Jim Garlick

That seems expected to me since the consumer is in charge of the rate of date transfer.

See flux-framework/rfc#427 for a proposed protocol.

> I think the design in the RFC assumed multiple channels would return credits at the same time, but I think that is impractical, so it's just credits for each...

It's a bit unfortunate if we have to wait for the reactor loop to tell us the initial buffer size before sending it to the remote user since we _should_...

Ooh nice, does it work for an even smaller buffer like 4K?

Is this ready for another look? I wasn't sure if you were still working on it.

> We could remove this default and require remote subprocess execution to set this flag as needed. However, it ends up that the vast majority of code in flux-core sets...

I think the meaning of RUNNING here is "process was started and we have a pid", so it's not necessarily exclusive of the process being also STOPPED. Would it simplify...

@trws, even though the source changes are massively out of date here, the .clang-format changes are an improvement over what's there. Any objection if I push to this branch to...

Eh, let's close this and I'll just propose the new .clang-format in a new PR referencing this one and with authorship attributed to you.