The second is usually better (in more detail about what you are doing). I usually prefer if (a_double == 0.0). Also note that floating-point approximation comparisons are often useful given the possibility of rounding in calculations (although doing this well may be nontrivial).
. , , , : . , , , . ( ) , 2, .
, ( ), (, IEEE , 1e + 10000 1e- 2000). , , (, double 1e-308 - 1e-300 7 ~ 15).
( - 2 53), , 2 (, 3,5, 1,375).
, / . , , , (.. 16 2, , 2). , 5 (, 1.2 , ). , ( , ) , , .