Re: another subtle signals issue

[Roland McGrath]

> Yeah. There's another issue too, which is the "preferred thread" thing. We 
> should probably _prefer_ threads that are interruptible as opposed to 
> threads that are in disk wait, the same way we prefer threads that are not 
> stopped. It might improve throughput.

I am really only concerned with the correctness issues, and don't have much
opinion on optimization choices like this.  It think the tradeoffs on who
it give it to vs the complexity of the scan and such depend heavily on how
many threads you have and what they are doing.