>>There is a performance increase using a named constant. It's not evaluated at run-time. See message#268676. George Tasker came up with a test that shows this.
>
>
>Yes, this is kind of related to that thread about defining SYS() functions and other similar things, something that I would usually use a local for if only needed in one method, but now I'm wondering if a constant might be better. If it weren't for that debugger problem...

I know, dejavu. If you have 'Code complete' by Steve McConnell he spends about 5-6 pages discussing named constants and how to use them.

I think if used properly the debugger issue is not much of a problem. One of our goals should be 'information hiding' ie. Hide all those (to us) meaningless numbers behind something we understand. The messagebox constants are a perfect example.

In 'Code complete' Mr. McConnell writes that we should avoid all literals (with the exception of 0 and 1). They should be replaced with named contants.