Buffer issue on nightly 1.12.0-DEV.148 (2024-03-08) about julia HOT 8 CLOSED

I guess the question is "what do you want to do with bufdata?". In julia v1.11+ IOBuffer(::Vector{UInt8}) doesn't wrap directly the vector passed as input, and looking at the docstring it doesn't look like it promised to do that. Much simpler demonstration:
julia v1.10:

julia> bufdata = UInt8[]
UInt8[]

julia> iobuf = IOBuffer(bufdata, read = true, write = true)
IOBuffer(data=UInt8[...], readable=true, writable=true, seekable=true, append=false, size=0, maxsize=Inf, ptr=1, mark=-1)

julia> println(iobuf, "hello")

julia> iobuf.data
6-element Vector{UInt8}:
 0x68
 0x65
 0x6c
 0x6c
 0x6f
 0x0a

julia> bufdata
6-element Vector{UInt8}:
 0x68
 0x65
 0x6c
 0x6c
 0x6f
 0x0a

julia> bufdata === iobuf.data
true

julia master:

julia> bufdata = UInt8[]
UInt8[]

julia> iobuf = IOBuffer(bufdata, read = true, write = true)
IOBuffer(data=UInt8[...], readable=true, writable=true, seekable=true, append=false, size=0, maxsize=Inf, ptr=1, mark=-1)

julia> println(iobuf, "hello")

julia> iobuf.data
8-element Memory{UInt8}:
 0x68
 0x65
 0x6c
 0x6c
 0x6f
 0x0a
 0x00
 0x00

julia> bufdata
UInt8[]

julia> bufdata === iobuf.data
false

from julia.

the issue here is that figuring out what to change it to is pretty hard. This code seems to be built extensively on the assumption that if you give a Vector to an IOBuffer, at some unspecified later point in time, after you have done a lot of things with that IOBuffer, the Vector you passed in to the IOBuffer at the beginning will have all the bytes that were read into the IOBuffer. I think the correct change, might be to remove the bufferdata field entirely, and replace the uses of String(copy(buffer.bufferdata)) with

seekstart(buffer.buffer)
read(buffer.buffer)

That said, given the API that this package exposes, I think this would likely be a breaking change for the package.

from julia.

there have been some changes to iobuffer recently. Is Luxor accessing the internals of it?

from julia.

I don't think it's doing much of that.

Starting at:

bufdata = UInt8[]
iobuf = IOBuffer(bufdata, read = true, write = true)

there's

Cairo.CairoSVGSurface(iobuf, w, h)

which is the stream for this (Cairo.jl):

function CairoSVGSurface(stream::T, w::Real, h::Real) where {T<:IO}
    callback = get_stream_callback(T)
    surf = CairoSurface(C_NULL, w, h, stream)
    GC.@preserve surf begin
        surf.ptr = ccall((:cairo_svg_surface_create_for_stream,libcairo), Ptr{Nothing},
                         (Ptr{Nothing}, Ptr{Nothing}, Float64, Float64), callback,
                         get_stream_ptr(surf), w, h)
    end
    return surf
end

from julia.

Thanks, Mosè! I think I vaguely understand.

Do you think the behaviour of this on master is likely to change, or will I have to change my code somehow?

from julia.

We aren't currently planning on reverting the change, but we perhaps should ask triage on this one. In theory the change only affected internals, but it has broken a lot of people's code that was (intentionally or not) depending on internals.

from julia.

I'm happy to change my code - if you tell me what to change it to! 😀

from julia.

Thanks, Oscar. I think I see what's going on.

I'm not able to rebuild this area of the code at the moment, so I'll close this issue for now.

from julia.