round() with BigFloat argument

[mbaz]

julia> ceil(BigFloat(2.1))
error: stack overflow
in ceil at math.jl:119

(The last line is repeated many times)

round() also fails in a similar manner; I haven't tested other similar functions.

[JeffBezanson]

We have all these fallbacks of the form f(x::Real) = f(float(x)), which basically converts a no method error to a stack overflow in cases like BigFloat that don't yet implement ceil. I wonder if these should use ::Integer instead to help avoid this trap.

[nolta]

Partially fixed by 0c1a073, but round still has the same problem.

[andreasnoack]

@JeffBezanson I think Integer is better. A no method error makes more sense here.

@nolta Maybe round should return an error.

Regarding 0c1a073, the new functions use the symbols in gmp whereas the old use gmp_wrapper. The latter is just renaming of functions in gmp and some memory allocation. Wouldn't it be more clean to remove all the renaming from gmp_wrapper.c and just have the memory allocation there? The rest can gmp directly, I think.

[ViralBShah]

I wonder if we can get rid of gmp_wrapper.c completely. One can always allocate the memory as an array of Uint8s and pass it in. I do that in the wrappers for DSFMT.

[aviks]

#1887 : previous discussion around removing gmp_wrapper.c

[JeffBezanson]

Fixed by f34c812.