Re: [bgl-discuss] call flush (6) does nothing

[Rob Ross <rross@xxxxxxxxxxx>]

On Mon, 7 Feb 2005, Ray Loy wrote:

> > this is probably a trend that you will see more of, rather than less.  
> ...
> > you might want to use mpi-io to explicitly write data rather than using 
> > stdout. 
> 
> unfortunately, one doesn't usually have control over libraries, etc.
> (or have the inclination to rewrite them).  seeing a warning or error
> that comes out on stdout/stderr in real time (or something close to
> it) can save a lot of wasted hours...

i agree that correctly receiving error messages from libraries in a timely
manner is very important.

however, that's what we are supposed to use error output variables (or
return codes) for.  then detection of problems can be handled by the 
library and responding to the error can be handled by the application 
(programmatically, rather than the user eyeballing the output).  one way 
to respond would be to print an error.

in that ideal world, we would just drop in a replacement "error handling
subsystem" and we would be done.  and the mpi implementor could spend
his/her time making communication more efficient rather than figuring out
how to best organize and label stdout from 64k processes.

unfortunately, the fact is that libraries and applications do sometimes
just print messages and mpi implementations are less likely to cater to
stdout over time.  actually the real problem is likely to be that MPI
implementations handle stdin/stdout in different ways on different
platforms, or in a way that is inconvenient for a particular application
to parse.

probably users will just end up jumping through more hoops for a while.  
sorry!

rob

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.