Let's talk about deprecations

[Cym13 via Digitalmars-d]

On Monday, 3 July 2017 at 16:08:43 UTC, bachmeier wrote:
> On Monday, 3 July 2017 at 13:14:02 UTC, Jonathan M Davis wrote:
>
>> So, in general, I think that it's a big mistake to keep 
>> deprecated stuff along on a permanent or semi-permanent basis. 
>> Keeping it around for about two years like we do now is 
>> already quite a long time in that regard. As long as the 
>> symbol doesn't require any maintenance, then it's not a big 
>> deal, but too often, _some_ maintenance does end up being 
>> required.
>
>
> For Phobos, changing the name would be less about keeping the 
> function around for a long time than about offering an easy fix 
> once the code does break. The current approach to deprecations 
> is that you get a warning that at some point the function might 
> be removed, then you update the compiler and you have broken 
> code without an obvious fix. A broken build gets your 
> attention, but you have a short-term fix. The total length of 
> the deprecation cycle wouldn't have to change.

I don't get it: how is linking old and unmaintained code "fixing" 
anything? Dead code is dead, if it's been two years already let 
it die and fix your own, anything else can't be future-proof.