[PATCH 0/4] Speed improvement in fetch/clone

[Goffredo Baroncelli]

On Wednesday 18 January 2006 20:46, you (John Arbash Meinel) wrote:

> I'm a little concerned that your http_get class doesn't handle HTTPS,
> and it doesn't handle all of the possible return codes correctly (nor
> the fact that HTTP/1.0 would want to disconnect unless it was passed
> keep-alive or somesuch).
> 

This patch was a _little hack_ to highlight which could be the gain on
changing from a "one request per connection" to "more request per connection".
The gain was impressive ( about 10x ) in case of bazaar[*]; but the patch 
isn't ( and probably won't be) ready for a merging in bazaar.

> I think you could refactor it into a plugin, which just overrides
> transport.http.get_url into a different function. Then people could
> install it if they wanted the performance improvement, but it wouldn't
> have to officially support everything that we would want the core of bzr
> to support.
> 
> I'm also curious how it compares with Martin's pycurl work.
> 
> But it is definitely nice to see performance improve with http branches.

With this kind of patch, there isn't any need for the rsync program ( and for
the setup of the server side ! ). I hope that the pycurl work will be included
soon

> John
Goffredo

-- 
gpg key@ keyserver.linux.it: Goffredo Baroncelli (ghigo) <kreijack_AT_inwind.it>
Key fingerprint = CE3C 7E01 6782 30A3 5B87  87C0 BB86 505C 6B2A CFF9