Overload ncodeunits on 0.7 #114
Conversation
|
Probably worth retagging 0.3.3 after merging this. |
|
The string stuff in this package is still woefully broken after JuliaLang/julia#24999. |
|
The stack overflow happens when filling the array to return: using Unicode
pool = CategoricalPool(["", "caf"])
v2 = CategoricalArrays.catvalue(2, pool)
g = graphemes(v2)
x = Base._similar_for(1:1, eltype(g), g, Base.iteratorsize(g))
x[1] = first(g)So the problem happens when trying to convert a convert(SubString{CategoricalString{UInt32}}, first(g))That's really weird since we're supposed to have no-op |
test/08_string.jl
Outdated
| @test_throws BoundsError ind2chr(v1, 0) | ||
| @test ind2chr(v2, 4) === 4 | ||
| if VERSION < v"0.7.0-DEV.2949" | ||
| @test_throws BoundsError ind2chr(v1, 0) |
There was a problem hiding this comment.
Shouldn't this be converted to length calls?
There was a problem hiding this comment.
Is there Compat for multi-arg length?
test/08_string.jl
Outdated
| @test collect(graphemes(v2)) == collect(graphemes("café")) | ||
| if VERSION < v"0.7.0-DEV.2949" | ||
| # FIXME: This causes a StackOverflowError after the string overhaul | ||
| @test collect(graphemes(v2)) == collect(graphemes("café")) |
There was a problem hiding this comment.
I'd rather keep this or mark it as @test_broken, else we'll forget about it.
src/value.jl
Outdated
| Base.nextind(x::CategoricalString, i::Integer) = nextind(get(x), i) | ||
| Base.prevind(x::CategoricalString, i::Integer) = prevind(get(x), i) | ||
| Base.nextind(x::CategoricalString, i::Integer) = nextind(x, Int(i)) | ||
| Base.nextind(x::CategoricalString, i::Int) = nextind(get(x), i) |
There was a problem hiding this comment.
No need to define a method for ::Int, the conversion used in the ::Integer method will be a no-op anyway.
|
If you know what needs to change here, do feel free to push directly to this branch, @nalimilan. Otherwise I'll see if I can get to this a little later. |
nalimilan
force-pushed
the
aa/strings
branch
from
209f11f to
f9da21b
Compare
|
I've added more changes to fix all errors and deprecations in 08_string.jl. I've stopped using "é" in the tests since it triggers failures in Base which should be handled separately. |
No description provided.