[Haskell-cafe] Is bumping the version number evil, if it's not mandated by the PVP?

[wren ng thornton]

Sebastian Fischer wrote:
> Hello,
> 
> I wonder whether (and how) I should increase the version number of a 
> library when the API does not change but the implementation gets more 
> efficient.

I think it depends on how much more efficient. Constant factors of 
improvement are always nice, but they aren't (usually) earth shattering 
and so are probably only worth a D bump. Asymptotic improvements may 
very well be worth a C or B bump--- especially if they introduce the 
possibility of new bugs due to large code changes (i.e., you haven't 
_proven_ bug-for-bug compatibility).

But it also depends on the audience. For example, some libraries are 
defined as being "an efficient implementation of Foo" rather than being, 
say, an implementation of a particular data type or a particular 
high-level API for writing nice code. If your library is _defined_ by 
its performance characteristics, then a C or B bump would be appropriate 
since the complexity is effectively part of the API even though it isn't 
captured in the type system.

-- 
Live well,
~wren