Remove Int64 type definition from Problem

[ovainola]

Problem function defined dimension as an Int64 type, and this
restricts using JuliaFEM on 32bit architecture. Changed type to
a more generic: Integer.

fixes issue #113

[coveralls]

Coverage decreased (-3.03%) to 87.172% when pulling b520af3 on ovainola:master into aff194b on JuliaFEM:master.

[ahojukka5]

We are also having problems with Float64 if we want to have JuliaFEM to work on 32 bit machines.

julia> f(x::Float32) = 2x
f (generic function with 2 methods)

julia> f(2.0)
ERROR: MethodError: no method matching f(::Float64)
Closest candidates are:
  f(::Float32) at REPL[3]:1
  f(::Int32) at REPL[1]:1

In general, should we use AbstractFloat instead of Float64 and Integer instead of Int64 to have also 32-bit compability?

[ovainola]

it'd be more general if we would change.

[ChrisRackauckas]

No. Float64 is the default float on 32-bit machines and 64-bit machines. The difference is only in integer sizes. However, instead of using the abstract Integer, you should probably be using Int. This is an alias, and on 32-bit Int == Int32 and on 64-bit Int == Int64. So then typeof(1) == Int is true on either machine. You should then, in the few cases where you expect overflow (the integer is greater than 2147483647) upconvert inside of your functions to Int64 manually via Int64(a).

[ChrisRackauckas]

Oh, I misread this at first. Yes, you won't have performance issues if you loosen function definitions, so it's usually recommended you go to Integer and AbstractFloat, or whatever is the highest abstract type which makes sense. For storing the values in types, you should make them concrete, so either parameterize or use the concrete types Float64 and Int as explained above.

[ChrisRackauckas]

you can make this dimension::Integer to make it more succinct.

[ahojukka5]

This particular file has been fixed in FEMBase, however, there's still use of Int64 in code, which should be replaced with Int. In order to make progress, we should activate Appvoyer to test in 32-bit environment to catch all problems with type definitions. See this PR (JuliaFEM/FEMBase.jl#46).