Re: [PATCH 2/2] remote-curl.c: fix rpc_out()

From: Tay Ray Chuan <rctay89@gmail.com>
Date: Nov 23, 2009, 17:54 UTC
Message-ID: <20091124015415.c22d07c1.rctay89@gmail.com>
In-Reply-To: <alpine.DEB.1.00.0911231137170.4985@pacific.mpi-cbg.de>

Hi,

On Mon, Nov 23, 2009 at 6:39 PM, Johannes Schindelin <Johannes.Schindelin@gmx.de> wrote:

> I guess you meant "not be enough", as an int can hold a pretty large
> number until it turns negative.

I really did mean 'enough' - enough to trigger the use of chunked encoding. I think the most important fix here is forcing rpc_out to (according to curl) memcpy at most size_t max bytes (the removal of the extraneous ';' addresses this). Pushing with chunked transfer would fail with this extra semicolon.

Removing the possibility of a negative size_t was a preventive measure, and, like you mentioned, requires a larger repository, so it's harder to test for.

I probably should separate these issues into separate patches.

Show 5 quoted lines

> So I think in this case it is more harm- than helpful to have a test case.
>
> For future reference: if you need a repository with special featurs
> for testing, it is best to generate it in a test script (see the many test
> cases labeled 'setup' in our test suite for examples).

Here's what I came up with: use the git repository which fetched the test suite, and use the environment variable GIT_REMOTE_REFSPEC to specify the remote refspec which the tester fetches git from.

  if test -z "$GIT_REMOTE_REFSPEC"; then
  	say 'skipping test, the remote for git is not specified'
  else
  	test_expect_success 'push with chunked encoding' '
  		OWD=$(pwd) &&
  		cd $TEST_DIRECTORY/../.git/ &&
  		REPO=$(pwd) &&
  		cd "$OWD" &&
  		echo "$REPO"/objects > .git/objects/info/alternates &&
  		git fetch "$REPO" "$GIT_REMOTE_REFSPEC"/*:refs/remotes/git/* &&
  		git push -v -v origin "refs/remotes/git/*:refs/remotes/git/*" \
  			>out 2>&1 &&
  		grep "POST git-receive-pack (chunked)" out
  	'
  fi

Thoughts?

-- 
Cheers,
Ray Chuan