Re: [Boost-build] warnings (was: Problem with disabling warnings ongcc)

From: Reece Dunn (msclrhd_at_[hidden])
Date: 2006-03-06 03:44:39

Vladimir Prus wrote:
>On Friday 03 March 2006 15:57, Reece Dunn wrote:
>
> > What if we have a <warnings>default option or equivalent that translated
> > to: <toolset>gcc/<warnings>default ==> <warnings>all
> >    <toolset>msvc/<warnings>default ==> <warnings>on
> >    <toolset>cw/<warnings>default ==> <warnings>on
> >
> > This would retain the meaning of off, on and all so the user can specify
> > these and provides a means of getting a reasonable level of warnings for
> > compilers where <warnings>all is a bit too much. <warnings>default could
> > then be tailored to each compiler so some of the more verbose warnings
> > could be disabled (e.g. msvc's unreferenced argument) without impacting 
>the
> > meaning of on or off.
>
>This is technically possibly, but that would mean that <warnings>all will 
>be
>still unsuable with msvc, because it will produce a large number of 
>spurious
>warnings.

The idea for <warnings>all is to enable *all* warnings. It is possible to 
use msvc with this option - it just means wrapping <vector>, <map>, Boost 
headers and other additional library headers within #pragma disable blocks. 
This makes it possible to use the <warnings>all <warnings-as-errors>on 
combination.

> > However, I am open to suggestions.
>
>I think we should <warnings>all add /W3 on msvc, and simulary lower the
>warning level for CW.

Then how will developers enable all warnings on these toolsets? The 
<warnings>on option already does this for msvc and cw. I would like ot keep 
the meaning of the 3 warning levels:
   off -- disable warnings
   on -- enable warnings
   all -- enable all warnings, no matter how spurious

Your proposal would change the meaning of that and would mean that msvc and 
cw users would have to use (non-portable) compiler flags to provide all 
warnings, making BBv2 harder to use and more confusing.

- Reece