specialize length(SubString) for DirectIndexString

[catawbasam]

passes 'make clean testall'

Double-checked code_typed( length,(SubString{ASCIIString},) ) to verify that this one actually affects execution :/

Current code_typed() output:

1-element Array{Any,1}:
 :($(Expr(:lambda, {:s}, {{:i,:n,:#s406,:#s404,:c,:#s403,:j,:#s405,:_var0,:_var1,:_var2,:_var3},{{:s,SubString{ASCIIString},0},{:i,Int64,2},{:n,Int64,2},{:#s406,(Char,Int64),18},{:#s404,(Char,Int64),18},{:c,Char,18},{:#s403,(Int64,Int64),18},{:j,Int64,18},{:#s405,Int64,2},{:_var0,Char,18},{:_var1,Int64,18},{:_var2,Int64,18},{:_var3,Int64,18}},{}}, :(begin  # string.jl, line 81:
        i = 1 # line 82:
        unless slt_int(top(getfield)(s::SubString{ASCIIString},:endof)::Int64,i::Int64)::Bool goto 0 # line 83:
        return 0
        0:  # line 85:
        n = 1 # line 86:
        unless true goto 2
        3:  # line 87:
        #s406 = next(s::SubString{ASCIIString},i::Int64)::(Char,Int64)
        #s405 = 1
        _var0 = tupleref(#s406::(Char,Int64),1)::Union(Int64,Char)
        _var1 = box(Int64,add_int(1,1))::Int64
        c = _var0::Char
        #s405 = _var1::Int64
        _var2 = tupleref(#s406::(Char,Int64),2)::Union(Int64,Char)
        _var3 = box(Int64,add_int(2,1))::Int64
        j = _var2::Int64
        #s405 = _var3::Int64 # line 88:
        unless slt_int(top(getfield)(s::SubString{ASCIIString},:endof)::Int64,j::Int64)::Bool goto 5 # line 89:
        return n::Int64
        5:  # line 91:
        n = box(Int64,add_int(n::Int64,1))::Int64 # line 92:
        i = j::Int64
        4: 
        unless false goto 3
        2: 
        1: 
        return
    end::Int64))))

Proposed code_typed() output:

1-element Array{Any,1}:
 :($(Expr(:lambda, {:s}, {{},{{:s,SubString{ASCIIString},0}},{}}, :(begin  # In[65], line 1:
        return top(getfield)(s::SubString{ASCIIString},:endof)::Int64
    end::Int64))))

[JeffBezanson]

Nice!

[StefanKarpinski]

Yes, this is clearly a big improvement. I'm trying to think if situations where this isn't true, but I can't think of any. I think the fact that the index of the i character is i implies that endof(s) == length(s).

[catawbasam]

Below are a few tests. I'd be happy to add them to test/strings.jl -- but have no experience piling changes onto an open pull request. Is that a safe thing to do?

julia> using Base.Test

julia> str="tempus fugit"
"tempus fugit"

julia> ss=SubString(str,1,length(str))
"tempus fugit"
julia> @test length(ss)==length(str)

julia> ss=SubString(str,1,0)
""
julia> @test length(ss)==0

julia> ss=SubString(str,14,20)
""
julia> @test length(ss)==0

julia> ss=SubString(str,10,16)
"git"
julia> @test length(ss)==3

julia> str2=""
""

julia> ss=SubString(str2,1,4)
""
julia> @test length(ss)==0

julia> ss=SubString(str2,1,1)
""
julia> @test length(ss)==0

[JeffBezanson]

Yes, you can keep pushing to the same branch and the new commits will appear as part of this pull request.

[catawbasam]

OK. tests pushed.