I have two types (FooType and BarType) which implement the AbstractArray interface. The former is just a vector of float numbers, the latter is a vector of FooType. When I try to call sort() on them, it works smoothly for FooType objects but doesn’t for BarType. The error is: ERROR: MethodError:…

What happens is that this fails: julia> similar(b, Float64, (3,)) ERROR: MethodError: Cannot `convert` an object of type Int64 to an object of type BarType This may have arisen from a call to the constructor BarType(...), since type constructors fall back to convert methods. Base.similar(b::BarTy…

it appears that you have yet to define Base.isless for the second type

But it makes no sense logically (imo), since the comparisons are between elements of BarType (FooType) and not BarType themselves! To make a concrete example, when you sort an array of integer numbers, you need to define the comparison operator (isless) for integer numbers, not arrays of integers. …

Thanks for the tip, I’ve reformatted the text, hope it’s more readable now. Just in case, I tried to define Base.isless for FooType too as suggested by @JeffreySarnoff , and it doesn’t work indeed. Base.isless(a::BarType, b::BarType) = a[1]<b[1] That is the code I used, I’m still convinced that …

I’m a bit surprised sort(b.data) works actually, if i understand correctly it is sorting a vector whos elements are also vectors? Base.isless(a::BarType, b::BarType) = sum(a.data[1]) < sum(b.data[1]) would be one kind-of-sensible way of sorting them (sum of first vector of BarType), but you may wa…

(its not the way you are looking at it) a < b is an element-centric operation (is the element a less than the element b? where you get to define the meaning of “less than” should a, b be elements of your own structured type. [a1, a2, a3] < [b1, b2, b3] is not directly obainable in general given th…

It might be confusing the sorting algorithm, foo(A) != foo(B) where !(foo(A) < foo(B)) & !(foo(B) < foo(A)) Base.:(==)(a::FooType, b::FooType) = a.data == b.data Base.isless(a::FooType, b::FooType) = a[1]<b[1]

Let my clarify again: Base.isless(a::FooType, b::FooType) = a[1]<b[1] may seem a nonsense way to compare two vectors, but it doesn’t matter here, the problem is different. The issue is not how elements are ordered, but the fact that I get an error when trying to sort a BarType (which is an array of …

I’ve another example which is emblematic: x = [FooType(2), FooType(3)] sort(x) #-> works This is the proof that the problem isn’t comparing objects of FooType, but is something else. The question remains open

I haven’t played with Offset Arrays at all, but from the description they probably have their own way of doing getindex. You may need to pass in a set of index values? There is a possible shortcut in that Base.getindex(b::BarType, i::Int) = b.data[i].data[1] should be enough to run sort(b::BarTyp…

Sorting custom types

New to Julia

tkoolen March 22, 2018, 11:08pm 4

By the way, please see: PSA: how to quote code with backticks.

Topic		Replies	Views
Sorting arrays of structures New to Julia sort	13	7183	September 17, 2019
Special purpose subtypes (of arrays) New to Julia array , type	4	565	January 31, 2019
Structure type + in place modification of columns of arrays New to Julia type , mutable-structure	4	574	July 8, 2021
Meaning of `isless`, `<`, `cmp`, `isequal`, `==`, total order, partial order, unordered Internals & Design	24	2450	January 7, 2018
Help me implement a sorting method New to Julia sort	6	724	April 10, 2019

Sorting custom types

Related topics