Re: [boost] [range][foreach] Range fixes broke foreach

From: Eric Niebler (eric_at_[hidden])
Date: 2007-10-24 11:13:24

Thorsten Ottosen wrote:
> Eric Niebler skrev:
>> Thorsten Ottosen wrote:
>>> Beman Dawes skrev:
>>>> Looks like the range fixes broke foreach. See 
>>>> http://mysite.verizon.net/beman/bgd-win32-trunk-results.html
>>> This is my fault, sinc eI changed the names of the ADL hooks back to 
>>> 1.34 names:
>>>
>>> namespace boost
>>> {
>>>      char * range_begin(mine::dummy&) {return 0;}
>>>      char const * range_begin(mine::dummy const&) {return 0;}
>>>      char * range_end(mine::dummy&) {return 0;}
>>>      char const * range_end(mine::dummy const&) {return 0;}
>>> }
>>>
>>> should be
>>>
>>> namespace boost
>>> {
>>>      char * boost_range_begin(mine::dummy&) {return 0;}
>>>      char const * boost_range_begin(mine::dummy const&) {return 0;}
>>>      char * boost_range_end(mine::dummy&) {return 0;}
>>>      char const * boost_range_end(mine::dummy const&) {return 0;}
>>> }
>>>
>>> The old names have been in use for so long that I don't see the point in 
>>> changing them anymore.
>>
>> I'm dumbfounded. Thorsten, how can you make an API-breaking change to a 
>> core library like Range at a time like this, and not tell anybody? 
> 
> I could have announced this better, I apologize for that. What time is this?

We're stabilizing for 1.35.

>> My 
>> code, docs and tests would all need to change because of this.
>>
>> !!!!!
>>
>> We agreed long ago that range_begin() was the name of the customization 
>> point, because the Range concept is larger than Boost. Has that changed?
> 
> And that is the problem: do you want to break the few boost libs in a 
> development trunk, or break all clients out there? The names where 
> settled for maybe two years ago.

Have you just changed the names of the adl hooks? There are lots of 
other breaking changes, like the changed meaning of range_iterator<>, 
the deprecation of range_result_iterator<> and the addition of 
range_mutable_iterator<>. And the behavior of range_end(char (&)[N]), 
too. Are you going to roll all of that back? If not, then you haven't 
really got a leg to stand on -- you're breaking clients anyway.

This change is a mistake. Please, back it out.

-- 
Eric Niebler
Boost Consulting
www.boost-consulting.com