v7, 2 partsDocumentation/remote-helpers: Rewrite description

Rewrite the description section to describe what exactly remote
helpers are and the need for them. Also mention the curl family of
remote helpers as an example.
---
 The previous version got corrupted and didn't apply cleanly for some reason.
 Additionally split it into two parts now.

 Documentation/git-remote-helpers.txt |   32 ++++++++++++++++++++++++++------
 1 files changed, 26 insertions(+), 6 deletions(-)

Hello,

On 2010-03-28 14:03, Ramkumar Ramachandra wrote: [...]

[...]

[...]

Oops, "needs" is repeated twice here...

Apart from that, it seems nice and helps to better understand what they should be used for.

It doesn't really concern this patch but If I'm not wrong, there still is no documentation written on how exactly git invokes the remote helpers. A good description was given previously by Ilari Liusvaara (so I've added you, Ilari in the cc list). The description can be seen here:

http://lists.zerezo.com/git/msg712892.html

Should we add this information on this man page?

Hi,

Oops. Sorry about the silly mistake.
Junio: Could you please make this correction and apply the patch? I'm
writing another patch documenting invocation which depends on this
patch.

Heya,

On Sun, Mar 28, 2010 at 19:17, Gabriel Filion <lelutin@gmail.com> wrote:

Yes we should. Also, I have almost finished a 'testgit' helper, which simply invokes git-fast-import and git-fast-export behind the scenes. So it's really a clumsy way to communicate with another git repository but using the remote-helper import/export API. I've also written tests for it, so it should be pretty clear from the tests and the simple helper how to implement your own.

On Sun, 28 Mar 2010, Gabriel Filion wrote:

That information ought to be in the documentation, but possibly not on this man page in particular. I think it would be better to document that part in the documentation of the code and programs that call the helper, not in the helper documentation. In general, I'd like the helpers to make as few assumptions about their caller as possible, and similarly keep the information about the caller out of their documentation.

Also, I think it would be more helpful in general to get that information into documentation of the transport and remote code and configuration of these things, because users are going to come at this from that side ("I want to access an SVN repository, how does that work?") rather than from the helper documentation side ("I want to exercise git-remote-svn's export support, just to run that code").

	-Daniel
*This .sig left intentionally blank*

Hi,

I agree that the callers need to document the subset of the invocations they make to remote helpers. I think we can defer this until we have a real remote helper in `git.git` that actually interfaces with a foreign versioning system.

I've thought about documenting the full set of invocations in the code for the developer, but there's a problem. Here's an excerpt from remote-curl.c, showing how it parses its command line arguments:

	remote = remote_get(argv[1]);

	if (argc > 2) {
		url = argv[2];
	} else {
		url = remote->url[0];
	}

Unfortunately, I don't see where else this documentation can fit in: if it were to go into a specific remote helper's code, then it'll have to be duplicated for all the remote helpers, since all of them parse options similarly. It certainly cannot go into remote.c or transport-helper.c, because they have little/ nothing to do with the actual argument parsing.

I could try modifying the documentation I've written to serve more to specify "how remote helpers are invoked" and less about "how callers invoke remote helpers", and try to fit it in this manpage. It's more of a developer manpage and less of an end-user manpage as it is. Or we could create another page about remote helpers intended to be read exclusively by developers. What are your thoughts on this?

-- Ram

Ramkumar Ramachandra wrote:

[...]

One possibility: new manpage, called giturl(7) or something, with:

 - the information from your patch, reformatted a little to be from
   the caller’s perspective;

 - the information currently in the GIT URLS and REMOTES sections 
   of git-pull(1) and and other urls-remotes.txt includers;

 - pointers to appropriate high-level and low-level documentation
   for more information.

This would at least avoid some duplication of text in explaining how the [remote "<name>"] setups work.

One possibility would be to put it in Documentation/technical/transport.txt or some similarly named new file. Later that file could expand to an overview of the transport layer, which would be nice to have.

Files in Documentation/technical do not get installed as manpages, which would make this less convenient when writing a new helper without a full documentation tree available.

More importantly, the “how to configure access to a foreign repository” aspect of what you are writing is really more pertinent to users than remote helper developers. Remote helper developers only need to know “first argument is a remote nickname or some nonsense with a colon; second argument is a transport-native address identifying the remote repository; second argument can be omitted only if a remote nickname was used”.

HTH, Jonathan

On Mon, 29 Mar 2010, Ramkumar Ramachandra wrote:

I think getting information on what the helper is supposed to do with its command-line arguments into this man page would be good, and the appropriate focus for the man page.

The current answer is this:
 argv[1] is the name of a remote, which may be a nickname or some more 
  direct name; in any case, remote_get() will produce the available data 
  for it.
 argv[2] is the URL, if there is one. Since a remote could have more than 
  one URL, the helper gets the URL it should be handling. Some systems 
  don't use URLs, and these will just look at the remote or something like 
  that.

The rest of what you wrote is true, but it's really more information on how the attributes of remotes are determined than what the helper should be worrying about.

	-Daniel
*This .sig left intentionally blank*

On Mon, Mar 29, 2010 at 9:09 PM, Daniel Barkalow <barkalow@iabervon.org> wrote:

Got it. I'll prepare a patch in the evening.

-- Ram

Hi,

regarding...

On Mon, Mar 29, 2010 at 2:03 AM, Ramkumar Ramachandra <artagnon@gmail.com> wrote:

and...

I think "family" is confusing. They're all symlinks of git-remote-curl, if I remember correctly.

But then, I suspect you'll be rewriting this portion, since Ilari's notes has a thing or two why helper names have the protocol in them.

Hi,

Actually, there's no Makefile target called git-remote-curl. See Makefile:1122 to Makefile:1126. They're copies in my build.

	REMOTE_CURL_PRIMARY = git-remote-http$X
	REMOTE_CURL_ALIASES = git-remote-https$X git-remote-ftp$X git-remote-ftps$X
	REMOTE_CURL_NAMES = $(REMOTE_CURL_PRIMARY) $(REMOTE_CURL_ALIASES)
	PROGRAM_OBJS += http-fetch.o
	PROGRAMS += $(REMOTE_CURL_NAMES)

Right. Thanks for pointing this out.

-- Ram

Hi,

On Sun, Mar 28, 2010 at 11:33 PM, Ramkumar Ramachandra <artagnon@gmail.com> wrote:

In accordance with Junio's 4th April "What's Cooking" report, could some remote helper experts kindly Ack this patch?

Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>

Thanks! -- Ram

Ramkumar Ramachandra <artagnon@gmail.com> writes:

I'll copy and paste this, but please make it a habit to sign off your patch when committing if you plan to be a regular contributor to the project.

I've re-read all the discussions that led to v7 patch and then re-read the patch [1/2]. I'll queue it with a minor tweak.

Thanks.

Sorry about that. I'll auto sign-off and pay extra attention next time.

-- Ram