I encountered some unexpected behavior regarding broadcasted functions for strings. Let’s say we have two strings: a = "abc" b = "abx" When checking for equality, the result was as I expected: a == b # result: false When I wanted to check for the equality for each individual glyph using a broad…

[image] antoine-levitt: For instance, Strings are “scalar” and structs are “containers” by default, even though sometimes you want structs to be scalar Every new type is assumed to be a “container” by default. If you want to opt out of this for a new type that you define, you define Base.bro…

I didn’t found a reasons, except that String is considered to be a scalar type. See e.g. How to check if a variable is scalar in julia - Stack Overflow or https://syl1.gitbook.io/julia-language-a-concise-tutorial/language-core/data-types#scalar-types But by the definition of isscalar using Broadc…

It seems to me that this would make working with vectors of strings painful - mainly in DataFrames, where subsetting data by doing something like df[df.col1 .== "some string", :] would presumably error if the string is considered an iterator.

Should strings be vectors of chars? a = "abcd" b = "abxd" Vector{Char}(a) .== Vector{Char}(b) 4-element BitVector: 1 1 0 1

julia> @code_lowered Vector{Char}(a) CodeInfo( 1 ─ %1 = Base.collect($(Expr(:static_parameter, 1)), s) └── return %1 ) identical to collect(Char,a).

There is always this tension of what types are considered “scalar” vs “containers” for the purposes of broadcasting. For instance, Strings are “scalar” and structs are “containers” by default, even though sometimes you want structs to be scalar - in which case you use Ref - and sometimes you want st…

Thanks @oheil for the excellent performance tip! I actually thought collect would be slower because it didn’t return a view. I obviously should have measured instead of guessed. Could you explain why the SubString view (it is a view, right?) makes it slower? I was expecting == to just perform read …

[image] fedario: Could you explain why the SubString view (it is a view, right?) makes it slower? The SubString isn’t that what makes it slower. But split fills up a Vector with chunks of SubString-views by traversing through the complete String searching for the next occurence of the split …

What I find jarring is the fact that strings are iterable but scalar-like for broadcast, and structs by default are not iterable but container-like for broadcast. I don’t use strings much so have no opinion on them, but I’ve often wished broadcast on structs was scalar by default. It seems to me th…

Scratching my head to understand this: [image] stevengj: "Raphaël" and "Raphaël" are canonically equivalent strings, but they don’t contain the same Char s (or even the same number of Char s). Is there a typo? Thank you for shedding some more light. Later edit: Your other post here helps…

I had the same question marks in head, but copy&paste did it: julia> "Raphaël" "Raphaël" julia> "Raphaël" "Raphaël" julia> collect("Raphaël") 8-element Vector{Char}: 'R': ASCII/Unicode U+0052 (category Lu: Letter, uppercase) 'a': ASCII/Unicode U+0061 (category Ll: Letter, lowercase) 'p': AS…

Dot calls / broadcasts for strings

Internals & Design

Topic		Replies	Views
Basic string iteration New to Julia	6	3995	June 17, 2021
`Char` and `broadcastable` Internals & Design	8	756	May 19, 2021
recent broadcast changes (iterate by default), scalar struct, and `@.` Internals & Design broadcast	68	8134	January 9, 2019
String indices : byte indexing feels wrong New to Julia strings , unicode	18	1600	December 5, 2023
Broadcasting on Vector{SubString{String}} does not work in Julia 1.7.0 rc1? General Usage strings , broadcasting	12	1997	October 16, 2021

Dot calls / broadcasts for strings

Related topics