Hi,

I'm pretty sure this worked before, but with git-1.5.4, when trying to
clone a repository via ssh from a machine that does not have git
installed in a standard system path, I get the following:

$ git clone -u /home/enewren/software/install/linux/git/bin/git-upload-pack
ssh://enewren@remote/var/scratch/enewren/votd
Initialized empty Git repository in /home/newren/devel/votd/.git/
remote: fatal: exec pack-objects failed.
remote: aborting due to possible repository corruption on the remote side.
fatal: early EOF
fatal: index-pack failed
fetch-pack from 'ssh://enewren@remote/var/scratch/enewren/votd' failed.

I tried doing a local clone of the same repository (worked fine), a
'git fsck -full' on the relevant repository (reported no errors), and
a few other things.  After a bit of digging, I found a really ugly
hack that works around my problem.  After recompiling git on the
remote machine with the attached patch, the clone operation succeeded.


Thanks,
Elijah

git-pull(1)
[...]
       --upload-pack <upload-pack>
              When  given,  and  the  repository  to  fetch from is handled by
              git-fetch-pack, --exec=<upload-pack> is passed to the command to
              specify non-default path for the command run on the other end.

-- 
Jakub Narebski
Poland
ShadeHawk on #git
-

The OP *did* use this option (rather, its short form, -u), see above.

The problem is that git-upload-pack (which is not a built-in) does not
call setup_path() and so does not extend PATH to contain the special
installation location. Now, when git-upload-pack tries to exec
git-pack-objects, it fails since this is not in PATH.

A quick work-around for Elijah is to add

   GIT_EXEC_PATH=/home/enewren/software/install/linux/git/bin

to .profile on the remote host.

-- Hannes

-

Hi,



I guess you meant .bashrc, as .profile is not sourced when using ssh 
transport (it does not spawn a shell) AFAIR.

Ciao,
Dscho

-

As far as I can tell, setting paths in .bashrc doesn't really work (or
else I'm just doing it wrong).  If it did, I would have never hit this
bug.  Observe the difference between (feel free to replace PATH with
GIT_EXEC_PATH; same general result occurs):

  $ ssh localhost
  # Wait for connection to be made, then run
  $ echo $PATH

AND

  $ ssh localhost 'echo $PATH'

AND

  $ ssh localhost 'source .bashrc
  echo $PATH'

The first and the third give the same result, but the second gives
something different.  It is the second form that git uses, meaning
that my paths never get set up.

In my little git wrapper script, I put in some code to work around
this little issue and find git-upload-pack for the user if it can (by
ssh'ing to the machine and sourcing their .bashrc in the ssh command
if necessary; might be a hack, but it makes things nicer for me than
always specifying the -u flag).

Cheers,
Elijah
-

I had this problem as well, and the only solution I found was setting
up my .ssh/environment file.
I guess when you specify a command for ssh, it executes this command
by itself, not within a shell environment.  Usually it is bash that
sources .bashrc and .profile, so you don't get your paths set up.
Unfortunately .ssh/environment is not a scripting environment, so you
have to just specify your full path there, based on what you get when
you echo $PATH from the shell.
Also, this requires putting the following in sshd_config:

PermitUserEnvironment yes

Which, and this is the biter, is not usually a default on most
systems, as far as I can see.. so it could mean bugging the sysadmin
if it's not your machine, which is not always convenient.  Can't see
any alternative though.

Steve
-

From: Johannes Sixt <johannes.sixt@telecom.at>

Since git-upload-pack has to spawn git-pack-objects, it has to make sure
that the latter can be found in the PATH. Without this patch an attempt
to clone or pull via ssh from a server fails if the git tools are not in
the standard PATH on the server even though git clone or git pull were
invoked with --upload-pack=/path/to/git-upload-pack.

Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
---

How about this (almost) one-liner instead?

-- Hannes

 upload-pack.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/upload-pack.c b/upload-pack.c
index 7e04311..51e3ec4 100644
--- a/upload-pack.c
+++ b/upload-pack.c
@@ -620,6 +620,9 @@ int main(int argc, char **argv)

 	if (i != argc-1)
 		usage(upload_pack_usage);
+
+	setup_path(NULL);
+
 	dir = argv[i];

 	if (!enter_repo(dir, strict))
-- 
1.5.4.rc3.24.g25a9a

-

Hi,


I'm fine with it.

Ciao,
Dscho
-

Hi,


But I also made this:

-- snipsnap --
[PATCH] Make upload-pack a builtin

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
---
 Makefile                               |    2 +-
 upload-pack.c => builtin-upload-pack.c |    8 +++++---
 builtin.h                              |    1 +
 git.c                                  |    1 +
 4 files changed, 8 insertions(+), 4 deletions(-)
 rename upload-pack.c => builtin-upload-pack.c (98%)

diff --git a/Makefile b/Makefile
index 83c359a..1792039 100644
--- a/Makefile
+++ b/Makefile
@@ -259,7 +259,6 @@ PROGRAMS = \
 	git-show-index$X \
 	git-unpack-file$X \
 	git-update-server-info$X \
-	git-upload-pack$X \
 	git-pack-redundant$X git-var$X \
 	git-merge-tree$X git-imap-send$X \
 	git-merge-recursive$X \
@@ -392,6 +391,7 @@ BUILTIN_OBJS = \
 	builtin-update-index.o \
 	builtin-update-ref.o \
 	builtin-upload-archive.o \
+	builtin-upload-pack.o \
 	builtin-verify-pack.o \
 	builtin-verify-tag.o \
 	builtin-write-tree.o \
diff --git a/upload-pack.c b/builtin-upload-pack.c
similarity index 98%
rename from upload-pack.c
rename to builtin-upload-pack.c
index 7e04311..207754c 100644
--- a/upload-pack.c
+++ b/builtin-upload-pack.c
@@ -10,6 +10,7 @@
 #include "revision.h"
 #include "list-objects.h"
 #include "run-command.h"
+#include "builtin.h"
 
 static const char upload_pack_usage[] = "git-upload-pack [--strict] [--timeout=nn] <dir>";
 
@@ -593,14 +594,14 @@ static void upload_pack(void)
 	}
 }
 
-int main(int argc, char **argv)
+int cmd_upload_pack(int argc, const char **argv, const char *prefix)
 {
 	char *dir;
 	int i;
 	int strict = 0;
 
 	for (i = 1; i < argc; i++) {
-		char *arg = argv[i];
+		const char *arg = argv[i];
 
 		if (arg[0] != '-')
 			break;
@@ -620,12 +621,13 @@ int main(int argc, char **argv)
 
 	if (i != argc-1)
 		usage(upload_pack_usage);
-	dir = argv[i];
+	dir = xstrdup(argv[i]);
 
 	if (!enter_repo(dir, strict))
 		die("'%s': unable to ch...
[ message continues ]

<snip>

I tried this patch yesterday, but it seems to suffer from the same
problem?  I had to manually apply the patch from Johannes Sixt on top
of this ("manually" since there was a trivial conflict) in order for
it to solve the problem.

Anyway, it works for me now.  Thanks for the quick fixes!

Elijah
-

Please post patches inline, not attached.

The problem you've encountered is to do with the PATH setup on the
remote end. See a previous thread
(http://thread.gmane.org/gmane.comp.version-control.git/72673/focus=72805)
for some solutions.


Dave.
-