add True, False

[jverzani]

In #260 a promote rule for true or false is setup. Currently this would promote true to a symbolic 1 and false to a symbolic 0 as Bool <: Integer. This provides an alternatative by promoting to a symbolic True and symbolic False.

I don't know which is a better choice, this is just meant to provide an option.

If this option is better, it might be nice to have some means to convert True or False to 1 or 0 with the N function.

[jverzani]

One option for N could be

@@ -124,7 +125,12 @@ end
 function N(::Val{<:Any}, b::Basic)
     is_constant(b) ||
         throw(ArgumentError("Object can have no free symbols"))
-    out = evalf(b)
+
+    # replace any True/False with 1/0
+    m = CMapBasicBasic()
+    m[True] = Basic(1); m[False] = Basic(0)
+
+    out = evalf(subs(b, m))
     imag(out) == Basic(0.0) ? N(real(out)) : N(out)
 end

It adds some allocations, but isn't too costly performance wise.

[isuruf]

Add a specialized N(::Val{:BooleanAtom}, b::Basic) ?

[jverzani]

That is here, it is when a BooleanAtom is part of an expression (like Basic(2)+True) that N falls back to calling evalf which doesn't treat True as numeric, as happens in Julia.

[jverzani]

Oops, I hadn't pushed all the changes I had locally. What is there now is one way that booleans could be handled.

[isuruf]

Can we make this map a global variable instead of creating it every time?

[jverzani]

I need your help with that. I was trying, but kept getting segfaults. The construction of Basic objects is a bit different than that of CMapBasicBasic, so I was having issue initializing the map in __init__.

One thing about what I have here that bothers me is I'd like to cheaply test if an expression has a BooleanAtom and then try to evaluate that expression after substitution with evalf, but it seemed easier to do the substitution and then compare the result to the original. I wasn't sure if substituting in a map with no matches would always return the original expression, as the test has that assumption.

[jverzani]

I just checked in what I thought I'd tried earlier and it seems to not cause an issue. I'll add a direct test to be sure.