operator bool() and set_zero(): test for zero and set to zero respectively.

[jeremy-murphy]

Checking for zero or non-zero and setting to zero are common operations, so fast implementations are useful.

[jzmaddock]

This should be explicit when BOOST_NO_CXX11_EXPLICIT_CONVERSION_OPERATORS is not set, otherwise use an implicit conversion to an "unmentionable" type, for example multiprecision::number uses:

typedef bool (self_type::*unmentionable_type)()const;

BOOST_MP_FORCEINLINE operator unmentionable_type()const
{
return is_zero() ? 0 : &self_type::is_zero;
}

Which prevents accidental conversion to an arithmetic type.

[jeremy-murphy]

Ahh, wow, that's a new one on me. What about naming a function that sets the polynomial to zero? I have used the verb clear but maybe set_zero is more appropriate? This is on the assumption that it is more efficient than assignment from zero, which I assume the compiler is not quite smart enough to optimize away for a class such as polynomial.

[jzmaddock]

On 12/05/2016 12:08, Jeremy W. Murphy wrote:

In include/boost/math/tools/polynomial.hpp
#39 (comment):

@@ -455,6 +454,18 @@ class polynomial :
normalize();
return *this;
}
+

• // Useful for fast test of equality to zero.
• operator bool() const

Ahh, wow, that's a new one on me. What about naming a function that
sets the polynomial to zero? I have used the verb |clear| but maybe
|set_zero| is more appropriate? This is on the assumption that it is
more efficient than assignment from zero, which I assume the compiler
is not quite smart enough to optimize away for a class such as polynomial.

clear() works for me.

[jeremy-murphy]

I still need to update the docs after we settle on the names.

[jeremy-murphy]

And what about the horrible workaround for gcc 4.6.x?

[jzmaddock]

On 12/05/2016 12:10, Jeremy W. Murphy wrote:

And what about the horrible workaround for gcc 4.6.x?

I don't believe that's required in this case - it was the presence of
other operator() overloads that caused the issue (it called the bool
operator rather than the correct one if memory serves).

[jeremy-murphy]

John, if you're happy with my choice of set_zero as the name for the modifying function, then this is all done from my end.

[jzmaddock]

Can you mark operator bool in the docs as explicit to match the code?

Re the name "set_zero", I confess I'm not overly keen on set_ and get_ prefixes, so I would be inclined to go with either just "zero" or even "clear". Does assigning from a literal zero produce the same result BTW?

Thanks for this, John.

[jeremy-murphy]

I'm definitely not in favour of the object-oriented habit of 'getters and setters' on a per-member basis, but I thought it was OK in this case. I thought "clear" was too tied to the underlying implementation and is not meaningful to a polynomial per se (although that's not a huge sin). "zero" seems ambiguous as to whether it is setting or querying? I don't love "set_zero" but the alternatives all seem more problematic.

[jzmaddock]

OK lets go with what we have and move on, merging to develop...