MySQL Bugs: #55024: Wrong type returned by SQLColAttribute(SQL_DESC

Bug #55024	Wrong type returned by SQLColAttribute(SQL_DESC_PRECISION...) in 64-bit Linux
Submitted:	6 Jul 2010 10:59	Modified:	13 Aug 2010 15:08
Reporter:	Bogdan Degtyariov	Email Updates:
Status:	Closed	Impact on me:	None
Category:	Connector / ODBC	Severity:	S3 (Non-critical)
Version:	5.1.6	OS:	Linux (x64)
Assigned to:	Bogdan Degtyariov	CPU Architecture:	Any
Tags:	64-bit, SQL_DESC_PRECISION

Description:
SQLColAttribute(SQL_DESC_PRECISION...) function returns wrong type identifiers that have negative value:

#define SQL_LONGVARCHAR       (-1) returned 4294967295
#define SQL_BINARY            (-2) returned 4294967294
#define SQL_VARBINARY         (-3) returned 4294967293
#define SQL_LONGVARBINARY     (-4) returned 4294967292
#define SQL_BIGINT            (-5) returned 4294967291
#define SQL_TINYINT           (-6) returned 4294967290
#define SQL_BIT               (-7) returned 4294967289

Obviously, they are returned as 32-bit unsigned integer.
This only happens in 64-bit Linux.

How to repeat:
See the test case uploaded below.

Suggested fix:
just fix it

test case

Attachment: test55024.c (text/plain), 2.76 KiB.

patch

Attachment: patch55024.diff (application/octet-stream, text), 4.74 KiB.

Bogdan, I remember we discussed this bug and agreed on solution. But now i'm pretty sure i opinion on the reason of the bug was wrong. Now i'm not even sure that is a bug.
See on the http://msdn.microsoft.com/en-us/library/ms713558%28v=VS.85%29.aspx
"SQLColAttribute returns descriptor information for a column in a result set. Descriptor information is returned as a character string, a 32-bit descriptor-dependent value, or an integer value."

i think if you you pass in your testcase SQLINTEGER value to the SQLColAttribute things should work fine. It should(i didn't test - only looked the code:)) fails only if SQLLEN is no of the same size with SQLINTEGER, i.e. - on 64bit systems(i pushed the testcase and you can see that it fails on win64 as well).
This SQLColAttribute is some alchemy. My interpretation that SQLLEN* parameter doesn't have to be really pointer to a SQLLEN. Otherwise many things have little sense. They don't have much either. btw corresponding descriptor field is SQLSMQLLINT. but if one passes SQLSMALLINT pointer - driver should crash...

Lawrin,

Thanks for the comment, but we have to make some decision. Crashing driver is not good in my opinion... returning corrupted result is even worse.

the patch has been pushed as rev#908. it fixes treating numeric value buffer as 32bit on 64bit platforms. That's not quite obvious from the specs, but other drivers do so, and parameter type is SQLLEN* anyway. Thus we changed that too.

An entry has been added to the 5.1.7 changelog:

SQLColAttribute(SQL_DESC_PRECISION...) function returned incorrect results for type identifiers that have a negative value:

#define SQL_LONGVARCHAR       (-1) returned 4294967295
#define SQL_BINARY            (-2) returned 4294967294
#define SQL_VARBINARY         (-3) returned 4294967293
#define SQL_LONGVARBINARY     (-4) returned 4294967292
#define SQL_BIGINT            (-5) returned 4294967291
#define SQL_TINYINT           (-6) returned 4294967290
#define SQL_BIT               (-7) returned 4294967289

They were returned as 32-bit unsigned integer values. This only happened on 64-bit Linux.