Fix #29451: parse(Int, s::AbstractString) when ncodeunits(s) > lastindex(s)

[samoconnor]

#29451 :

i <= ncodeunits(s) should be i <= endpos ?

julia/base/parse.jl

Lines 75 to 76 in 1a1d6b6

	if c == '0' && i <= ncodeunits(s)
	c, i = iterate(s,i)::Tuple{Char, Int}

When I do parse(Int, "0") where "0" is an AbstractString and ncodeunits > lastindex, i get this:
TypeError: in parseint_preamble, in typeassert, expected Tuple{Char,Int64}, got Nothing

Bug introduced here ? 1a1d6b6#r30713661

Needs a sanity check from someone who knows the detail of the parser code @Keno? @StefanKarpinski?

[samoconnor]

@KristofferC, do you think the test needs to include a working struct TestAbstractString <: AbstractString and call parse(Int, :: TestAbstractString) ? Or would something like this suffice as a minimal failing example? ...

julia> s = SubString("0x")
"0x"

julia> Base.lastindex(s::SubString) = s.string == "0x" ? 1 : thisind(s, s.ncodeunits)

julia> function Base.iterate(s::SubString, i::Integer=firstindex(s))
           i == lastindex(s)+1 && return nothing
#     Was: i == ncodeunits(s)+1 && return nothing

           @boundscheck checkbounds(s, i)
           y = iterate(s.string, s.offset + i)
           y === nothing && return nothing
           c, i = y
           return c, i - s.offset
       end

julia> s
"0"

julia> parse(Int, s)
ERROR: TypeError: in parseint_preamble, in typeassert, expected Tuple{Char,Int64}, got Nothing
Stacktrace:
 [1] parseint_preamble(::Bool, ::Int64, ::SubString{String}, ::Int64, ::Int64) at ./parse.jl:77

[KristofferC]

There is an existing tstStringType and also a GenericString in the string tests, would it be possible to modify those to trigger the bug here?

[samoconnor]

How about this:

struct Issue29451String <: AbstractString end

Base.ncodeunits(::Issue29451String) = 12345
Base.isvalid(::Issue29451String, i::Integer) = i == 1
Base.lastindex(::Issue29451String) = 1
Base.iterate(::Issue29451String, i::Integer=1) = i == 1 ? ('0', 2) : nothing

@test Issue29451String() == "0"
@test Issue29451String()[1] == '0'
@test isvalid(Issue29451String(), 1)
@test !isvalid(Issue29451String(), 2)
@test parse(Int, Issue29451String()) == 0

Or even

struct Issue29451String <: AbstractString end

Base.ncodeunits(::Issue29451String) = 12345
Base.lastindex(::Issue29451String) = 1
Base.iterate(::Issue29451String, i::Integer=1) = i == 1 ? ('0', 2) : nothing

@test Issue29451String() == "0"
@test parse(Int, Issue29451String()) == 0

[StefanKarpinski]

Nice test case! Looks good to me once it passes tests.

[samoconnor]

CI looks good except for something about rmtmpdir and Pkg/pkg on Windows. Unrelated?