arbtt feedback

[Gwern Branwen]

On Sun, Sep 16, 2012 at 6:20 AM, Joachim Breitner
<mail at joachim-breitner.de> wrote:
> that’s ok; but I’d like to have such discussion archived. Let me
> rephrase the question: Do you mind if I forward (i.e. redirect) this
> thread to the list as well? You do not have to subscribe.

I don't mind.

> GHC in testing is at 7.4. But 0.6.2-1 should have been compiled with
> that.

/shrug

When I run update && upgrade, GHC is in the kept back list so far.

> I looked at the code again and the problem is that it currently keeps
> track of the whole list for several reasons:
>  * There can be multiple report being processed.
>  * Some global data (i.e. total number of records) is calculated first
> and then shared between possibly multiple report passes.
>  * Some reports also refer to the non-selected time (“% of total time
> selected”). So making arbtt-stats O(1) is a non-trivial refactoring.

Why would either of the latter two force retention of the full
dataset? An integer counter like 'total number of records' should be
computable without retaining the list, and likewise integer counters
for selected or non-selected time.

> You need time >= 1.4 (which should come with GHC 7.4); I fixed the cabal
> file accordingly.

Well, I'll try again when Debian decides it's safe to upgrade. Until
then, it's not *that* much of a priority - I'll only really need the
arbtt data in March when my big experiment concludes.

-- 
gwern
http://www.gwern.net