[#DERBY-4040] SQLChar.getLength returns wrong length for some data values

Derby

SQLChar.getLength returns wrong length for some data values

Created: 30/Jan/09 12:04 PM Updated: 01/Jul/09 04:25 PM

Component/s:

SQL

Affects Version/s:

10.3.3.1, 10.4.2.0, 10.5.1.1

Fix Version/s:

10.3.3.1, 10.4.2.1, 10.5.1.1

Time Tracking:

Not Specified

File Attachments:

File Name	Date Attached	Attached By	Size
derby-4040-1a-SQLChar_length_and_test.diff	2009-02-03 12:09 PM	Kristian Waagan	9 kB
derby-4040-1a-SQLChar_length_and_test.stat	2009-02-03 12:09 PM	Kristian Waagan	0.2 kB
derby-4040-2a-backport.diff	2009-02-18 05:02 PM	Kristian Waagan	8 kB

Issue Links:

Reference

This issue is related to:

~~DERBY-3741~~ SQL LENGTH function materializes CLOB into memory

Bug behavior facts:	Regression, Wrong query result
Resolution Date:	19/Feb/09 10:47 AM
Labels:

Description

« Hide

SQLChar.getLength() returns the wrong length of the data value in some cases.
For the bug to be triggered, the value must contain characters that are represented by two or three bytes, and Derby must store the value as a stream.

In my ad-hoc testing, I found that a value of length 10886 with CJK characters doesn't trigger the bug, but 10887 does trigger the bug. Further, for a range of lengths the bug can also been observed for CLOB columns (when doing 'select length(clobcolumn)').

All

Comments

Work Log

Change History

Subversion Commits

Sort Order:

[ Permalink | « Hide ]

Kristian Waagan added a comment - 30/Jan/09 12:10 PM

Bug was introduced by a checkin for ~~DERBY-3741~~.

[ Permalink | « Hide ]

Kristian Waagan added a comment - 03/Feb/09 12:09 PM

'derby-4040-1a-SQLChar_length_and_test.diff' fixes the bug by always decoding the stream is the data value is stored as a stream, and it also adds a few more tests (adjusted existing test).
The bug was that the byte length was used as a character length. Since these two lengths are the same for ASCII data values, it wasn't detected earlier. The limited range of the length information in the header helped conceal the bug too.

Ran regression tests without failures. I'll look at backporting the fix when it has made it into trunk.
Patch ready for review.

[ Permalink | « Hide ]

Kristian Waagan added a comment - 09/Feb/09 09:42 AM

Committed patch 1a to trunk with revision 742357.

[ Permalink | « Hide ]

Kristian Waagan added a comment - 18/Feb/09 05:02 PM

Attached the diff for the backport, 'derby-4040-2a-backport.diff'.
I had to do a manual merge, as the method InputStreamUtil.skipFully doesn't exist on the 10.4 and 10.3 branches. Instead of using that method, I simply chose to use InputStream.skip.
Regression tests ran successfully.

Committed to the 10.4 branch with revision 745571.

[ Permalink | « Hide ]

Kristian Waagan added a comment - 19/Feb/09 10:47 AM

Backported the fix from 10.4 to the 10.3 branch with revision 745811.

The tinderbox test for 10.4 ran cleanly.
I'm resolving the issue, and will await the tinderbox results from 10.3 before I close.

[ Permalink | « Hide ]

Kristian Waagan added a comment - 20/Feb/09 08:32 AM

10.3 tinderbox ran without any new failures (lang/subquery did fail).

Closing issue.