[Monotone-devel] Re: RFC: Fake IDs

[Graydon Hoare]

Nathaniel Smith wrote:

> Well, as Zack mentioned, it is rather a pain to pass that database
> handle around everywhere (and it would be nice to pass it less
> places!).  Furthermore, there are times when you don't really mean
> that anyway -- e.g., in the pluck code, I use 3 fake rids (for marking
> the synthetic ancestor/child/workspace rosters), and there _are_ no
> other rids in play, all I actually want is 3 rids that are distinct
> from each other.  (Or maybe I actually use a real rid for one of
> those, since I happen to have it handy... it doesn't really matter,
> the point is that the constraint doesn't care about the rest of the
> db.)

Ok. I've argued with myself about my knee-jerk perception of risk here 
and convinced myself that I've been irrationally mis-perceiving relative 
probabilities. You're right: a fixed plaintext collision -- even all 
zero -- is actually *less* likely than the birthday-collision of "any 
two files". It's unnerving, but I have to accept the reasoning. Score 
one for unreasonable psychology.

I propose an Explicit and Sound Mechanism:

Make a class fake_rid_source that wraps a u64 counter and sequentially 
assigns "%40.40llX" rids. Construct one and have it hand out fake rids 
within the scope that you're using them. That makes the scope 
relationships of your fakes somewhat explicit (which is half of what I 
wanted with "checking the database") and still lets you generate a few 
within a function and throw them out without dragging a db handle into it.

Good enough?

-graydon