[Dlang-internal] Regression control & breaking changes policies

[Walter Bright via Dlang-internal]

On 12/6/2016 5:56 PM, Dicebot wrote:
> It is harlmess by a simple criteria that I challenge you to give an
> example of harm it causes.

Let's say there's a bug report that XXX generates corrupt code (such as not 
initializing a variable). This may work most of the time, or appear to work, but 
exhibit an error only rarely. But when it does fail, it could cause much 
disruption to the user's business.

Suppose that a fix is created, but comes with it a regression that some unusual 
(but correct) code is now broken.

We have the benefit of fixing a silent potentially disastrous problem, while 
introducing a problem that may be a nuisance, but will at least be obvious to 
the user.

In other words, if the bug fixed is costlier than the bug introduced, then that 
is an example of net harm in reverting.

Almost all changes/fixes bring with them at least *some* difference that 
*somebody* may depend on, however remote the change and however we might roll 
our eyes that somebody managed to depend on it. :-)

(For example, somebody might be doing hashes on the object file generated.)