[Monotone-devel] Re: Proposal for human readable revision IDs

[Steven Grimm]

On 12/09/05, Jon Bright <address@hidden> wrote:

> I also agree with this. For me,
>
> 568b-2462-456e-9a57-4326-93df-936d-4835
>
> would be much more readable than
>
> 568b2462456e9a57432693df936d4835 

To me, they're both meaningless gobbeldygook that I can't say out loud 
to someone when I want to tell them which recent checkin has a fix for 
their bug. I think the only differences between the two are that the 
first one takes less screen real estate and I can select the first one 
by double-clicking, which I can't reliably do with the hyphens there.

I guess it depends on what you want to do with revision IDs. It's not 
like you're ever actually *reading* one per se; it has no intrinsic 
meaning. There are only a few reasons I ever care about a revision ID in 
any version control system:

1. To supply it to a version-control system command or another app 
(bug-tracking system, etc.) I doubt many people will ever type a hex ID 
in by hand, so hyphens don't matter here; you'll be cutting and pasting. 
And that's easier without the hyphens because they can act as word breaks.

2. To refer to it verbally. Which is impractical with SHA1 IDs no matter 
how you render it; if I want to refer to a revision in speech I have to 
tag it.

3. To compare it against another one. Only in one case out of four 
billion is it insufficient to compare, say, the first and last four 
digits of an SHA1 hash, which is easy enough to do without the hyphens. 
The chances of making a mistake in a visual comparison are orders of 
magnitude higher than the chances that any 8 digits will actually be the 
same between two hashes.

My vote (as a mere user of monotone, not yet a contributor) is to keep 
them as they are.

Is there any significant performance penalty to having zillions of tags 
in the system? I think most people's objections to the hashcodes would 
be be satisfied with a hook script that auto-tags revisions with a much 
shorter, likely-but-not-guaranteed unique, value at commit time 
(username + timestamp, DB name + autoincremented serial number, etc.) I 
know if I had that, I'd probably never refer to a hex revision ID. What 
happens when there's a name collision between tags during a synchronize, 
anyway?

-Steve