There are a few things that I'd need to get this cleaned up:
- A hotplug managing thread. I'm spawning the first ztex thread on startup as the hotplug manager which in turn will handle (re)connects, but this should be a proper top level thread.
- Proper support for adding threads on the fly; I added a MAX_THREADS def and alloc memory for all of them on startup in thr_info[], because otherwise the other threads (long polling, watchdog, etc) would become interleaved with the mining threads and break all the "for i in len(mining_threads)" that expect 0 based and sequencial miner_threads. A simple solution to prevent the need to pre-alloc would be to move all non mining threads to the top of the array instead.
- A little more dynamic variables handling the curses output, specifically being able to add lines to the device statuses.
- When unplugging I'm setting deven to DEV_DISABLED, but there should really be a DEV_DISCONNECTED state too. I didn't need it because I search for devices being disabled and ztex, reusing those entries and threads as needed (because the only time ztex devices are disabled is when they are disconnected or malfunctioning) but if cgminer ever supports hotplugging for multiple device types we really need to know exactly which devices[] entries we can reuse.
There's more, but if you just look at my changes to cgminer.c you'll see what I mean.
What I've done works for me but I don't feel it is even close to clean to make a pull request out of it (without even going into the fact I changed a lot of tab indentations to double space
) but I would really love to work with the core developer(s) in order to come up with a plan that stands a chance of being accepted in the mainline.Update code at the usual place: https://github.com/nelisky/cgminer/tree/ztex
