[gdal-dev] severe memory leak when opening corrupt modis hdf (python bindings)

[Vincent Schut]

On 10/31/2013 12:29 PM, Vincent Schut wrote:
> Hi,
>
> when opening a corrupt hdf4 modis product (I've put an example here: 
> http://sarvision.nl/~schut/gdal_corrupt_hdf/MYD09GQ.A2010007.h11v09.005.2010010052202.hdf) 
> with python, there is a severe memory leak (multiple GB!)
>
> it is a dataset that gives "ERROR 4: 
> `/data/modis/MOLA/MYD09GQ.005/2010.01.07/MYD09GQ.A2010007.h11v09.005.2010010052202.hdf' 
> not recognised as a supported file format." when running gdalinfo on 
> it. When opening it with python (gdal.Open('...'), the result is None, 
> but there is a significant grow in memory usage that is only freed 
> after exiting python.
>
> Can someone confirm? Just repeatedly running ds = 
> gdal.Open('MYD09GQ.A2010007.h11v09.005.2010010052202.hdf') should do 
> it. Well, maybe not repeatedly. Running it once already gets my memory 
> usage more than 2GB higher...
>
> This is on gentoo linux, gdal from svn (updated yesterday), python 
> 2.7.5, hdf 4.2.9, and also tested on Arch Linux.
>
> Can someone confirm?
>
Update: it seems to be a hdf library problem, when I run hdfls on the 
file, it also grows my memory with several GB. I also tried other 
versions of libhdf (4.2.6, 4.2.8) but no difference...

>
> Best regards,
> Vincent.
> _______________________________________________
> gdal-dev mailing list
> gdal-dev at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/gdal-dev