My rough interpretation (which is by no means the official one, as I think this is still up for debate), is as follows:,
1) convert(T,x) is the most specific: it will convert x to type T, throwing an error if outside the range (though it can round, for instance when converting to a Rational or FloatingPoint type). This can be called implicitly, e.g. when assigning to a typed array. 2) T(x) does mostly the same thing, but is slightly more general * when T <: FloatingPoint, you can pass extra rounding arguments, e.g. Float64(x,RoundDown) * in some cases, x can be a string: at the moment this seems to only work for BigFloat and BigInt, but perhaps this should apply to other types. 3) x % T is defined only for integer types: the idea is to be a "module conversion", and can be faster because no exceptions can be thrown. Additionally, if x is a FloatingPoint type, then there are also the extra methods round(T, x), trunc(T, x), floor(T, x), ceil(T, x) and the non-exported Base.unsafe_trunc(T, x). s On Saturday, 21 March 2015 00:22:43 UTC+1, Milan Bouchet-Valat wrote: > > Le vendredi 20 mars 2015 à 14:03 -0700, Julia User a écrit : > > I looked through some of the Base source code and see different things > > for conversions. > > > > c%UInt8 > > convert(UInt8,x) > > UInt8(t) > > > > to name a few.... > > > > Any official recommendation for someone starting out new - anything > > which is planned to be depreciate soon. > I think all of them are OK, but the last one does not have the same > behavior: > > julia> UInt8(256) > ERROR: InexactError() > in call at base.jl:36 > > julia> convert(UInt8, 256) > ERROR: InexactError() > in convert at int.jl:160 > > julia> 256 % UInt8 > 0x00 > > The first one ends up calling the second one, so I'd say use it since > it's shorter. The last one does not check for overflow, so use it only > if you really want this behavior. > > In short: in doubt, use the first one. > > > Regards >
