Re: [INTERFACES] bug in ODBC driver ?

Byron Nikolaidis <byronn@insightdist.com> el día Mon, 14 Dec 1998 12:53:52
-0500, escribió:

>
>
>Sergio Kessler wrote:
>
>> Hi (Byron ?),
>>
>> I think I've found a bug in the postODBC driver, version
>> 6.40.0001.
>>
>> The problem is with the function SQLGetData.
>>
>> According to the docs of ODBC, when retrieving long data
>> (blobs), if the buffer is not enough, this function must
>> return SQL_SUCCESS_WITH_INFO and SQLSTATE 01004
>> (String data, right truncated) and in the variable
>> length_or_ind, it must return the _remaining_ byte
>> length of the data.
>>
>> The function is not working this way, ie I call it
>> (in Delphi):
>>
>> SQLGetData( hstmt, FieldNo, SQL_BINARY, PChar(TargetValue),
>>             15, @strlen_or_ind);
>>
>> Note that I put 15 as the buffer size, I _know_ that the
>> total size of the field (a text field in postgres) is 23
>> bytes, so this call must return SQL_SUCCESS_WITH_INFO
>> and SQLSTATE 01004 and the value of strlen_or_ind must
>> be 8.
>> But it return SQL_SUCCESS and strlen_or_ind = 23.
>>
>> Also, the driver, don't must write more bytes than
>> the buffer size, ie. if the buffer size is 15 the
>> driver should write no more than 15 bytes.
>>
>> Maybe this is the cause why other people are insulting the
>> BDE with blob fields (I don't use the BDE, btw)
>>
>> Hope this help.
>>
>
>Well, I looked into this.
>
>First, the multiple call logic *only* works on LONGVARBINARY fields.  In
>other words, you can only call SQLGetData multiple times for a field if
>its a Large Object type.  (At the time, this addition was awesome). It
>will not work on BINARY, or VARBINARY.  For those, it will simply return
>SQL_SUCCESS (it should return SQL_SUCCESS_WITH_INFO and data truncated).
>
>Second, if you do try to use multiple calls on LONGVARBINARY, it works,
>but it returns SQL_NO_TOTAL for the amount remaining.  Even though this
>is not very useful, it is valid according to the odbc spec if the driver
>doesn't know how much is left.
>
>So there really is nothing wrong, except the return for binary and
>varbinary types.  However, time permitting,  I will try to enhance the
>shortcomings in this area.

According with the Odbc spec (Odbc Sdk 3.5), the multiple call logic
must apply to _all_ length variable fields.

Only in the case of a fixed length field, the driver assume that
the buffer is enough and ignore the TargetValueLength parameter.

I'm using a Text type field, wich is clearly (I hope) variable
length, and as I'm using Borland tools, I unchecked the option
'Text as LongVarChar' in the driver.
I don't know how the driver manage this type of field internally.
Maybe I should check it ? But if it give me SQL_NO_TOTAL is not
very useful to me, as you said.

Regards,
Sergio Kessler