patchrefspec: allow @ on the left-hand side of refspecs

> The object ID parsing machinery is aware of "@" as a synonym for "HEAD"
> and this is documented accordingly in gitrevisions(7).  The push
> documentation describes the source portion of a refspec as "any
> arbitrary 'SHA-1 expression'"; however, "@" is not allowed on the
> left-hand side of a refspec, since we attempt to check for it being a
> valid ref name and fail (since it is not).
> 
> Teach the refspec machinery about this alias and silently substitute
> "HEAD" when we see "@".  This handles the fact that HEAD is a symref and
> preserves its special behavior.  We need not handle other arbitrary
> object ID expressions (such as "@^") when pushing because the revision
> machinery already handles that for us.

> 
> Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
> ---
> I probably type "git push upstream HEAD" from five to thirty times a
> day, many of those where I typo "HEAD", so I decided to implement the
> shorter form.  This design handles @ as HEAD in both fetch and push,
> whereas alternate solutions would not.

> 
> check_refname_format explicitly rejects "@"; I tried at first to simply
> ignore that with a flag, but we end up calling that from several other
> places in the codebase and rejecting it and all of those places would
> have needed updating.
> 
> I thought about putting the if/else logic in a function, but since it's
> just four lines, I decided not to.  However, if people think it would be
> tidier, I can do so.
> 
> Note that the test portion of the patch is best read with git diff -w;
> the current version is very noisy.
> 
>  refspec.c             |   6 ++-
>  t/t5516-fetch-push.sh | 104 +++++++++++++++++++++---------------------
>  2 files changed, 58 insertions(+), 52 deletions(-)
> 
> diff --git a/refspec.c b/refspec.c
> index e8010dce0c..57c2f65104 100644
> --- a/refspec.c
> +++ b/refspec.c
> @@ -62,8 +62,12 @@ static int parse_refspec(struct refspec_item *item, const char *refspec, int fet
>  		return 0;
>  	}
>  
> +	if (llen == 1 && lhs[0] == '@')
> +		item->src = xstrdup("HEAD");
> +	else
> +		item->src = xstrndup(lhs, llen);
> +

> On 07/29, brian m. carlson wrote:
> > The object ID parsing machinery is aware of "@" as a synonym for "HEAD"
> > and this is documented accordingly in gitrevisions(7).  The push
> > documentation describes the source portion of a refspec as "any
> > arbitrary 'SHA-1 expression'"; however, "@" is not allowed on the
> > left-hand side of a refspec, since we attempt to check for it being a
> > valid ref name and fail (since it is not).
> > 
> > Teach the refspec machinery about this alias and silently substitute
> > "HEAD" when we see "@".  This handles the fact that HEAD is a symref and
> > preserves its special behavior.  We need not handle other arbitrary
> > object ID expressions (such as "@^") when pushing because the revision
> > machinery already handles that for us.
> 
> So this claims that using "@^" should work despite not accounting for it
> explicitly or am I misreading?  Unless I'm mistaken, it looks like we
> don't really support arbitrary rev syntax in refspecs since "HEAD^"
> doesn't work either.

> On Mon, Jul 30, 2018 at 10:50:51AM -0700, Brandon Williams wrote:
> > On 07/29, brian m. carlson wrote:
> > > The object ID parsing machinery is aware of "@" as a synonym for "HEAD"
> > > and this is documented accordingly in gitrevisions(7).  The push
> > > documentation describes the source portion of a refspec as "any
> > > arbitrary 'SHA-1 expression'"; however, "@" is not allowed on the
> > > left-hand side of a refspec, since we attempt to check for it being a
> > > valid ref name and fail (since it is not).
> > > 
> > > Teach the refspec machinery about this alias and silently substitute
> > > "HEAD" when we see "@".  This handles the fact that HEAD is a symref and
> > > preserves its special behavior.  We need not handle other arbitrary
> > > object ID expressions (such as "@^") when pushing because the revision
> > > machinery already handles that for us.
> > 
> > So this claims that using "@^" should work despite not accounting for it
> > explicitly or am I misreading?  Unless I'm mistaken, it looks like we
> > don't really support arbitrary rev syntax in refspecs since "HEAD^"
> > doesn't work either.
> 
> Correct, it does indeed work, at least for me:
> 
> genre ok % git push castro HEAD^:refs/heads/temp
> Total 0 (delta 0), reused 0 (delta 0)
> To https://git.crustytoothpaste.net/git/bmc/git.git
>  * [new branch]            HEAD^ -> temp
> 
> genre ok % git push castro @^:refs/heads/temp
> Total 0 (delta 0), reused 0 (delta 0)
> To https://git.crustytoothpaste.net/git/bmc/git.git
>  * [new branch]            @^ -> temp
> 
> Note that in this case, I had to specify a full ref since it didn't
> exist on the remote and the left side wasn't a ref name.

> 
> Now it doesn't work for fetches, only pushes.  Only the left side of a
> push refspec can be an arbitrary expression.
> -- 
> brian m. carlson: Houston, Texas, US
> OpenPGP: https://keybase.io/bk2204