Re: [PATCH] describe: fall back to 'HEAD' if no appropriate tag exists

[Andy Whitcroft <apw@shadowen.org>]

Junio C Hamano wrote:
> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
> 
>> Now, if no tag exists to say something like '<tag>-gfffffff', say
>> 'HEAD-gfffffff' instead of erroring out.
>>
>> Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
>> ---
>>  describe.c |    2 +-
>>  1 files changed, 1 insertions(+), 1 deletions(-)
>>
>> diff --git a/describe.c b/describe.c
>> index ab192f8..8b08a3f 100644
>> --- a/describe.c
>> +++ b/describe.c
>> @@ -136,7 +136,7 @@ static void describe(const char *arg, in
>>  			return;
>>  		}
>>  	}
>> -	die("cannot describe '%s'", sha1_to_hex(cmit->object.sha1));
>> +	printf("HEAD-g%s\n", find_unique_abbrev(cmit->object.sha1, abbrev));
>>  }
> 
> Hmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm.
> 
> At least with tags, the user can assume NAME-gREV means commit
> whose abbrev is REV that is descendant of NAME.  HEAD is not
> necessarily so.
> 
> Having said that, in order to avoid barfing, we have to have
> something there, and HEAD is already special in many aspects
> anyway (e.g. by only saying HEAD you cannot tell which branch's
> tip you are talking about), it might be good enough.
> 
> I am just wondering if there is some other obvious substitute
> that is better than HEAD.  "GIT-g%s" is not it ("g" already
> stands for GIT).
> 
> Another possibility is just to do
> 
>         puts(sha1_to_hex(cmit->object.sha1))
> 
> in this case.  I tend to like that better somehow; it makes
> things more explicit.

Heh, we've just been putting together 'versioner' for some software and
when git-describe fails we substitute in the complete sha1 at that
point.  Based mostly on the "who'd release something versioned without
tagging anyhow".  I think either the full sha1 or _just_ the gNNNNN
without prefix would be fine.

-apw