On Friday, September 19, 2003, at 07:13  AM, Christos Zoulas wrote:

> On Sep 19, 10:08am, perry@piermont.com ("Perry E. Metzger") wrote:
> -- Subject: Re: CVS commit: src/sys/arch/alpha
>
> If there are technical reasons for not putting PT_STEP in the
> kernel, another way would be to put it in the libc ptrace() stub, so
> that it is available for other applications.
>
> Jason, what do you think about that?

I think it's wrong.  I think that if breakpoint-based stepping is going 
to happen, then the debugger should KNOW that, and have explicit 
control over it.

For example, if the kernel writes over an instruction word, and the 
debugger doesn't know about it, you have no way to properly disassemble 
the function you're stepping through.  On the other hand, if the 
debugger writes over an instruction word, it has a record of it, and 
can compensate during disassembly.

         -- Jason R. Thorpe <thorpej@wasabisystems.com>