ogrinfo returns 'Corrupt data' error for wkt of length 1023 characters

[amayuk]

ogrinfo returns 'Error 1: GetNextRawFeatures(): Corrupt data' when used with a wkt feature which is of length 1023 characters.

This was seen with data in a SQL server 2008 express database. To create a table and populate it with test data run the following script:

CREATE TABLE [dbo].[snhd_wkt2](
	[ID] [int] NULL,
	[XMIN] [float] NULL,
	[YMIN] [float] NULL,
	[XMAX] [float] NULL,
	[YMAX] [float] NULL,
	[WKT] [varchar](max) NULL
) ON [PRIMARY]

INSERT INTO [dbo].[snhd_wkt2] VALUES
('1', '424137.474088898', '140734.844325094', '424381.173359863', '140870.555478181', 'Polygon ((424141.423 140736.164,424141.415 140736.164,424141.432 140736.174,424141.860 140736.574,424142.512 140737.224,424143.353 140738.223,424144.103 140739.283,424144.787 140740.403,424145.356 140741.582,424145.851 140742.792,424146.230 140744.052,424146.544 140745.311,424146.692 140746.611,424146.791 140747.921,424146.758 140749.230,424146.634 140750.530,424146.387 140751.829,424146.041 140753.089,424145.604 140754.309,424145.480 140754.589,424144.870 140755.568,424144.120 140757.328,424143.246 140759.127,424140.929 140763.916,424137.474 140770.954,424158.928 140773.253,424179.037 140778.462,424194.835 140782.231,424243.481 140812.932,424280.105 140836.445,424296.257 140847.032,424330.903 140864.627,424359.448 140870.555,424372.013 140868.036,424379.195 140862.118,424381.173 140848.472,424374.528 140833.036,424356.397 140826.748,424323.004 140824.059,424266.460 140798.386,424235.945 140782.051,424209.016 140767.695,424196.088 140751.180,424173.472 140739.683,424151.573 140734.844,424141.423 140736.164))')

Create an ODBC connection (in this case called 'SpatialTest' to the SQL Server database and run the following ogrinfo command (in this case with a user/password both of 'test'):

ogrinfo -ro -so "ODBC:test/test@spatialtest" snhd_wkt2

The following is returned:

ERROR 1: Column Geometry requested for geometry, but it does not exist.
INFO: Open of `ODBC:test/test@spatialtest'
      using driver `ODBC' successful.

Layer name: SNHD_wkt2
Geometry: Unknown (any)
ERROR 1: GetNextRawFeature(): Corrupt data
Feature Count: 1
ERROR 1: GetNextRawFeature(): Corrupt data
Layer SRS WKT:
(unknown)
ID: Integer (10.0)
XMIN: Real (0.0)
YMIN: Real (0.0)
XMAX: Real (0.0)
YMAX: Real (0.0)

If a blank character is entered into the WKT string (i.e. making is longer than 1023 characters) then ogrinfo works without any errors.

[warmerdam]

Hi, could you add a note (or set the version flag for this ticket) indicating what version of GDAL/OGR you are using? This sounds a bit like bug #990.

[warmerdam]

Roni,

Please make sure you note anything of significance that you learn about this problem.

[warmerdam]

Chaitanya,

Please attempt to reproduce this problem on SQL Server Express as time permits. It is low priority.

[ashrafroni]

I need all the table names and field names of that database. I am close to this bug. But this can occur only in special situation of database table.

[amayuk]

There is just one table in the database. The fields are described above.

[ashrafroni]

You have tried to create a database with 6 column but in the ogrinfo output this is looking like 5 column. Is it a copy error else you got right this output from ogrinfo?

[amayuk]

Yes, that's the correct output.

Further investigation shows this error occurs for wkt of length (x*511)+1, where x>1.

[lgao]

Yes, for a WKT of length x*511+1, cpl_odbc will fail to trim a '\0' before appending the last char ")". ogr2ogr also reports the same error.

Code: int CPLODBCStatement::Fetch( int nOrientation, int nOffset )

{ ...

if( cbDataLen > (int) (sizeof(szWrkData) - 1)

cbDataLen == SQL_NO_TOTAL )

{

nChunkLen = sizeof(szWrkData)-1; if (nFetchType == SQL_C_CHAR)

while ( (nChunkLen > 1)

&& (szWrkData[nChunkLen - 1] == 0) )

--nChunkLen; trimming the extra terminators

} else

nChunkLen = cbDataLen;

}

In the second to the last reading, cbDataLen = 512, sizeof(szWrkData)=512, so we skip the trimming. In the reading cbDatalen =1, and the char is ")". For example, a DBS WKT "POLYGON(('......'))" will becomes "POLYGON(('......')\0)" so 'Corrupt data'

[lgao]

Replying to lgao: oops, I meant sizeof(szWrkData)-1 = 512, so we skip the trimming...

[warmerdam]

Ashraf,

Do you have time to follow up on this? Or should I move it to Chaitanya?

[ashrafroni]

Frank, I am busy with opengraphrouter in GSoC can you please send this bug to Chaitanya. After this month I will be available from 1st July. If you allow me that time definitely I will solve this.

[chaitanya]

A part of this issue was handled in #1902. Only strings of length 511 were tested there. Thanks to amayuk and lgao's detailed description of the problem.

This problem occurs for any type of column of length (x*511)+1, where x>1.

Corrected in r17380 in trunk and r17381 in 1.6 and 1.5 branches.