[ On Tuesday, February 15, 2000 at 09:57:57 (-0700), Tobias Weingartner wrote: ]
> Subject: Re: more on fixing 'cvs add'
>
> Greg, I'm going to disagree with this proposal.
OK.
> In some sense, and correct
> me if I'm wrong here, it looks like you wish to remove 'cvs add' for all the
> directory cases. IE: since directories are not versioned, you wish to have
> 'cvs add/rm' work only with files, dealing with directories as an artifact.
Yes, exactly.
> 1) I'm hoping to see directories versioned to some degree in the future.
I'm hoping exactly the opposite stays true. Directory versioning is
extremly horrible stuff to get into and has been shown to be entirely
unecessary in a simple tool like CVS.
> Treating them, however wrong currently, as a pseudo object at the user level
> will be easier to integrate into the UI in the future.
That's the problem indeed that I'm working hard against! :-)
> 2) Searching up the tree is not always possible. There are filesystems out
> there in use, where searching up the tree is a very expensive, and sometimes
> impossible, operation. Certain "instances" of AFS come to mind, and other
> such filesystems.
I think you're wrong there, and even if you're right I doubt it matters
at all to CVS. Given the existance of a getcwd() routine that always
returns a "fully qualified" pathname starting at '/' it is always 100%
guaranteed that you can search in each parent directory, by definition
of what a pathname to CWD is. You just strip off each basename
component until you find a valid CVS administrative sub-directory in the
resulting path. It's absolutely trivial and very easy to add
administrative directories for the intervening parent directories.
Note that it is necessary and critical to maintaining the state of the
workspace that the intervening administrative files be created, BTW,
especially if the optimisation of ignoring the contents of unknown
directories is to be kept in place.
> I would rather see the semantics that the above would return an error from
> CVS, saying something about admin files not found.
Sorry, but that's not possible to meet the goals of my proposal, nor is
it necessary for a correctly and safely functioning system.
> While not 100% pure (in the sense you still 'cvs add' directories), I believe
> this model offers the least confusion for current users of CVS. And judging
> by the list traffic, the less confusion about these sort of things, the better
> in the long run.
I beg to differ. The current confusion will melt away silently. You
need to look back at much older archives to see how long people have
been confused by the strangeness of the current 'cvs add'
implementation. I've been wanting to fix this since 1994 and now I know
how to fix it.
--
Greg A. Woods
+1 416 218-0098 VE3TCP <[EMAIL PROTECTED]> <robohack!woods>
Planix, Inc. <[EMAIL PROTECTED]>; Secrets of the Weird <[EMAIL PROTECTED]>
