Re: [PATCH v3 00/20] refs backend rebase on pu

[David Turner <dturner@twopensource.com>]

On Thu, 2016-02-04 at 20:39 +0000, Ramsay Jones wrote:
> 
> On 04/02/16 20:25, David Turner wrote:
> > On Thu, 2016-02-04 at 18:42 +0700, Duy Nguyen wrote:
> > > On Wed, Feb 3, 2016 at 3:08 AM, David Turner <
> > > dturner@twopensource.com> wrote:
> > > > Are there any more reviews on this?  I do have some changes
> > > > from
> > > > this
> > > > set, but they're pretty minor so I don't want to post a new one
> > > > (unless
> > > > folks would rather see those changes before reviewing).  Let me
> > > > know.
> > > 
> > > Last note. Since this is new code, maybe you can wrap
> > > translatable
> > > strings with _(), basically any visible string that machines do
> > > not
> > > need to recognize.
> > 
> > Fixed, thanks. 
> 
> Another minor point, could you please squash this in:
> 
> diff --git a/refs.c b/refs.c
> index 3d4c0a6..4858d94 100644
> --- a/refs.c
> +++ b/refs.c
> @@ -17,7 +17,6 @@ static const char split_transaction_fail_warning[]
> =
>  /*
>   * We always have a files backend and it is the default.
>   */
> -struct ref_storage_be refs_be_files;
>  static struct ref_storage_be *the_refs_backend = &refs_be_files;
>  /*
>   * List of all available backends
> 
> The above (on Linux, anyway) is a 'tentative definition' of the
> refs_be_files variable and so a common symbol definition is issued
> in refs.o. This then gets 'combined' with the *actual* symbol
> definition in  refs/files-backend.c. So everything 'works', except
> that I have used some unix (let alone non-unix) systems which would
> not output a common symbol for the above and would fail to link
> with a 'multiple symbol definition' error.
> 
> [Also note that an external declaration is already in effect from
> the refs/refs-internal.h header file! ;-) ]
> 
> ATB,
> Ramsay Jones

Fixed, thanks.