Re: more svn_ra.h notes

Greg Stein <gstein@lyra.org> writes:

> *) svn_ra_do_checkout() takes a URL. Unlike update, this *is* necessary for
> checkout since it doesn't have SVN/repository to look at. However, I
> think it is redundant with svn_ra_open's repository_name.
> [ I think the latter is probably best called repository_URL since I'd
> like to see that all repositories are identified through some kind of
> URL construction. ]

Ah, right. svn_ra_open should take an svn_string_t *repository_URL
argument. This URL could be http:// or file:/, depending on the RA
implementation. That makes sense.

Perhaps, then, svn_ra_do_checkout() shouldn't take a URL, but a path.
This path would be an absolute path in the Subversion filesystem.

Of course, this might be awkward programmatically. If user types

   svn checkout http://foo.bar.com/project/path

Then we'd have to decompose the input into

  svn_ra_open (http://foo.bar.com)
  svn_ra_do_checkout (project/path)
 
Hmm. Is this lame?

> *) "recurse" flags need to be passed to each operation

You mean ra_do_checkout() and ra_do_update() here? Are you thinking
about DAV's {0, 1, infinity} recursion levels?

This just seems odd to me. When I tell CVS to checkout
repository:/path, I don't need to tell it how much to recurse.
Recursion is the norm. In Subversion, are users ever going to want to
checkout a path *non*-recursively?

> *) what other flags? if there are many, then maybe we should reconsider how
> to pass them?

Huh? Are you asking for more flags?
Received on Sat Oct 21 14:36:15 2006