pipeline errors with a IOBuffer as first argument

[joa-quim]

According to the docs pipeline accepts a IOBuffer as first argument, but (and see also)

julia> run(pipeline(IOBuffer("a xyz b"), `grep xyz`))
ERROR: MethodError: `uvtype` has no method matching uvtype(::Base.AbstractIOBuffer{Array{UInt8,1}})
 in _jl_spawn at process.jl:253
 in anonymous at process.jl:415
 in setup_stdio at process.jl:403
 in spawn at process.jl:414
 in spawn at process.jl:293
 in run at process.jl:530

[vtjnash]

documentation isn't really intended to be legalistic. the operating system only understands how to use native I/O streams (like Pipe), while Julia can create many other types of I/O stream.

[Keno]

This seems like a reasonable feature request though. The semantics seem pretty clear (write the contents of the IOBuffer as STDIN).

[vtjnash]

we already have an API for that (writeall). but that's not the right semantics for this (IOBuffer is a File abstraction, not a Stream abstraction). then there's also the problem of defining the error handling model.

re-closing as "not a bug".

[tkelman]

According to the docs pipeline accepts a IOBuffer as first argument

We should at least clarify the docs then? Maybe not intended to be legalistic, but if people are getting confused that's a doc bug.

[joa-quim]

To clarify IOBuffer says

IOBuffer() → IOBuffer
    Create an in-memory I/O stream.

but than I can't use it in pipeline. How can I use then? I mean how can I put my string in STDIN and call pipeline on it?

[vtjnash]

SO or julia-users is a better place for questions (you're looking for open(p, "w"), not IOBuffer)

[joa-quim]

Sorry, it seamed appropriate to ask the question here since it's an consequence of doc contradiction. And BTW I had asked this before in julia-users. (and I still need help)

[nalimilan]

Reopening as a doc issue. Though adding a method which would raise an error with an explanation would be even more helpful.

[stevengj]

See also #26061, where @vtjnash suggested pipeline(exename, stdin=IOBuffer(input)), and the implementation could be modeled on the code in that PR.

[KristofferC]

Works now, #30278.