Re: [PATCH 07/14] parse-options: allow ll_callback with OPTION_CALLBACK

From: Derrick Stolee <stolee@gmail.com>
Date: Apr 16, 2019, 14:24 UTC
Message-ID: <6b731f57-eada-8c1c-8e5d-de386b0c414e@gmail.com>
In-Reply-To: <CACsJy8DDKuK5VYhh0GNYSJK1_y3MZgK5Vcq99N4jYcusVFnvQQ@mail.gmail.com>

On 4/16/2019 4:52 AM, Duy Nguyen wrote:

Show 23 quoted lines

> On Mon, Apr 15, 2019 at 9:06 PM Derrick Stolee <stolee@gmail.com> wrote:
>>
>> On 1/26/2019 7:35 PM, Nguyễn Thái Ngọc Duy wrote:
>>> @@ -238,7 +249,10 @@ static enum parse_opt_result parse_short_opt(struct parse_opt_ctx_t *p,
>>>                       len++;
>>>               arg = xmemdupz(p->opt, len);
>>>               p->opt = p->opt[len] ? p->opt + len : NULL;
>>> -             rc = (*numopt->callback)(numopt, arg, 0) ? (-1) : 0;
>>> +             if (numopt->callback)
>>> +                     rc = (*numopt->callback)(numopt, arg, 0) ? (-1) : 0;
>>> +             else
>>> +                     rc = (*numopt->ll_callback)(p, numopt, arg, 0);
>>>               free(arg);
>>>               return rc;
>>>       }
>>
>> Hi Duy,
>>
>> This "else" condition is unreachable. This block is only hit when we have a "-<n>"
>> option, using OPT_NUMBER_CALLBACK, which is implemented by filling "callback", never
>> "ll_callback".
> 
> It does not mean ll_callback cannot be used in the future though.

That's not a very good reason to add it now. YAGNI.

Show 14 quoted lines

> We
> have three options
> 
> 1. drop the else clause
> 2. replace with "else BUG();"
> 3. implement proper else clause
> 
> Option #1 to me sounds wrong. If you don't support something, yell up.
> Silently ignoring it only makes it harder to track down to this
> unsupported location when it becomes reachable, however unlikely that
> is.
> 
> Which leaves options #2 and #3. If you think this one line is risky
> enough, I'll send a patch to replace it with BUG().

It's not about risk, but the fact that it is pointless. The only way to get to this block is to create a 'struct option' with type OPTION_NUMBER manually (ignoring the OPT_NUMBER_CALLBACK macro), which _should_ be unsupported. If someone goes to the pain of adding a way to instantiate with a low-level callback, then they should add this 'if' statement.

If you are going to add protection from an incorrect instantiation of 'callback' in a use of OPT_NUMBER_CALLBACK, then the proper place to do that is probably in parse_options_check(), where you were already doing callback-vs-ll_callback checks in the case of OPTION_CALLBACK and OPTION_LOWLEVEL_CALLBACK.

However, the only places where the OPT_NUMBER_CALLBACK option appears are builtin/grep.c and t/helper/test-parse-options.c. I don't imagine a need to add low-level callbacks any time soon.

>> I recommend reverting this diff segment, but please let me know if I'm missing something.

I still think this is the easiest way to remove the dead code.

I'll try to submit a patch later. I'm still not in full work mode, so I may be slow.

Thanks, -Stolee