isXXX methods and integer_valued

[simonbyrne]

I guess this can get filed under bikeshedding, but should integer_valued (and real_valued) have an is prefix (like isfinite)? There is already isinteger (testing for an integer type), so should it be isintegervalued?

On a related note, the naming of the isXXX methods are somewhat ambiguous: some are determined by values (e.g. isfinite, isempty, islower, iseven, issym), and some are determined by type (e.g. isinteger, isreal, isbool): perhaps the type ones would be clearer as isXXXtype?

[johnmyleswhite]

I like all of these ideas, although it's a pretty substantial change.

[JeffBezanson]

Revisiting this, it now seems reasonable to only have isinteger and isa(x,Integer).

[pao]

To clarify--isinteger would be the value test, and isa(x, Integer) would be the type test, right?

[JeffBezanson]

Yes.

[StefanKarpinski]

+1

[simonbyrne]

I should point out that the current isinteger(x) is slightly different than isa(x,Integer) in that it also applies to AbstractArrays, e.g.

isinteger(1:10) == true

This usage matches that of Matlab and Octave. Arguably, such a function is not very "Julian" anyway, in that if you want different behaviour for an integer array, this should be defined at dispatch, and not within a function.

[simonbyrne]

I just did a quick scan through base:

• isinteger isn't used at all
• isreal is used twice in linalg/dense.jl (once on arrays, once on numbers)
• iscomplex used quite a bit in linalg, particularly for interfaces to Lapack and factorizations (most of these seem to operate on arrays)
• isbool is used once in sparse.jl (i'm not sure on this usage, but I assume it is to operate on individual elements).