Re: http-push as a builtin ? (Was: Is there a reason to keep walker.c ?)

On Mon, 28 Jan 2008, Mike Hommey wrote:

On Mon, Jan 28, 2008 at 08:17:49AM +0100, Mike Hommey wrote:

quoted

On Sun, Jan 27, 2008 at 04:23:17PM -0500, Daniel Barkalow wrote:

quoted

On Sun, 27 Jan 2008, Mike Hommey wrote:

quoted

Hi,

While working on the http code refactoring, I got to wonder if the
walker.c "wrapper", that is only used for the http transport, is still
worth keeping. If there are plans for others transport to use this code,
obviously, it would be worth keeping, but on the contrary, I think it
would simplify the http transport code even more. What do you think ?

It would be a good base for sftp (i.e. dumb file access over ssh). In 
fact, I think stuff should ideally be moved into walker.c such that the 
HTTP-specific code just handles access to files by filename and the logic 
of what files to request in what order is in walker.c. I think this would 
get the simplification you're looking for while making it easy to add sftp 
or any other situation where you have only slow remote filesystem-like 
access to the repository.

I like this idea. I'll probably implement that, then.

BTW, would there be objections to have http-push as a builtin ?

Not from me. Actually, it would be ideal to call its functions directly 
from transport.c and deprecate the separate command. (And possibly 
separate the control structure from the HTTP code and move the former into 
walker.c where it could be used by sftp)

	-Daniel
*This .sig left intentionally blank*