[Gnu-arch-users] Arch revision namespace

[Mikhael Goikhman]

On 24 May 2005 16:34:55 -0700, Tom Lord wrote:
> 
>   > As a policy *tool* I'm very much in favour of a namespace that knows
>   > about people and projects and branches and ... whatnot. However the
>   > management of revision identity and the namespace should not be coupled
> 
> They should be coupled differently, is all.
> 
> A good convention for naming commits seems to be:
> 
>       <user-name>/<checksum>
> 
> where the `<user-name>' is nearly anything a client cares to pick and
> `<checksum>' is a contents-summary of the resulting revision.  This 
> both generalizes the requirements on and simplifies the implementation
> of the revision-builder part of the system.
> 
> Arch 1.x is bogus by too narrowly constraining `<user-name>' and
> omitting `<checksum>' altogether -- but that's easily remedied.

And how exactly adding a completely random suffix to the namespace makes
it non-bogus and maybe more intuitive?

> Of course, by one mechanism or another, clients must be able to
> compute a list of the names of the ancestors of a given commit from
> the commit itself.  This returns to the familiar question of whether
> and how to support some sort of archive-side ancestry-list caching.

Yes, having to figure out the tree commit history, the latest archive
revision, merged in revisions, and find unmerged partners, all make this
<checksum> stuff look totally inconvenient and unpleasant to work with.

I still think the current Arch namespace is close to the optimal.

[Heh, Tom even uses the git terminology now, "commit" instead of
"revision/changeset".]

Regards,
Mikhael.