Re: [patch] debug mode

[François Dumont <francois dot cppdevs at free dot fr>]

On 08/18/2010 03:17 PM, Paolo Carlini wrote:
> On 08/18/2010 03:14 PM, Paolo Carlini wrote:
>    
> > OOps, I misread your patch, you are by and large already doing this. But
> > we want to check debug-mode performance in this case, right? Thus, *in
> > these specific cases*, I think the tests *should* #define _GLIBCXX_DEBUG.
> >
> >      
> Or *both* normal mode and debug-mode, useful for comparison: You can do
> it in various ways, either duplicating the tests, one of each pair
> defining _GLIBCXX_DEBUG and prefixed or suffixed debug in the name, or
> in more sophisticated ways... up to you.
>
> Paolo.
>    
Ok, I see what you mean even if I don't think the normal mode is useful. 
As far as I understand performance tests are here to check that 
modifications in the library do not impact performances. So I guess that 
library maintainers have scripts that run regularly to check that from 
one run to the other performances are always the same or better. I don't 
think those scripts will compare normal with debug mode tests. So I will 
force _GLIBCXX_DEBUG for the purpose of this patch.

I might propose in a future patch the vector_construct and 
deque_construct tests in normal mode too because it might be useful to 
check that memmove is used internally, I need to check this first.

Bests