Re: [jamboost] The right configuration mechanism

From: Andrey Melnikov (melnikov_at_[hidden])
Date: 2005-08-16 07:43:28

Vladimir Prus wrote:
> On Saturday 13 August 2005 03:43, Andrey Melnikov wrote:
> 
> 
>>>And note that this decision should be the same for all toolsets.
>>
>>configure msvc : : foo/bar ;
>>
>>AFAIK now foo/bar has different meaning for different toolsets. 
> 
> 
> No, it means "compiler binary, possibly with path" for all C++ compilers.

Do you mean just to avoid "possibly"?

>>So 
>>requiring foo/bar to be interpreted as path for all toolsets breaks
>>backward compatibility.
>>
>>Users don't write configure directives often, and use these parameters
>>even more rarely. If backward compatibility is broken, I think we can
>>require an explicit parameter name here, prohibiting just foo/bar.
> 
> 
> I think requiring explicit parameter name is ultimately best approach --
> configuration directives indeed are rarely written, so user will consult the
> docs anyway.
> 
> 
>>>>>I'm not sure if ordinary users want to have all installed toolsets
>>>>>configured. Opinions?
>>>>
>>>>How could it hurt to have them configured? It costs nothing AFAICT!
>>>
>>>Except execution time. IIRC, SCons used to search for all tools it knows
>>>about and it was a performance problem. OTOH, we'll be searching just for
>>>a few compilers, and not for bison/lex/f77 and whatnot.
>>
>>AFAIK, now if I have 10 using rules in user-config, 10 toolset files are
>>parsed and 10 init rules are executed even if I build only with a single
>>toolset. Is this true?
> 
> 
> Yes. 

I wonder how much time for incremental builds we will save if we won't 
use full evaluation? From my measurements it seems that we won't save 
too much - an incremental build of a project consisting of 3 targets and 
5 cpp files takes less than 3 seconds on my old 900mhz box. But I only 
have 3 toolsets configured: MSVC 7.1, 8.0 and BCB.

Andrey