History File over 2GB

[greg andruk]

Russ Allbery wrote:

> I don't think I understand.  MODE CANCEL is still an interface to innd.
> Why does it help any to have that part of innd doing expensive work rather
> than the article processing part?

You still want to tell innd what's going away so it can knows what to 
refuse.  For the time being, I'm leaving in the assumption that history 
only has one writer.

> Or do you mean offloading that support as well, into an overchan-like
> daemon that just processes cancels?  Hm.  I think all of our overview
> methods lock against multiple writers, so that might work.

Yeah, the actual deletions would be done by the process feed.  It was 
really the lack of something better than ctlinnd cancel that convinced 
me to leave cancels out of the controlchan thingy.  That, and being too 
lazy to look into how to feed supersedes to the thing.