[Carpet] CCTK_TraverseString: invalid group/variable name

[Luca Baiotti]

Erik,

I am aware of the problem of different users having different wishes and 
I don't think that the default solution can be good for everyone.

Indeed, I would suggest to add parameters for CarpetIO* to specify the 
warning level. There are other level 1 warnings which I don't care 
about, like

WARNING level 1 in thorn IOUtil processor 28 host node0017.admin
   (line 599 of 
/data1/baiotti/Cactus/configs/belladonna_NoDebug/build/IOUtil/CheckpointRecovery.c): 

   -> Recovery directory 
'/data20/baiotti/Meudon-45km-5lev-dx.25_noRefBoundInStar_IF/checkpoint' 
doesn't exist


(because I was setting recovery to autoprobe).

Did I convince you?


Luca



Erik Schnetter wrote:
> Luca,
> 
> people's opinions on whether a simulation should output a warning or 
> should abort when a problem is detected differs greatly.  Personally, I 
> prefer simulations to abort as early as possible if there is an error, 
> but it would not be appropriate to impose this opinion onto others.
> 
> Cactus has the notion of an "error level", which is the severest warning 
> level which is accepted before the simulation aborts.  It defaults to 
> zero.  If you set it to one, the simulation would abort on this error.
> 
> Alternatively, you can write stdout and stderr to different files.  If 
> you check stderr of your simulations regularly, then you will detect 
> these problems.  If you find that the stderr file is too large, then we 
> need to change this, because only severe problems should be reported 
> there.  Sometimes Cactus repeats the same error message multiple times 
> in there, which is not useful and buries the real errors.
> 
> I would prefer to have these generic mechanisms work.  If they don't 
> work, then we could introduce parameters into every thorn letting people 
> choose the warning level of its messages.  However, that wouldn't solve 
> the problem that Cactus should do something sensible by default.
> 
> -erik