Re: [Boost-users] boost::iequals bug in 1.35?

From: Pavol Droba (droba_at_[hidden])
Date: 2008-04-03 07:50:56

Thorsten Ottosen wrote:
> Pavol Droba skrev:
>> Thorsten Ottosen wrote:
>>> Pavol Droba skrev:
>>>> Hi,
>>>>
>>>> char* is not longer supported as a default range type by the range library.
>>>> you can use const char* literals only.
>>>>
>>>> You can pass char* only if you convert it to a range using as_literal or as_array
>>>> helpers, that are part of the range library.
>>>>
>>>> We have to change this since thare were ambiguieties in char* usage.
>>>>
>>>> Best regards,
>>>> Pavol.
>>>>
>>>> L Pocaille wrote:
>>>>> The following code was working in boost 1.34
>>>>>
>>>>>    char smurf_name[512];
>>>>>    strcpy(smurf_name, "Schtroumpf");
>>>>>    bool samething= boost::iequals("Schtroumpf", smurf_name);
>>>>>    assert(samething);
>>>>>
>>>>> but it no longer works in 1.35. 
>>> Pavol,
>>>
>>> why don't you use as_literal() internally in the string library?
>>> Would that not preserve the behavior?
>>>
>> Actualy I do. But it seems, that this way it works only with const-chars .
> 
> Oh? On every compiler?
> 
> Looking at the example I'me a bit surprised that this have ever worked 
> *unless smurf_name was passed by value to force a decay*. AFAICT, the 
> range returned by as_literal(smurf_name) will be very long because we 
> don't scan arrays, only pointers, for a null.
> 

Ooops, I was supposed to fix this, but it seems, that I forgot ;(

> I remember some problems with this earlier. But I'm suprprised that our 
> regression of the string algorithms library would not have cought this.

There used to be tests for char* arguments, but since they were deemed as invalid usage,
I have removed them from the test suite.

Pavol.