Le samedi 29 novembre 2014 à 12:46 -0800, Christoph Ortner a écrit : > > > When I want to just write a quick script to test and idea, then I > don't want to check that I've created my indices, array lengths etc as > the correct type. I think for functions such as linspace not > performing an automatic type conversion is a big mistake. > > > I thought that in previous Julia verions `int32(1.1)` would throw > an error (maybe I misremember this). Now it just converts it to 1. I > then agree that this is a bit dangerous. It should throw an error > because 1.1 cannot be converted to int. This would resolve the issue > --- but of course there may be others that I don't appreciate. The future of int32(), int() and the like is not completely clear at the moment, especially with regard to the fact that they currently coerce, i.e. don't fail when exact conversion is not possible. See https://github.com/JuliaLang/julia/issues/1470
Regards > > Christoph > > > > > On Saturday, 29 November 2014 13:22:05 UTC, Tamas Papp wrote: > In general, I am more concerned about the following scenario: > type A > makes sense, type B in general, does not, yet a subset of type > B values > can be converted to A in a reasonable way, so an the > implementation > performs this silently and everything seems fine. > > But at the same time, had I known that I left a type B value > in there, I > would be concerned, eg because I am assuming that it is exact > when it > isn't (integers vs floats) is problematic, or that conversion > is > possible only for a subset of the domain, and other values are > possible > but rare enough for my unit tests not to discover them, and > will strike > me on the nose at some future point like a rake carelessly > left in the > grass. > > Best, > > Tamas > > On Sat, Nov 29 2014, Stefan Karpinski <[email protected]> > wrote: > > > Our conversions throw errors when the value cannot be > converted exactly these days so it could be done pretty > safely. But yes, this is just because there's no applicable > method of linspace. > > > > > >> On Nov 29, 2014, at 7:07 AM, Tamas Papp <[email protected]> > wrote: > >> > >> I think of this as a feature: using a float when only an > integer makes > >> sense is most likely a bug in my code. Automatic conversion > would mask > >> the problem. > >> > >> Best, > >> > >> Tamas > >> > >>> On Sat, Nov 29 2014, [email protected] wrote: > >>> > >>> One of the primary paradigms of Julia is the use of > multiple dispatch, that > >>> is choosing which method to call based on the types of the > parameters > >>> provided. To quote from the manual > >>> > http://docs.julialang.org/en/latest/manual/methods/#defining-methods > "No > >>> automatic casting or conversion of function arguments is > ever performed: > >>> all conversion in Julia is non-magical and completely > explicit." > >>> > >>> Cheers > >>> Lex > >>> > >>>> On Saturday, November 29, 2014 8:17:26 PM UTC+10, > Christoph Ortner wrote: > >>>> > >>>> > >>>> In Julia 3.3 `linspace(0, 1, 10) ` works, but > `linspace(0, 1, 10.)` > >>>> does not. > >>>> > >>>> Why is the type not automatically converted to Int when > possible? Is this > >>>> intended or a bug? > >>>> > >>>> Christoph > >>>> > >>>> > >>>> > >>>> > >>>> > >>>> > >>>> > >>>>
