-1". It does not appear that these names for encodings are a 1:1 match with C library locale names (like "en_CA.utf8" for instance) which may cause some issues.

Some particular names of interest:

  • "": The current locale. Use this when converting from/to the users locale.
  • "UTF-8": Unicode in multi-byte encoding. Most of the time this will be

our internal linga-franca.

  • "POSIX": I think this is roughly ASCII (perhaps with some extended characters?).
  • "UCS-2": Two byte unicode. This is a wide character format and only suitable for use with the wchar_t methods.

On some systems you can use "iconv --list" to get a list of supported encodings.

iconv()

It is proposed to implement the CPLString::recode() method using the iconv() and related functions when available.

There is an excellent implementation of this API as GNU libiconv(), which is used by the C libraries on Linux. Also some operating systems provide the iconv() API as part of the C library (all unix?); however, the system iconv() often has a restricted set of conversions supported so it may be desirable to use libiconv in preference to the system iconv() even when it is available.

If iconv() is not available, a stub implementation of the CPLString services will be provided which:

  • implements UCS-2 / UTF-8 interconversion using either mbtowc/wctomb, or an implementation derived from <a href="​http://www.cl.cam.ac.uk/~mgk25/unicode.html">​http://www.cl.cam.ac.uk/~mgk25/unicode.html</a>.
  • Implements recoding from "" to and from "UTF-8" by doing nothing, but issuing a warning on the first use if the current locale does not appear to be the "C" locale.
  • Implements recoding from "ASCII" to "UTF-8" as a null operation.
  • Implements recoding from "UTF-8" to "ASCII" by turning all non-ASCII multi-byte characters to '?'.

This hopesfully gives us a weak operational status when built without iconv(), but full operation when it is available.

The --with-iconv=<arg> option will be added to configure. The argument can be the path to a libiconv installation or the special value 'system' indicating that the system lib should be used. Alternatively, --without-iconv can be used to avoid using iconv.

OFTString/OFTStringList Fields

It is declared that OGR string attribute values will be in UTF-8. This means that OGR drivers are responsible for translating format specific representations to UTF-8 when reading, and back to the format specific representation when writing. In many cases (of simple ASCII text) this requires no transformation.

This implies that the arguments to methods like OGRFeature::SetField( int i, const char *) should be UTF-8, and that GetFieldAsString() will return UTF-8.

The same issues apply to OFTStringList lists of strings. Each string will be assumed to be UTF-8.

OGR Driver Updates

The following OGR drivers could benefit immediately from recoding to UTF-8 support in one way or another.

  • ODBC (add support for wchar_t / NVARSHAR fields)
  • Shapefile
  • GML (I'm not sure how the XML encoding values all map to our concept of encoding)
  • Postgres

I'm sure a number of the other drivers, particularly the RDBMS drivers, could benefit from an update.

Implementation

Frank Warmerdam will implement the core iconv() capabilities, the CPLString additions and update the ODBC driver. Other OGR drivers would be updated as time and demand mandates to conform to the definitions in this RFC by interested developers.

The core work will be completed for GDAL/OGR 1.6.0 release.

References

Note: See TracWiki for help on using the wiki.