git pull through http / git

git pull through http / git

[Nicolas Ferre]

Hi,

I see differences when I use git to a git.kernel.org between the one I 
fetch using http protocol and the one shown through the gitweb interface.

I cloned (or git remote add) the mmc tree and my config shows this url :
url = http://www.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc.git

but I see updates in :
http://git.kernel.org/?p=linux/kernel/git/drzeus/mmc.git;a=summary
I cannot get.

I tried from several mirrors without success.

Thanks for your help, regards,
-- 
Nicolas Ferre

Re: git pull through http / git

[Jan Engelhardt]

On Monday 2008-07-07 18:15, Nicolas Ferre wrote:

> Hi,
>
> I see differences when I use git to a git.kernel.org between the one I fetch
> using http protocol and the one shown through the gitweb interface.
>
> I cloned (or git remote add) the mmc tree and my config shows this url :
> url = http://www.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc.git
>
> but I see updates in :
> http://git.kernel.org/?p=linux/kernel/git/drzeus/mmc.git;a=summary
> I cannot get.

I don't think many care about http downloading, just because it is 
available as such.
git:// and rsync:// are available though and should be usable even 
without the http info that needs to be constantly refreshed when one 
wants to use http....

Re: git pull through http / git

[H. Peter Anvin]

Jan Engelhardt wrote:
> 
> I don't think many care about http downloading, just because it is 
> available as such.
> git:// and rsync:// are available though and should be usable even 
> without the http info that needs to be constantly refreshed when one 
> wants to use http....
> 

I thought rsync:// had the same problem (dumb transport) as http://.

I have made some sketches on a git-aware http protocol to deal with the 
"idiot IT department-managed firewall" problem; however, I have not had 
time to implement it.

	-hpa

Re: git pull through http / git

[Jan Engelhardt]

On Monday 2008-07-07 20:03, H. Peter Anvin wrote:

> Jan Engelhardt wrote:
>> 
>> I don't think many care about http downloading, just because it is available
>> as such.
>> git:// and rsync:// are available though and should be usable even without
>> the http info that needs to be constantly refreshed when one wants to use
>> http....
>
> I thought rsync:// had the same problem (dumb transport) as http://.

I thought the opposite (read: no way to prove it).
rsync has a possibility to list a directory, whereas this is not 
possible with HTTP (you always get a fancy html index page or a 
Directory Listing Denied), hence it can figure out all objects that are 
possibly located in the repository.

> I have made some sketches on a git-aware http protocol to deal with 
> the "idiot IT department-managed firewall" problem; however, I have 
> not had time to implement it.

Not wanting to downplay your work, but I'd be happier if they
would not get to have their githttp in the hope  that these
IT departments get a clue that http is not a fits-it-all solution.

Because once the workers can't do their job because the world is 
Git-dominated, they have to turn it on. But maybe then they still don't 
get the clue...hmm
tricky situation :)

Re: git pull through http / git

[H. Peter Anvin]

Jan Engelhardt wrote:
> 
>> I have made some sketches on a git-aware http protocol to deal with 
>> the "idiot IT department-managed firewall" problem; however, I have 
>> not had time to implement it.
> 
> Not wanting to downplay your work, but I'd be happier if they
> would not get to have their githttp in the hope  that these
> IT departments get a clue that http is not a fits-it-all solution.
> 
> Because once the workers can't do their job because the world is 
> Git-dominated, they have to turn it on. But maybe then they still don't 
> get the clue...hmm
> tricky situation :)
> 

Given that one of the companies so affected is Intel, which you'd think 
would have more of a clue not just for git but for God-knows-how-many 
other services...

	-hpa

Re: git pull through http / git

[Nicolas Ferre]

Jan Engelhardt :
> On Monday 2008-07-07 18:15, Nicolas Ferre wrote:
> 
>> Hi,
>>
>> I see differences when I use git to a git.kernel.org between the one I fetch
>> using http protocol and the one shown through the gitweb interface.
>>
>> I cloned (or git remote add) the mmc tree and my config shows this url :
>> url = http://www.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc.git
>>
>> but I see updates in :
>> http://git.kernel.org/?p=linux/kernel/git/drzeus/mmc.git;a=summary
>> I cannot get.
> 
> I don't think many care about http downloading, just because it is 
> available as such.
> git:// and rsync:// are available though and should be usable even 
> without the http info that needs to be constantly refreshed when one 
> wants to use http....

If only I had the choice...

You certainly know why I am using http. This feature is very useful for 
me and I was wondering why it was out of sync with the web interface.

Regards,
-- 
Nicolas Ferre

Re: git pull through http / git

[J.H.]

Keep in mind Kernel.org uses caching on gitweb and thus gitweb itself is
not necessarily accurate / up to date.  git.kernel.org also round-robins
across 4 machines which may be in various states of update, so your
fundamentally looking at a number of different possible points that may
be slightly out of sync with each other.

- John 'Warthog9' Hawley
Chief Kernel.org Admin & Original Author of the Gitweb Caching code

On Tue, 2008-07-08 at 09:22 +0200, Nicolas Ferre wrote:
> Jan Engelhardt :
> > On Monday 2008-07-07 18:15, Nicolas Ferre wrote:
> > 
> >> Hi,
> >>
> >> I see differences when I use git to a git.kernel.org between the one I fetch
> >> using http protocol and the one shown through the gitweb interface.
> >>
> >> I cloned (or git remote add) the mmc tree and my config shows this url :
> >> url = http://www.kernel.org/pub/scm/linux/kernel/git/drzeus/mmc.git
> >>
> >> but I see updates in :
> >> http://git.kernel.org/?p=linux/kernel/git/drzeus/mmc.git;a=summary
> >> I cannot get.
> > 
> > I don't think many care about http downloading, just because it is 
> > available as such.
> > git:// and rsync:// are available though and should be usable even 
> > without the http info that needs to be constantly refreshed when one 
> > wants to use http....
> 
> If only I had the choice...
> 
> You certainly know why I am using http. This feature is very useful for 
> me and I was wondering why it was out of sync with the web interface.
> 
> Regards,

Re: git pull through http / git

[Nicolas Ferre]

J.H. :
> Keep in mind Kernel.org uses caching on gitweb and thus gitweb itself is
> not necessarily accurate / up to date.  git.kernel.org also round-robins
> across 4 machines which may be in various states of update, so your
> fundamentally looking at a number of different possible points that may
> be slightly out of sync with each other.

Ok, I understand.
I am talking about a 3 days old commit. Is this normal to have no update 
  from the http side ?
I yes, I will wait...

Regards,
-- 
Nicolas Ferre

Re: git pull through http / git

[H. Peter Anvin]

Nicolas Ferre wrote:
> J.H. :
>> Keep in mind Kernel.org uses caching on gitweb and thus gitweb itself is
>> not necessarily accurate / up to date.  git.kernel.org also round-robins
>> across 4 machines which may be in various states of update, so your
>> fundamentally looking at a number of different possible points that may
>> be slightly out of sync with each other.
> 
> Ok, I understand.
> I am talking about a 3 days old commit. Is this normal to have no update 
>  from the http side ?
> I yes, I will wait...
> 

You're trying to do a git pull via http (as opposed to gitweb)?  Then 
that is dependent on the owner of the repository having enabled
"git update-server-info".

Overall, avoid pulling git via http.  It's just plain broken.

	-hpa