v2, 3 partsImprove your performance with our patch

Ever since I saw spam to the kernel mailing list with that subject, I've wanted to title a suitable patch series that. This series makes executing the git startup code a whole lot faster on distros where libcurl has all the possible features, by not linking the regular programs against it.

I have in the works a series to implement my foreign-scm series as an extension to this series, mainly adding handling for the ways that importing content is different from simply getting it from a remote repository.

Daniel Barkalow (2):
  Add support for external programs for handling native fetches
  Use an external program to implement fetching with curl

Linus Torvalds (1):
  git-http-fetch: not a builtin

 Documentation/git-shim.txt           |   37 +++++++++
 Makefile                             |   12 ++-
 git.c                                |    3 -
 builtin-http-fetch.c => http-fetch.c |    5 +-
 shim-curl.c                          |  132 +++++++++++++++++++++++++++++++
 transport-shim.c                     |  142 ++++++++++++++++++++++++++++++++++
 transport.c                          |  136 +--------------------------------
 transport.h                          |    3 +
 8 files changed, 328 insertions(+), 142 deletions(-)
 create mode 100644 Documentation/git-shim.txt
 rename builtin-http-fetch.c => http-fetch.c (95%)
 create mode 100644 shim-curl.c
 create mode 100644 transport-shim.c

Daniel Barkalow venit, vidit, dixit 28.07.2009 08:08:

ROTFL :)

I'm glad it passed my spam filters... Do you offer free shipping and a rebate if I order a year's supply?

Michael

Hi,

On Tue, 28 Jul 2009, Daniel Barkalow wrote:

May I re-register my complaint about the naming?

I mean, yes, I could think of something even worse when it comes to the (ridiculously bad!) tradition of naming things "porcelain", "plumbing" and "potty", especially when it comes to "pushing objects" and then "pulling".

But now that we had a good laugh, it is time to get serious again, and give this child a _way_ better name.

I mean, you already had a good name for your foreign scm series. Why did you give that up in favor of this lousy naming?

Ciao, Dscho

2009/7/28 Johannes Schindelin <Johannes.Schindelin@gmx.de>:

How about git-http?

- Reece

Hi,

On Tue, 28 Jul 2009, Reece Dunn wrote:

If 'git-fetch' learns to search for a helper by the name of the protocol (as it should learn), this will obviously fail with svn://.

Ciao, Dscho

Johannes Schindelin wrote:

A Shim has nothing to do with plumbing or toilets. From <http://en.wikipedia.org/wiki/Shim_%28engineering%29>:

In engineering, a shim is a thin and often tapered or wedged piece of material, used to fill small gaps or spaces between objects. Shims are typically used in order to support, adjust for better fit, or provide a level surface.

I suspect that it is being used in the "adjust for better fit" sense in this case.

Regards,

Rogan

2009/7/29 Rogan Dawes <lists@dawes.za.net>:

Yes, a shim is a layer that acts as a bridge from one thing to another. In this case, it is a bridge between the transport code and the transport mechanism. Using shim as an external name is confusing -- a shim to what? The fast-import code and git-svn/hg/bzr/cvs/... could be a shim (as it is acting as a bridge between other version control systems).

Here, git-remote-* where * is the protocol name is a better naming scheme as it tells you about what the logic is doing. The documentation can be under git-remote.txt.

- Reece