IOBuffer of SubString

[hayd]

It seems like this might be useful to define, for example if you don't know whether the type is a string or a substring.

I had a go implemented with the following in this commit:

IOBuffer(s::SubString) = IOBuffer(s.string.data[s.offset:s.endof], true, false)

however, that seems not to work. :(

[ivarne]

How doesn't it work?

[hayd]

@ivarne great question, I will try and come up with a simple example (was hoping it was somethng obvious). I think it's unicode related.

The bug was with unicode in markdown tables with the above definition (from the commit), without unicode it works (IIRC).

[ivarne]

Looking at how the .endof and .offset fields are used leads me to think that

IOBuffer(s::SubString) = IOBuffer(s.string.data[(s.offset:(s.offset+s.endof)) + 1], true, false)

will work much better.

Edit: Changed - 1 to + 1, sorry if you look at the email!

[stevengj]

That seems suboptimal because it is creating a copy of the data (at least until #9150 lands). And if you're willing to make a copy, you could just do IOBuffer(s::AbstractString) = IOBuffer(utf8(s)).

Doing this without making a copy seems problematic right now because IOBuffer requires an Array{Uint8}, and doesn't permit subarrays.

[ivarne]

I just thought IOBuffer needed a real Array, but if it can work directly with a SubArray, it might be worth it.

[stevengj]

It looks like the code would work with any AbstractVector supporting pointer and stride where the stride is 1 (i.e. contiguous). This should include contiguous subarrays, although I'm not sure what methods are defined for SubArray as I haven't used that type much.

[vtjnash]

i think we can make IOBuffer parameterized by the type of the AbstractVector{Uint8} that it contains, rather than forcing Vector{Uint8}. that would also allow it to wrap a SharedArray too.

[JeffBezanson]

Doesn't IOBuffer kind of need to own the vector it wraps? I guess a read-only or non-growable IOBuffer on a SubArray would be ok though.

[stevengj]

@JeffBezanson, in the IOBuffer(::String) case the buffer is read-only.

[vtjnash]

right. when creating a IOBuffer from an existing object, it should typically default to read-only

[vtjnash]

this was implemented in #11554