From: Greg Brockman <gdb@mit.edu>
Date: Wed, 21 Jul 2010 15:11:10 GMT
Subject: Re: [PATCH 1/4] Allow creation of arbitrary git-shell commands
Message-ID: <AANLkTiktvRmx-2iSVCTvrDYrAAw0kGCtsBhb1mg1ULbx@mail.gmail.com>
In-Reply-To: <7vr5iym6jj.fsf@alter.siamese.dyndns.org>

>> +                     user_argv[0] = prog;
>> +                     execv(user_argv[0], (char *const *) user_argv);
>> +                     free(prog);
>> +             }
>> +             free(user_argv);
>> +             /*
>> +              * split_cmdline modifies its argument in-place, so 'prog' now
>> +              * holds the actual command name
>> +              */
>> +             die("unrecognized command '%s'", prog_cpy);
>> +     } else {
>> +             /*
>> +              * split_cmdline has clobbered prog and printed an
>> +              * error message, so print the original
>> +              */
>> +             die("invalid command format '%s'", prog_cpy);
>
> Hmm, we might want to fix split_cmdline to report the breakage better
> then (perhaps not as part of this patch series).  Instead of seeing
>
>    error: cmdline ends with \.
>    fatal: invalid command format 'foo bar \'.
>
> wouldn't it be nicer to see a single error message:
>
>    fatal: invalid command 'foo bar\': cmdline ends with \.
>
That would definitely be preferable.  If possible, I would rather not
do it as part of this patch series, but would be happy to work on
fixing split_cmdline afterwards.