Re: [patch] Make vector::at() assertion message more useful

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

On Sun, Aug 18, 2013 at 5:38 AM, Chris Jefferson <chris@bubblescope.net> wrote:
> On 18/08/13 08:09, Václav Zeman wrote:
>>
>> On 08/17/2013 05:10 PM, Gabriel Dos Reis wrote:
>>>
>>> On Sat, Aug 17, 2013 at 5:41 AM, Paolo Carlini <paolo.carlini@oracle.com>
>>> wrote:
>>>>
>>>>
>>>> Hi,
>>>>
>>>> Paul Pluzhnikov <ppluzhnikov@google.com> ha scritto:
>>>> __throw_out_of_range(__N("vector::_M_range_check"));
>>>>>
>>>>> +        {
>>>>> +          char __s[256];
>>>>> +          __builtin_snprintf(__s, sizeof(__s),
>>>>> +                             __N("vector::_M_range_check: %zu >=
>>>>> %zu"),
>>>>> +                             __n, this->size());
>>>>> +          __throw_out_of_range(__s);
>>>>> +        }
>>>>
>>>> The idea makes sense, but while we are at it I think the message could
>>>> be more clear, say what the two numbers are. Also, I don't think we can
>>>> unconditionally call snprintf, it's C99 and supported targets lack it. Maybe
>>>> with some libiberty magic? I don't think such magic automagically triggers
>>>> when __builtin_snprintf is expanded, or does it? Please investigate that.
>>>
>>> I agree that we can't embark a significant IO library as part of a purely
>>> data structure components.
>>>
>>> Furthermore, __builtin_snprintf doesn't always expand to a
>>> compiler magic -- it can be a normal function call.
>>>
>>> This would be a regression, not an improvement.
>>
>> What about adding a simple template function that converts integer into
>> std::string using division by 10? You could then concatenate the message
>> from std::string fragments, avoiding IO library but having to pull in
>> std::string instead. Would that be acceptable?
>
> Rather than making up new functions, why not use a std::ostringstream?

I have a hunch that we are entering an infinite loop.

-- Gaby