Add a `pass_through` option about iocapture.jl HOT 6 CLOSED

This seems like a good feature to have here. I would indeed call it something other than tee though, since that does feel a bit jargony.

from iocapture.jl.

I'm going to play around a little and see if I come up with something that works

from iocapture.jl.

For context: https://discourse.julialang.org/t/a-tee-option-for-iocapture/108642

from iocapture.jl.

Well, here's a start:

output = IOBuffer()
temp = IOBuffer()
buffer_redirect_task = @async begin
    write(temp, pipe)
    temp_data = take!(temp)
    write(output, temp_data)
    write(default_stdout, temp_data)
end

This seems to capture the data from the pipe while also passing it through the normal stdout.

Is this a good idea, or does this have the potential to block something or slow things down at a low level? If there's an expert on the IO internals that can sign off on this, I can easily build that into a full PR.

from iocapture.jl.

This might work, too, and avoid the allocation of temp_data?

output = IOBuffer()
temp = IOBuffer()
buffer_redirect_task = @async begin
    n_bytes = write(temp, pipe)
    seek(temp, 0)
    write(output, read(temp, n_bytes))
    seek(temp, 0)
    write(default_stdout, read(temp, n_bytes))
end

from iocapture.jl.

Ugh… neither of these actually works. I should have tested it on a longer-running calculation to start with. It seems like the writing to default_stdout gets buffered until the task ends. I've tried adding a flush(default_stdout), too.

Update: this one works better:

bufsize = 128
buffer = Vector{UInt8}(undef, bufsize)
buffer_redirect_task = @async begin
    while !eof(pipe)
        nbytes = readbytes!(pipe, buffer, bufsize)
        data = view(buffer, 1:nbytes)
        write(output, data)
        write(default_stdout, data)
    end
end

from iocapture.jl.