Re: [PATCH] clone: do not segfault when specifying a nonexistent branch

From: Stefan Beller <stefanbeller@googlemail.com>
Date: Oct 6, 2013, 09:27 UTC
Message-ID: <52512CF8.2070205@googlemail.com>
In-Reply-To: <CACsJy8BX_fWdsCGa4jnh4CbkSMxp7btOFjwzB9K0eRtjUR_F-Q@mail.gmail.com>

On 10/05/2013 01:55 AM, Duy Nguyen wrote:

Show 8 quoted lines

> On Fri, Oct 4, 2013 at 9:20 PM, Stefan Beller
> <stefanbeller@googlemail.com> wrote:
>> I think we should emit a warning additionally?
>>
>> Signed-off-by: Stefan Beller <stefanbeller@googlemail.com>
> 
> I think it's nice to credit Robert for reporting the fault in the
> commit message (something like "reported-by:" or "noticed-by:"...)

I'll do so in a resend.

Show 22 quoted lines

> 
>> ---
>>  builtin/clone.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/builtin/clone.c b/builtin/clone.c
>> index 0aff974..b764ad0 100644
>> --- a/builtin/clone.c
>> +++ b/builtin/clone.c
>> @@ -688,7 +688,7 @@ static void write_refspec_config(const char* src_ref_prefix,
>>
>>         if (option_mirror || !option_bare) {
>>                 if (option_single_branch && !option_mirror) {
>> -                       if (option_branch) {
>> +                       if (option_branch && our_head_points_at) {
>>                                 if (strstr(our_head_points_at->name, "refs/tags/"))
>>                                         strbuf_addf(&value, "+%s:%s", our_head_points_at->name,
>>                                                 our_head_points_at->name);
> 
> This prevents the segfault, but what about remote.*.fetch? Should we
> setup standard refspec for fetch or..?
>

Looking at the code a few lines below, this comment comes up:

	/*
	 * otherwise, the next "git fetch" will
	 * simply fetch from HEAD without updating
	 * any remote-tracking branch, which is what
	 * we want.
	 */

This behavior was good for the case (!option_branch && !remote_head_points_at)
Now we extend that behavior doing nothing to
	 ((!option_branch || !our_head_points_at) &&  !remote_head_points_at)

I am not sure how to handle that case best. The user has given a non existing branch, so it doesn't make sense to track that branch, but only have that registered as a remote*.fetch?

Reading the documentation enhancements of 31b808a (2012-09-20, clone --single: limit the fetch refspec to fetched branch), doesn't talk about this corner case. So maybe the remote.*.fetch shall be set, but no branch should be checked out, when running git clone --depth 1 -b test https://github.com/mitfik/coredump.git /tmp/coredump.git

Does that make sense?

Stefan