This problem still exists in the hdf5 1.8.9 pre-release.

And contrary to my original subject, the problem isn't just on Windows. 

Is there any chance this can be fixed in 1.8.9?

Thanks,
Dave

Begin forwarded message:

> From: Dave Wade-Stein <[email protected]>
> Subject: Re: [Hdf-forum] h5diff 1.8.8 problem on Windows
> Date: February 23, 2012 1:50:56 PM MST
> To: HDF Users Discussion List <[email protected]>
> 
> Jonathan,
> 
> Thanks for the response.
> 
> The diff that h5diff reports is for the taggedPtcls, not charge attribute, 
> and when I used h5dump to dump the files, those data are exactly the same, 
> except for the type. Below are the excerpts from the h5dump of each file:
> 
>       ATTRIBUTE "currentTags" {
>          DATATYPE  H5T_STD_U32LE
>          DATASPACE  SIMPLE { ( 4 ) / ( 4 ) }
>          DATA {
>          (0): 204, 205, 2, 3
>          }
>       }
> 
>       ATTRIBUTE "currentTags" {
>          DATATYPE  H5T_STD_U64LE
>          DATASPACE  SIMPLE { ( 4 ) / ( 4 ) }
>          DATA {
>          (0): 204, 205, 2, 3
>          }
> 
> If differing datatypes cause h5diff to get "confused" and report diffs that 
> aren't there, shouldn't h5diff refuse to do the diff on the data with 
> differing datatypes?
> 
> It would be much better if the output of h5diff were:
> 
> Warning: different storage datatype
> <currentTags> has file datatype H5T_STD_U32LE
> <currentTags> has file datatype H5T_STD_U64LE
> Unable to compare attribute currentTags due to differing datatypes.
> 
> Instead, what we get is
> 
> Warning: different storage datatype
> <currentTags> has file datatype H5T_STD_U32LE
> <currentTags> has file datatype H5T_STD_U64LE
> attribute: <currentTags of </taggedptclsCell>> and <currentTags of 
> </taggedptclsCell>>
> size:           [4]           [4]
> position        currentTags of </taggedptclsCell> currentTags of 
> </taggedptclsCell> difference          
> ------------------------------------------------------------
> [ 1 ]          205             0               205            
> [ 2 ]          2               205             203            
> [ 3 ]          3               0               3       
> 
> So, h5diff is saying the values in the second file are (204, 0, 205, 0) when 
> in fact they are (204, 205, 2, 3) as shown in my h5dump excerpt above.
> 
> Thanks,
> Dave
> 
> On Feb 23, 2012, at 12:51 PM, Jonathan Kim wrote:
> 
>> Hi Dave,
>> 
>> According to the verification, h5diff 1.8.7 had a bug and h5diff 1.8.8 is 
>> correct.
>> The h5diff 1.8.7 didn't catch such differences from attributes and it got 
>> fixed for h5diff 1.8.8.
>> The warning is simply informational purpose from the type size difference, 
>> so user can be aware of the detail.
>> 
>> If you check data in the 'charge' attribute of the '/taggedptcls' dataset 
>> from both files, you will see those attributes contain the difference data.
>> 
>> Jonathan
>> 
>> 
>> On 2/22/2012 9:30 PM, Dave Wade-Stein wrote:
>>> 
>>> The two files being compared are not in fact, different. But I've noticed 
>>> that h5diff 1.8.8 fails when comparing H5T_STD_U32LE to H5T_STD_U64LE (see 
>>> bold text below). When I see that error message, I know it's going to be 
>>> wrong.
>>> 
>>> dws@qawin32 /winsame/dws/vorpalall/vptests/tests/ptcls
>>> $ C:/winsame/contrib-vs9/hdf5-1.8.8-par/bin/h5diff -v  
>>> taggedParticles2p_taggedptcls_2.h5 
>>> C:/winsame/dws/vorpalall/vptests/vpresults-win32/ptcls/taggedParticles2p/taggedParticles2p_taggedptc
>>>     ls_2.h5 /taggedptcls
>>> dataset: </taggedptcls> and </taggedptcls>
>>> attribute: <charge of </taggedptcls>> and <charge of </taggedptcls>>
>>> 0 differences found
>>> Warning: different storage datatype
>>> <currentTags> has file datatype H5T_STD_U32LE
>>> <currentTags> has file datatype H5T_STD_U64LE
>>> attribute: <currentTags of </taggedptcls>> and <currentTags of 
>>> </taggedptcls>>
>>> size:           [4]           [4]
>>> position        currentTags of </taggedptcls> currentTags of </taggedptcls> 
>>> difference              
>>> ------------------------------------------------------------
>>> [ 1 ]          289             0               289            
>>> [ 2 ]          2               289             287            
>>> [ 3 ]          3               0               3              
>>> 3 differences found
>>> …
>>> 
>>> Now if I use h5diff 1.8.7 to diff those same files, I get the correct 
>>> answer (and no complaint about datatypes):
>>> 
>>> dws@qawin32 /winsame/dws/vorpalall/vptests/tests/ptcls
>>> $ C:/winsame/contrib-vs9/hdf5-1.8.7-ser/bin/h5diff -v  
>>> taggedParticles2p_taggedptcls_2.h5 
>>> C:/winsame/dws/vorpalall/vptests/vpresults-win32/ptcls/taggedParticles2p/taggedParticles2p_taggedptcls_2.h5
>>>  /taggedptcls
>>> dataset: </taggedptcls> and </taggedptcls>
>>> attribute: <charge of </taggedptcls>> and <charge of </taggedptcls>>
>>> 0 differences found
>>> attribute: <mass of </taggedptcls>> and <mass of </taggedptcls>>
>>> 0 differences found
>>> attribute: <numPtclsInMacro of </taggedptcls>> and <numPtclsInMacro of 
>>> </taggedptcls>>
>>> 0 differences found
>>> attribute: <numSpatialDims of </taggedptcls>> and <numSpatialDims of 
>>> </taggedptcls>>
>>> 0 differences found
>>> attribute: <ptclEnds of </taggedptcls>> and <ptclEnds of </taggedptcls>>
>>> 0 differences found
>>> attribute: <time of </taggedptcls>> and <time of </taggedptcls>>
>>> 0 differences found
>>> attribute: <vsLimits of </taggedptcls>> and <vsLimits of </taggedptcls>>
>>> 0 differences found
>>> attribute: <vsNumSpatialDims of </taggedptcls>> and <vsNumSpatialDims of 
>>> </taggedptcls>>
>>> 0 differences found
>>> attribute: <vsTimeGroup of </taggedptcls>> and <vsTimeGroup of 
>>> </taggedptcls>>
>>> 0 differences found
>>> attribute: <vsType of </taggedptcls>> and <vsType of </taggedptcls>>
>>> 0 differences found
>>> 
>>> I see this problem on both 32-bit and 64-bit Windows. Has anyone else seen 
>>> it?
>>> 
>>> Dave
> 
> 
> 

Dave Wade-Stein, Ph.D.
Software Developer II
Quality Assurance and Releases Group
Tech-X Corporation      
[email protected]





_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org

Reply via email to