Re: [PATCH 1/2] refspec: consolidate ref-prefix generation logic

From: Jonathan Nieder <jrnieder@gmail.com>
Date: May 31, 2018, 00:43 UTC
Message-ID: <20180531004309.GA36515@aiede.svl.corp.google.com>
In-Reply-To: <20180516234822.182663-2-bmwill@google.com>

Hi,

Brandon Williams wrote:

Show 14 quoted lines

> When using protocol v2 a client constructs a list of ref-prefixes which
> are sent across the wire so that the server can do server-side filtering
> of the ref-advertisement.  The logic that does this exists for both
> fetch and push (even though no push support for v2 currently exists yet)
> and is roughly the same so lets consolidate this logic and make it
> general enough that it can be used for both the push and fetch cases.
>
> Signed-off-by: Brandon Williams <bmwill@google.com>
> ---
>  builtin/fetch.c | 13 +------------
>  refspec.c       | 29 +++++++++++++++++++++++++++++
>  refspec.h       |  4 ++++
>  transport.c     | 21 +--------------------
>  4 files changed, 35 insertions(+), 32 deletions(-)

I assume this is meant to be a refactoring with no functional change. Alas, it's causing fetch-by-SHA-1 to fail for me:

 $ rm -fr /tmp/r
 $ git init /tmp/r
 Initialized empty Git repository in /tmp/r/.git/
 $ bin-wrappers/git -C /tmp/r -c protocol.version=2 fetch \
	https://kernel.googlesource.com/pub/scm/git/git \
	6373cb598e1a4e0340583ad75d5abba01ff79774
 fatal: no matching remote head

The parent commit succeeds.

Known issue?

Thanks, Jonathan