Re: [Libmesh-devel] [Libmesh-users] write_equation_systems() with several Systems and EquationSystems

[Derek G. <fri...@gm...>]

(switching conversation to devel)

On Thu, Dec 6, 2012 at 3:19 PM, Roy Stogner <roy...@ic...>wrote:

> If we just switched to devel as the default we'd merely replace "my
> program crashed" with "my program was way too slow", which may even be
> worse in that it's less likely to elicit an email allowing us to help
> fix the problem.
>

Please do not switch the default away from being opt. I prefer the default
for a high-performance numerical library to be the high-performance mode.
 At the point where I run into a problem I can choose to turn on debugging
stuff to send to the developers or figure out what's going on.

I've actually never liked that PETSc does it the other way around.  In all
of my years of using PETSc I have never sent a stack trace to the PETSc
team... nor have I ever needed a debug version of PETSc in general.  That's
not to say that no one does... it all depends on what you're doing.  We
always disable debugging in PETSc.

I could definitely see someone doing a comparison of libMesh to DealII or
something else and being like "see!  libMesh is slow" all because they
didn't notice they weren't running in "opt"....

Derek