Re: [Gnu-arch-users] Re: Linus

[Thomas Zander]

Hi Niel, long time no see :)

On Sunday 12 October 2003 23:43, Neil Stevens wrote:
> On Sunday October 12, 2003 2:47 pm, Tom Lord wrote:
> >     > From: Neil Stevens <address@hidden>
> >     >
> >     > > As far as I know there's nothing preventing arch from being
> >     > > used _exactly_ like CVS, with everyone committing to the same
> >     > > archive (and there's no obstacle to moving to more distributed
> >     > > development in that case either).
> >     >
> >     > CVS has server-side hooks.  A move from CVS to arch loses that
> >     > functionality.
> >
> > I have some questions: do CVS server-side hooks get to vote on whether
> > or not a transaction completes?
>
> Yes.  Example: KDE uses them to restrict write access for sensitive
> modules like www, restrict use of "official" branch names to select
> people, and prevent common mistakes of people importing code into the
> wrong place.

More accurately; it gets to vote if a transaction can start.
Since CVS does dir-level locking (it creates lock files in the dirs that it 
locks during the transaction) it is needed to make this distinction. The 
checks for acls and other stuff is done pre-commit.
Other things I have seen is that the sourcefiles being committed are changed 
just before they are put in the repository; think source formatting or 
filtering to expand things like '$id;" to a full name.

Notice that there is also a post-commit thing going on in CVS; sending 
emails about the commit is a good example of that.

> > Do they get to postpone a second
> > transaction until after the hooks from a first transaction completes?
>
> I don't believe so.

Indeed the pre-commit checks don't delay others.  The only delay is due to 
the target dirs (remember CVS works per file, and thus you can do a commint 
on only 1 dir) being locked.

> > If the answers are both `no', then arch has server-side hooks in the
> > CVS sense too.   If either or both answers are `yes', then, yes, it's
> > reasonable to say there's a functional difference here.

I'd say that the server side (where "server" is used loosly) functionality 
is missing largely;

- sending emails after a commit/import
- having ACLs where a particular dir is restricted by the admin.
- having ACLs where particular actions (removing revisions for example) are 
restricted by the admin.

-- 
Thomas Zander