[Carpet] CarpetIOHDF5 testsuites are broken

[Luca Baiotti]

Hi,

I have investigated further the CarpetIOHDF5 testsuite problem.

I have found out that, removing all recent patsches (in darcs) back to

Tue May  9 15:06:37 CEST 2006  Erik Schnetter <schnetter at cct.lsu.edu>
   * Carpet: Add aliased function QueryProlongating()

(included in the unpull) and going back to the corrispondent cvs version 
of cactus, the testsuite passes (I am referring to the
Carpet/CarpetIOHDF5/test/CarpetWaveToyRecover_test_1proc.par testsuite).

After applying the above patch the testsuite fails. The data (e.g. 
phi.x.asc) differ at the initial time (after recovery) of as much as 
0.1% and are visible to the eye in ygraph without zooming.

The data produced by running the testsuite with all the patches up to 
the one above included and the data produced with all the patches (up to 
today) coincide; (i.e. there have been no more changes in the results 
produced by running the testuite from the above patch up to now; and the 
testsuite data in the repository have not been changed since Feb 6th 2006.

Can someone more expert than I further investigate why the above patch 
broke the testuite? And commit a fix?


Ciao,

Luca.


Thomas Radke wrote:
> Erik Schnetter wrote:
> 
>>The parameters regrid_during_recovery and  
>>use_grid_structure_from_checkpoint are supposed to be used together.   
>>Using only one of them will probably lead to inconsistencies.  
> 
> 
> Sorry, forgot to mention this in my previous email: Luca and me did use 
> both parameters.
> 
> 
>>>Erik, when you pushed this testsuite in May, was it passing for you ?
>>
>>
>>Yes, it did.  At least I think so.
>>
>>
>>>I
>>>had to unpull all patches up to "CarpetLib: Do not restrict to points
>>>they are used for boundary prolongation" from 26 February in order to
>>>get the other two testsuites passing again.
>>
>>
>>The new recovery test case fails for me as well.
> 
> 
> How about the other two CarpetIOHDF5 testsuites ? Could you check when 
> they broke for you ? Was it after this new feature had been introduced 
> (beginning of May) or before (back in February, with the "CarpetLib: Do 
> not restrict to points they are used for boundary prolongation" patch) ?
> 
> As a side note: can we have the Carpet development version periodically 
> tagged in darcs, say, every month when you send out the status report ? 
> I think that would simplify the process of reverting back to an older 
> state. Currently I have to do this on a patch basis. But usually I don't 
> know what patch could have broken something, I only know (with some 
> certainty) that something still used to work at a specific time.