Standard scheme for maintainer-only debugging code?

[Joseph S. Myers]

On Tue, 10 Jun 2003, Nathanael Nerode wrote:

> Speed.  Sometimes (quite often actually) such code is in deeply inner 
> loops.  For one reason or another, one instance of such code might 
> trigger hundreds of times per compilation.  Combine that with hundreds 
> of such instances, and you *want* this out of release versions of the 
> compiler.

What is the actual performance cost of this (with/without
__builtin_expect)?  In particular is it reasonable to have this code all
available (controlled by command-line options) in default enable-checking
builds (as on mainline)?

-- 
Joseph S. Myers
jsm28@cam.ac.uk