[patch] debug mode

[François Dumont]

Hi

     Thanks for applying the patch.

     - I will check the exception/ testcases to see how it is done. 
Moreover I also discovered the dg-require-debug-mode Deja GNU extension. 
I might use it rather than making the test failed with a 
__builtin_assert without debug mode.
     - Ok for potential performance issue of mutexes in debug mode, I 
will check it later. For the moment I simply would like to generalize 
the patch that I have done to containers to algorithms. This will be the 
subject of a future mail this week I hope.
     - I also notice that there were no forward_list debug-mode and even 
started to work on it. But I prefer to concentrate on one subject at a time.

François

On 01/-10/-28163 08:59 PM, Paolo Carlini wrote:
> Thanks, the below is what I actually committed, only minor formatting /
> stylistic changes, after testing on x86_64-linux multilib, in various modes.
>
> A few additional comments:
> - I suspect that the testing infrastructure could be made more
> consistent with the one used in the exception/ testcases, in particular,
> I don't like so much the repeated cont_traits.h and debug_cont_traits.h
> headers...
> - If you are interested in performance, I would recommend again
> benchmarking and analyzing in detail our uses of mutexes in debug-mode,
> which we quickly added on top of the original design to make sure the
> behavior wrt mt programs is indistinguishable from that of the normal
> containers, besides performance of course. I'm pretty sure there are big
> margins of improvement there...
> - Well, if you care about debug-mode more generally, some help on adding
> a proper debug-mode forward_list would be appreciated, is in my todo
> list, but I didn't get around to work on it so far. The task isn't
> *completely* trivial - will require extending a bit the generic
> debug-mode infrastructure - because of the before_begin concept, which I
> believe could be dealt with similarly to end in the other containers.
> Let me know in case...
>
> Thanks again,
> Paolo.
>
> ////////////////////////
>