[Carpet] recovering grid structure from checkpoint

[Luca Baiotti]

Hi,

I would like to stress that the 
Carpet/CarpetIOHDF5/test/CarpetWaveToyNewRecover_test_1proc.par
and
Carpet/CarpetIOHDF5/test/CarpetWaveToyRecover_test_1proc.par
have as the only difference:

14,16d13
< Carpet::regrid_during_recovery                   = no
< CarpetIOHDF5::use_grid_structure_from_checkpoint = yes
<

and that they read the same checkpoint file (and that the checkpoint 
file was produced after the above parameters were introduced). But their 
results are different, as Thomas has checked.

So, even if the provided functionality is desirable, I suggest everybody 
not to set this parameters as above, until the problem is fixed. (I and 
Thomas do not know how to fix this problem)


Ciao,

Luca.



Thomas Radke wrote:
> Hi,
> 
> I fixed two of the three CarpetIOHDF5 testsuites which test 
> checkpointing/recovery. I found that the third one, which also recovers 
> the grid structure from the checkpoint file, gives different results 
> compared to the default grid structure setup procedure. While all the 1D 
> output files are identical, differences occur in all norms, eg.
> 
> 
>>diff /home/tradke/cactus/arrangements/Carpet/CarpetIOHDF5/test/CarpetWaveToyNewRecover_test_1proc/phi.norm1.asc /home/tradke/cactus/TEST/wave/CarpetIOHDF5/CarpetWaveToyNewRecover_test_1proc/phi.norm1.asc
>>
>>
>>1,3c1,3
>>< 64 0.421052631578947 0.845190454122255
>>< 96 0.631578947368421 0.817531253193611
>>< 128 0.842105263157895 0.795234619496462
>>---
>>
>>>64 0.421052631578947 0.845251782965402
>>>96 0.631578947368421 0.817538528282232
>>>128 0.842105263157895 0.795164735804112
> 
> 
> Bug or feature ?