Opened 6 years ago

Closed 6 years ago

#4196 closed defect (fixed)

HDF5 Identify does not handle files with user data at the beginning

Reported by: rprinceley Owned by: warmerdam
Priority: normal Milestone: 1.9.0
Component: GDAL_Raster Version: svn-trunk
Severity: normal Keywords:
Cc: gaopeng, antonio

Description

In an HDF5 file, the superblock need not start at beginning, we are required to search for it at specific offsets. HDF5Dataset::Identify should probably use H5Fis_hdf5 function to identify an HDF rather than look for the signature in poOpenInfo->pabyHeader.

We have some NOAA data that does not work with GDAL because the superblock starts at 1024 (prefixed with XML).

Change History (5)

comment:1 Changed 6 years ago by Even Rouault

The issue is that Identify() is supposed to run fast and not open any file, what H5Fis_hdf5() would do. H5Fis_hdf5() calls H5F_locate_signature() which has this interesting info :

/*-------------------------------------------------------------------------
 * Function:	H5F_locate_signature
 *
 * Purpose:	Finds the HDF5 superblock signature in a file.	The signature
 *		can appear at address 0, or any power of two beginning with
 *		512.

Do you have a link to those NOAA data ? Perhaps we could call H5Fis_hdf5() only if the extension is .h5 if those files do use it ?

comment:2 Changed 6 years ago by antonio

Cc: antonio added

comment:3 Changed 6 years ago by rprinceley

I have checked in the change made to support NOAA in 1.8-esri (r22971). These HDFs start with an XML block "<HDF_UserBlock>", the Identify function looks for this and calls H5Fis_hdf5. We are using H5Fis_hdf5 to search for the superblock in this case to avoid assumptions about the XML block size.

    if( poOpenInfo->pabyHeader )
    {
        if( memcmp(poOpenInfo->pabyHeader,achSignature,8) == 0 )
            return TRUE;

        if( memcmp(poOpenInfo->pabyHeader,"<HDF_UserBlock>",15) == 0)
        {
            if( H5Fis_hdf5(poOpenInfo->pszFilename) )
              return TRUE;
        }
    }

Will try to attach a sample to this ticket (checking if we have permission to distribute samples).

comment:4 Changed 6 years ago by Even Rouault

Milestone: 1.9.0

r22973 /trunk/gdal/frmts/hdf5/hdf5dataset.cpp: HDF5: Fix to identify datasets whose header doesn't start with the typical signature but with some XML content (#4196)

comment:5 Changed 6 years ago by Even Rouault

Resolution: fixed
Status: newclosed
Note: See TracTickets for help on using tickets.