[boost] indeterminate tribool to bool

From: Fernando Cacciola (fernando_cacciola_at_[hidden])
Date: 2005-07-28 18:12:42

Hi,

I'm reading a draft of a proposal to standarize interval arithmetic (from 
the authors of Boost.Inverval).
They propose to include the specialization "interval<bool>" to represent 
"false<->maybe<->true", that is, a tristate boolean expressed as an interval 
in the "lattice" range [false,true]

They specifically mention that they propose interval<bool> instead of 
Boost.TriBool becasue they argue that conversion from indeterminate to bool 
shouldn't be false but fail with an exception.
It turns out that in any actual case I can think of, throwing when trying to 
narrow an indeterminate tribool into true or false is indeed much prefered 
than returning false.

The tribool documentation says that you should write:

if ( b )
  certainly_true();
else if (!b )
   certainly_false();
else uncertain();

The authors of the proposal argue that even though you should follow that 
pattern, if you don't, so your code
is written like:

if ( b )
     certainly_true();
else certainly_false();

then if b is indeterminate the false branch should not execute because 
that's hardly what the programer intended in the case of a tri-state logic.

>From my own experience I cannot but agree. If you're implementing a complex 
triboolean predicate that nests child triboolean predicates (various levels 
down), you can rightly decide to use the simplified pattern (with just the 2 
certain branches) and let the exception mechanisms wind up the "uncertain 
answer" up to the caller; but for this you need ideterminate->bool to 
fail...
Of course there is always the choice of using the "correct" pattern that 
properly test for "certainly false".

So, why doesn't tribool throw in this case?

Fernando Cacciola
SciSoft