Re: [Libmesh-devel] Mixing debug/optimized objects

[Roy S. <roy...@ic...>]

On Wed, 16 Mar 2011, John Peterson wrote:

> .) class Elem has a couple of functions:    void libmesh_assert_valid_neighbors() const;
>   void libmesh_assert_valid_node_pointers() const;
> only defined in debug mode.  This seems to be a similar situation to what we had in Mesh so we should probably fix it.

Not quite as bad - the MeshBase functions were virtual and so led to
screwed up vtables and horrid runtime errors even when they weren't
used.  The Elem functions should only lead to link-time errors and in
far fewer cases.

Still worth fixing, though.

> .) The MeshTools namespace: A number of functions in the namespace are only defined in DEBUG mode, but because it's a namespace I don't *think* it
> would have the same problem with mixed opt/dbg binaries but I'm not sure.  These are all void functions so it would be pretty easy to give them
> empty bodies in DEBUG mode.

In non-DEBUG mode, you mean?  These probably aren't a problem, but for
now I'd go ahead and define them everywhere; we'll just use them only
in DEBUG mode.

> .) class InfFE has 2 static bool variables that are only defined in DEBUG mode.  They appear to only be used in src/fe/inf_fe_static.C.  It should
> be safe to remove the #ifdefs from the header file... In optimized mode a grand total of two extra unused bools will be stored.

This would definitely lead to horrid runtime errors; let's take out
the ifdefs.

> Let me know what you think and I will make these changes.

Sounds good.  I still like the idea of being able to strip
debugging-only functions out of libMesh, but it'll take more work to
do it properly - we'll need an extra configure-time option, I think.
---
Roy