tar_filesystem.jl

using Tar

mutable struct InMemoryFile
    size::Int
    pos::Int
    str::Union{Nothing, String}
end

mutable struct InMemoryFileSystem
    d::Dict{String, InMemoryFile}
    tarball_io::IOStream
    out_io::IOBuffer
    buf::Vector{UInt8}
end

function readfile(ref::InMemoryFileSystem, path::AbstractString)
    file = ref.d[path]
    file.str === nothing || return file.str
    seek(ref.tarball_io, file.pos)
    Tar.read_data(ref.tarball_io, ref.out_io; size=file.size, buf=ref.buf)
    file.str = String(take!(ref.out_io))
end

function create_inmemory_filesystem(path::AbstractString)
    tarball = open(path, "r")
    buf = Vector{UInt8}(undef, Tar.DEFAULT_BUFFER_SIZE)
    system = InMemoryFileSystem(Dict{String, InMemoryFile}(), tarball, IOBuffer(), buf)
    Tar.arg_read(tarball) do tar
        Tar.read_tarball(predicate, tar; buf=buf) do hdr, _
            if hdr.type == :file
                p = position(tar)
                Tar.skip_data(tar, hdr.size)
                system.d[hdr.path] = InMemoryFile(hdr.size, p, nothing)
            end
        end
    end
    return system
end

# Test

Tar.create(joinpath(homedir(), ".julia/registries/General"), "General.tar")
system = create_inmemory_filesystem("General.tar")
data = readfile(system, "E/Example/Package.toml")
print(data)

This is really nice! I've created a version that works on compressed tar archives here: https://gist.github.com/JBlaschke/ec0f4ece417bdfe1428a85c78c0804f7