`replace` does not handle multiple patterns for String

[Keno]

julia> replace("Keno", 'e'=>"abc", 'f'=>'g', 'a'=>'z')
ERROR: MethodError: no method matching similar(::String, ::Type{Any})
Closest candidates are:
  similar(::Array{T,1}, ::Type) where T at array.jl:357
  similar(::Array{T,2}, ::Type) where T at array.jl:358
  similar(::Array, ::Type, ::Tuple{Vararg{Int64,N}}) where N at array.jl:360
  ...
Stacktrace:
 [1] _similar_or_copy(::String, ::Type{Any}) at ./set.jl:415
 [2] replace(::String, ::Pair{Char,String}, ::Vararg{Pair,N} where N; count::Nothing) at ./set.jl:530
 [3] replace(::String, ::Pair{Char,String}, ::Pair{Char,Char}, ::Pair{Char,Char}) at ./set.jl:527
 [4] top-level scope at REPL[12]:1

This is surprising, because it doesn't match the API of replace on generic collections:

julia> replace([1,2,3],1=>4,2=>5,3=>6)
3-element Array{Int64,1}:
 4
 5
 6

Now, replace on string is a bit special, because it can take Strings and Regexes also, as patterns, and perhaps we should only allow a single pattern for those cases (because otherwise the order of replacements is ambiguous), but I think it should work for Char, and at least give a better error message for multiple patterns of another type.

[sudo-rushil]

I thought of using a recursive solution: Unfortunately, I've run into a problem.

julia> function replace(str::String, old_new::Pair...)
              for pair in old_new
                  str = replace(str, pair)
              end
              str
          end

However, it encountered this error.

julia> replace("Keno", 'e'=>"abc", 'f'=>'g', 'a'=>'z')
"Kzbcno"

I would have assumed that the desired output would be "Kabcno", but I'm not sure. It raises the question; what should be the precedence of replacements? I'll keep thinking on it and get back to you.

[Keno]

For the Char replacement case, it should definitely apply the patterns only in one go.

[sudo-rushil]

Got it. How's this?

julia> function replace(str::String, old_news::Pair...)
           out::String, mapping::Dict{Char,Any} = "", Dict(old_news)
           for c in str
               if c in keys(mapping)
                   out *= mapping[c]
               else
                   out *= c
               end
           end
           out
       end
replace (generic function with 10 methods)

julia> replace("Keno", 'e'=>"abc", 'f'=>'g', 'a'=>'z')
"Kabcno"

[Keno]

It has the correct interface, probably, but a complete implementation needs to look a bit different, as that implementation has O(n^2) performance characteristics. There's also a question of how exactly it integrates with the rest of the generic API.

[sudo-rushil]

I could speed it up to linear time, if it made a set of the keys in advance, but would that be good enough to solve the integration problem?

[tkf]

FYI previous discussions:
#30457
#25396

[DNF2]

I don't understand the error message here, either:

julia> replace("hello", ('l'=>'n'), ('o'=>'r'))
ERROR: MethodError: no method matching replace(::String, ::Pair{Char,Char}, ::Pair{Char,Char})
Closest candidates are:
  replace(::AbstractString, ::Pair, ::Pair) at set.jl:583
  replace(::Any, ::Pair...; count) at set.jl:526
  replace(::Union{Function, Type}, ::Pair, ::Pair; count) at set.jl:582

Doesn't the signature match the first candidate?

[sudo-rushil]

It turns out, that first candidate only throws a method error on the arguments. From base/set.jl:584:

replace(a::AbstractString, b::Pair, c::Pair) = throw(MethodError(replace, (a, b, c)))

[DNF2]

Is that a good idea, or a common pattern? It's very confusing, and quite frustrating.

[rfourquet]

Is that a good idea, or a common pattern? It's very confusing, and quite frustrating.

It doesn't seem like a good idea indeed (at least in general). IIRC in this case this is to handle ambiguities which arise when this method is not defined. Maybe throwing an ArgumentError or another exception would be better (disclaimer: I believe to be the one to blame for this ugly throw(MethodError(...)) ;-) )

[sudo-rushil]

I might be wrong but doesn't the call to findfirst have linear complexity?

[francescoalemanno]

edit: removed old post with code snippet, turned it to a PR see @ #35414

@sudo-rushil yes, this implementation is not the fastest, but it guarantees correct semantics.