Re: Status of UTF-8 support in 0.83 dev

> 
> Frediano,
> 
> Sorry my YES was too fast: I should have checked with a 
> simple ODBC program.
> 
> Even inserting does not fully work for now:
> 
> When binding with a small ColumnSize (like 2), I get invalid 
> data in the
> SQL Server Profiler trace (only pieces of the value).
> 
> Looks like the conversion from UTF-8 to UCS-2 relies on the ColumnSize
> and does not work properly when that size is small...
> (Maybe the conversion buffer used is allocated according to 
> ColumnSize,
> as a number of bytes?)
> 

Bingo, yes, code in src/odbc/sql2tds.c.
Oh... I committed patch I sent you yesterday.

> See attached example: Using ColumnSize=2, it does not insert all data,
> but when using ColumnSize>=6 it works...
> 

I updated utf8 test, see 
http://freetds.cvs.sourceforge.net/freetds/freetds/src/odbc/unittests/utf8.c?r1=1.3&r2=1.4
currently it doesn't work, it give 

$ ./utf8
IF OBJECT_ID(N'mytab王鴻') IS NOT NULL DROP TABLE mytab王鴻
CREATE TABLE mytab王鴻 (k int, c NCHAR(10), vc NVARCHAR(10))
INSERT INTO mytab王鴻 VALUES (1,N'aaa',N'aaa')
INSERT INTO mytab王鴻 VALUES (2,N'abcéáô',N'abcéîô')
INSERT INTO mytab王鴻 VALUES (3,N'abc王鴻',N'abc王鴻傑王鴻傑')
DELETE FROM mytab王鴻
insert #1
insert #2
insert #3
DELETE FROM mytab王鴻
insert #1
insert #2
utf8.c:106 SQLExecute
SQL error 42000 -- [FreeTDS][SQL Server]Character set conversion is not available between client character set '%.*s' and server character set '%.*s'


mmm... error is a bit wrong...

> Binding first value (4 bytes) with ColumnSize=4, it works, and second
> value (6 bytes) it works with ColumnSize=6...
> 
> Since ColumnSize defines a number of CHARACTERS, something is 
> wrong here.
> 
> Seb
> 

I got it working replacing (sql2tds.c)

    } else {
        if (is_unicode_type(dest_type))
            curcol->on_server.column_size = curcol->column_size * 2;
        else
            curcol->on_server.column_size = curcol->column_size;
    }

with

    } else {
        if (is_unicode_type(dest_type)) {
            curcol->on_server.column_size = curcol->column_size * 2;
            curcol->column_size *= 4;
        } else
            curcol->on_server.column_size = curcol->column_size;
    }


It merely allocate correct buffer... I think in this case it would be better to either
- avoid buffer allocation, fill column_data if possible... I think the main problem here is 
- use server encoding for buffer, I think this is related to bcp todo ("Native bcp has no iconv support; character bcp files are assumed be encoded with the client's charset.  More flexibility one both sides would be good.").
I think that column size == count in bytes is required cause:
- its possible to compute characters number from byte count and buffer but it's not easy from characters count and buffer to byte count (it require terminated strings)
- currently on_server.column_size is always a byte count
- memcpy/memmove it's easier if you have byte count
- client libraries provide byte count for strings and buffers

There are some cases where however having different encoding in libTDS buffer is useful
- having different client encoding
- avoiding converting characters twice (for instance if client use multibyte and wide at the same time as odbc does)

freddy77

_______________________________________________
FreeTDS mailing list
[email protected]
http://lists.ibiblio.org/mailman/listinfo/freetds