Thanks Roy, I'll take a look at the PETSc error handler. Thanks Jed:
On Fri, Feb 7, 2014 at 5:30 PM, Jed Brown <[email protected]> wrote: > > This is a tough question. PETSc's error handler is invoked collectively > on a communicator. If you call a function that is collective on comm1, > but the error is invoked on COMM_SELF (or some comm2---necessarily > smaller than comm1), it means that we don't have collective information > about the error, which is very difficult to recover from (i.e., > consistency checks that would allow recovery would also be expensive). > if it helps, I'm using only CommWorld > > Francesco, what errors do you want to recover from? There is a > difference between errors as a result of logic errors/incorrect > programming/incorrect problem specification, semantic incompatibility, > and errors that could arise as the problem or system change (i.e., don't > imply that bugs need to be fixed in the code or run-time configuration). > Many of the latter can be configured to never raise an error. > Basically, I am trying to perform a test where I solve a linear system for several values of a parameter of the problem. For some of these values the system may be singular: I don't care (for now), just discard those values and go on to the next ones (without crashing the application). Francesco ------------------------------------------------------------------------------ Managing the Performance of Cloud-Based Applications Take advantage of what the Cloud has to offer - Avoid Common Pitfalls. Read the Whitepaper. http://pubads.g.doubleclick.net/gampad/clk?id=121051231&iu=/4140/ostg.clktrk _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
