Fun Project: xTalk GUID Generator

[Brian Yennie]

I think what makes the GUID spec so bewildering is the fact that GUID  
needs are highly application-specific.

Your point is actually right on target. GUID generation is non- 
perfect and probabilistic. Just generating random data isn't really  
conceptually any worse, it just gives a much higher probability of  
conflicts, since random number generators aren't pure and can be  
influenced by things like the random seed, different machines, system  
shutdowns, etc.

What the GUID spec does is create a system in which it's practically  
impossible (as opposed to theoretically impossible) to generate the  
same GUID twice. In order to do this it basically calls on a handful  
of different unique data and mashes it all together. It also strives  
for variation (since checking the IP address has nothing to do with  
generating a random number, for example) so that there is no one  
point of failure. But in the end, it's just the difference between  
"unlikely" and "super super super super super unlikely", and really  
it's the application that is going to define how many "super"s you  
need =).

For the vast majority of applications, an incrementing counter or  
timestamp with a little app logic is all you need to get unique IDs  
for your uses.

>  From what I can dimly understand of the various versions of the
> spec, there are time-based, name-based and random-based versions.
>
> All of them are built from 1 x 60 bit value, 1 x 14 bit value and 1 x
> 48 bit value, plus 4 bits for the version id, and 2 bits for the
> 'variant' (I haven't worked out exactly what that is, but apparently
> it should be 1 0.)
>
> In the random version, each of these values (apart from the version
> and variant) is generated randomly, or pseudo-randomly.
>
> So my naive question is, for the random version, what would be wrong
> with simply generating a random sequence of 128 1s and 0s, sticking
> the version and variant bits in the right places, and then
> baseConverting the whole thing in groups of 4 to Hex digits. Add the
> formatting dashes, and it's done.
>
> This is probably very naive indeed, perhaps someone here understands
> the issues well enough to comment.