Comments on: Yet Another Reason Not To Be Lazy Or Imperative

By: Carter Schonwald

Carter Schonwald — Mon, 27 Aug 2012 21:23:06 +0000

Hey Bob! (nifty paper)
Nifty! I actually had a soft pen and paper proof about that sort of well defined GC traversal order enabling good memory locality a few months ago. Its a really *easy* (super scare quotes) property to prove informally :), and at the time I was really shocked that theres not been any work over that insight. Got a bit side tracked so I never finished formalizing it :). I’m glad someones up and done this.

Either way, its a grand first step in really making reasoning about how to get optimal memory locality while writing high level code tractable.

I’m having a bit of fun presently building up a bit of intuition in this space as I incremental write some high performance functional programming style tools for numerical computation. Its an area in desperate need of good PL/FP love, and I’m trying do my part.

-Carter

By: Robert Harper

Robert Harper — Mon, 27 Aug 2012 19:57:49 +0000

one could say “provably efficient implementation”, which would be appropriately descriptive, but it’s a bit of a mouthful too (and i just hate having to qualify with “provably” to basically say “we’re serious about this, not some jokers with histograms.”

By: Robert Harper

Robert Harper — Mon, 27 Aug 2012 19:49:37 +0000

it doesn’t exactly slide off the tongue, but it’s definitely better grammar.

By: chrisdone

chrisdone — Mon, 27 Aug 2012 19:34:23 +0000

s/one might reasonable/one might reasonably

By: Mike Shulman

Mike Shulman — Mon, 27 Aug 2012 17:19:54 +0000

I suppose “provably correct implementation” is too many syllables?

By: Robert Harper

Robert Harper — Mon, 27 Aug 2012 12:13:02 +0000

Oh, don’t get me started on this. I’ve reluctantly adhered to the pre-existing terminology, despite reservations. If you or anyone can think of a better name, I’m eager to learn it. Admittedly “proved implementation” is more accurate than “provable implementation”, but it’s not much of an improvement. I find it ridiculous that we have to emphasize that we can actually prove the validity of our claims, in contrast to standard practice of, at most, showing a few well-chosen histograms and declaring victory.

By: andrejbauer

andrejbauer — Mon, 27 Aug 2012 05:25:11 +0000

This is a sociological remark. You PL types always say things like “provable implementation”. Nobody in logic would every say that. I suspect you mean “implementation which was proved correct by cool people”. I can live with the contraction “proved implementation”, but not with “provable” in place of “proved”. So what do you mean when you say “provable X” with “X” a noun?

By: Andy Adams-Moran

Andy Adams-Moran — Mon, 27 Aug 2012 05:04:27 +0000

Is “Improvement in a lazy context: An operational theory for call-by-need”, http://dl.acm.org/citation.cfm?id=292547 relevant here? We were using cost semantics to give a theory to call-by-need (and so we tracked lookup/update in the heap instead of I/O reads/writes), but many of the same contextual recursion principles should hold (a context lemma being key to our approach).