[Nix-dev] Haskell NG

[roconnor at theorem.ca]

On Thu, 18 Dec 2014, Peter Simons wrote:

> Hi Russel,
>
> >> 2) Haskell packages support 'deepOverride'.
> >
> > How would you feel about using my overrideScope functionality from my
> > haskellPackagesFixpoint branch instead?
>
> the first version of my re-factored Haskell code defined deep overrides exactly
> the way you did it. Check out the "definePackage" function in [1]. You'll find
> that I studied your work thoroughly.
>
> My impressions are:
>
> 1. overrideScope and deepOverride achieve the same goal. I discovered no case
>    where one function worked but the other one didn't.
>
> 2. deepOverride is a standard function every callPackage derivation supports.
>    overrideScope, on the other hand, is not.
>
> 3. deepOverride is implemented in a simple generic algorithm (for appropriate
>    definitions of the term "simple") that requires no special support from the
>    haskellPackages record to function. overrideScope, on the other hand, needs
>    the magic nixClass attribute (__unfix__ in my code) to build a stack of
>    extension functions.
>
> Ultimately, I felt that I should stick to the standard deepOverride mechanism
> mostly due to Occam's razor'ish reasoning: I'd rather not introduce a new
> sophisticated tying-the-knot mechanism to haskellPackages just to do the same
> thing that deepOverride can do already.
>
> Does that make sense?

That is fair reasoning.

I will add that I expect overrideScope to run exponentially faster than 
deepOverride in worst case.  A similar change to replace-dependency took 
the evaluation time for replacing bash in a NixOS system from 65 minutes 
to instantaneous <https://githum.com/NixOS/nixpkgs/pull/4313>.  Replacing 
bash in a NixOS system is probably *the* worst case scenario because bash 
so deep in the dependency graph.  Until deepOverride starts taking 
appreciable time, the efficencies of overrideScope is probably not yet a 
pursuasive argument.

So please keep an eye out for long evaluation times when using 
deepOverride as you continue your work. :)  Haskell applications have 
deeper and more fine grained dependencies than regular applications so 
you might be more likely to run into evaluation time problems than others.

What I really need to do is convince Eelco Dolstra and Michael Raskin that 
semantics of overriding packages and semantics of object oriented 
programmed coincide and they should adopt my oop library for implementing 
the overriding mechanism.

I'm trying to think of a way of selling my oop library to them that is a 
little more light weight that flying to europe to give them a presenation. 
Maybe I should use the Ocaml package set as a testbed to demonstrate 
overrideScope.

-- 
Russell O'Connor                                      <http://r6.ca/>
``All talk about `theft,''' the general counsel of the American Graphophone
Company wrote, ``is the merest claptrap, for there exists no property in
ideas musical, literary or artistic, except as defined by statute.''