Fwd: git-receive-pack --stateless-rpc

Fwd: git-receive-pack --stateless-rpc

[Yiyi Sun]

Hi,

I am creating a Git smart HTTP protocol wrapper and run into two
problem. I tried the msysGit mailing list and was told to ask them to
the main git mailing list.

Here is the story. During 'git push',  I successfully processed the
initial GET request of '/info/refs?service=git-receive-pack' by
sending back the output of 'git receive-pack --stateless-rpc
--advertise-refs'. Then msysGit did a the POST request. I
invoked 'git receive-pack --stateless-rpc', which returned:

0030 000eunpack ok
0019ok refs/heads/master
00000000

Question #1, Is the result supposed to be something below instead?

000eunpack ok
0019ok refs/heads/master
0000

Question #2, msysGit cannot parse the result, it displays error message:

fatal: protocol error: bad line length character:
003
fatal: The remote end hung up unexpectedly

It seems that msysGit can read the http response of HTTP GET, but not
of HTTP POST.


Can you please help ?

Thanks,

yysun

Re: git-receive-pack --stateless-rpc

[Shawn Pearce]

On Thu, Jun 17, 2010 at 6:45 AM, Yiyi Sun <yiyisun@gmail.com> wrote:
> I am creating a Git smart HTTP protocol wrapper and run into two
> problem. I tried the msysGit mailing list and was told to ask them to
> the main git mailing list.
>
> Here is the story. During 'git push',  I successfully processed the
> initial GET request of '/info/refs?service=git-receive-pack' by
> sending back the output of 'git receive-pack --stateless-rpc
> --advertise-refs'. Then msysGit did a the POST request. I
> invoked 'git receive-pack --stateless-rpc', which returned:
>
> 0030 000eunpack ok
> 0019ok refs/heads/master
> 00000000

That doesn't look right to me.  There appears to be a space at the
start of the packet (between 0030 and 000e), which would confuse the
protocol.  Otherwise, that appears to be a valid response from
receive-pack assuming the client asked for side-band-64k in the
request, and it pushed only refs/heads/master.

> Question #1, Is the result supposed to be something below instead?
>
> 000eunpack ok
> 0019ok refs/heads/master
> 0000

Depends on whether or not the client asked for side-band-64k.  If it
did, we wrap it up in a another pkt-line framing like above, and end
with another flush-pkt.  If the client didn't ask for side-band-64k,
then the above would be the correct response.

> Question #2, msysGit cannot parse the result, it displays error message:
>
> fatal: protocol error: bad line length character:
> 003
> fatal: The remote end hung up unexpectedly

This is weird.  "003" isn't valid, its not 4 bytes long.  What was
that trailing final 4th byte right there?  If its a space like it is
above (though then its really in the wrong spot, isn't it?) that would
throw it all off.

It sounds to me like your wrapper is mangling the data that is passing
through it on the way back to the client.

-- 
Shawn.

Re: git-receive-pack --stateless-rpc

[Shawn O. Pearce]

Yiyi Sun <yiyisun@gmail.com> wrote:
> Thanks Shawn for you quick reply. Now I have the answer to question #1.
> 
> You are so good that spotted the space between 0030 and 000e. It is a
> byte of value 0x1. I can reproduce the issue using the attached file
> w/o goimg through my program:
> 
> --save the file (named input) to and goto the directory
> git init --bare test
> cd test
> cat ../input | git receive-pack .
> 
> It returns message below, where you can see a funny char there.
> 
> 00700000000000000000000000000000000000000000 capabilities^{}  report-status dele
> te-refs side-band-64k ofs-delta
> Unpacking objects: 100% (6/6), done.
> 0030☺000eunpack ok
> 0019ok refs/heads/master
> 00000000
> 
> 
> For the "003" error. I tired to grab the HTTP output w/o using msysGit. It is
> 
> HTTP/1.1 200 OK
> Server: ASP.NET Development Server/10.0.0.0
> Date: Thu, 17 Jun 2010 13:18:44 GMT
> X-AspNet-Version: 4.0.30319
> Cache-Control: private
> Content-Type: application/x-git-receive-pack-result
> Content-Length: 52
> Connection: Close
> 
> 0030\x01000eunpack ok
> 0019ok refs/heads/master
> 00000000
> 
> It looks normal. Besides msysGit can candle the output of 'git
> receive-pack --stateless-rpc
> --advertise-refs' via HTTP GET the same way my program returns the result.
> 
> Regards,
> 
> yysun
> 
> On Thu, Jun 17, 2010 at 10:00 AM, Shawn Pearce <spearce@spearce.org> wrote:
> >
> > On Thu, Jun 17, 2010 at 6:45 AM, Yiyi Sun <yiyisun@gmail.com> wrote:
> > > I am creating a Git smart HTTP protocol wrapper and run into two
> > > problem. I tried the msysGit mailing list and was told to ask them to
> > > the main git mailing list.
> > >
> > > Here is the story. During 'git push',  I successfully processed the
> > > initial GET request of '/info/refs?service=git-receive-pack' by
> > > sending back the output of 'git receive-pack --stateless-rpc
> > > --advertise-refs'. Then msysGit did a the POST request. I
> > > invoked 'git receive-pack --stateless-rpc', which returned:
> > >
> > > 0030 000eunpack ok
> > > 0019ok refs/heads/master
> > > 00000000
> >
> > That doesn't look right to me.  There appears to be a space at the
> > start of the packet (between 0030 and 000e), which would confuse the
> > protocol.  Otherwise, that appears to be a valid response from
> > receive-pack assuming the client asked for side-band-64k in the
> > request, and it pushed only refs/heads/master.
> >
> > > Question #1, Is the result supposed to be something below instead?
> > >
> > > 000eunpack ok
> > > 0019ok refs/heads/master
> > > 0000
> >
> > Depends on whether or not the client asked for side-band-64k.  If it
> > did, we wrap it up in a another pkt-line framing like above, and end
> > with another flush-pkt.  If the client didn't ask for side-band-64k,
> > then the above would be the correct response.
> >
> > > Question #2, msysGit cannot parse the result, it displays error message:
> > >
> > > fatal: protocol error: bad line length character:
> > > 003
> > > fatal: The remote end hung up unexpectedly
> >
> > This is weird.  "003" isn't valid, its not 4 bytes long.  What was
> > that trailing final 4th byte right there?  If its a space like it is
> > above (though then its really in the wrong spot, isn't it?) that would
> > throw it all off.
> >
> > It sounds to me like your wrapper is mangling the data that is passing
> > through it on the way back to the client.

Oh, yea.  That byte 0x01 is the sideband stream number.  I forgot
about that needing to be there. :-)

I think your wrapper program is dropping the data there or isn't
being 8 bit clean and is mangling the value.  That's part of
the stream.

-- 
Shawn.