optimization of list comprehensions

[Ulf Wiger (AL/EAB)]

In the face of Mats's convincing evidence and my 
own inability to calculate percentages, I must concede
the point, at least as regards AXD 301 & derivatives.

One should perhaps note that, at least for programmers
reared in the AXD 301 environment, funs were initially
slow and had a tendency to break during upgrade. Also,
using list comprehensions would be about twice as slow
as hand-coded iterator functions.

These problems have long since been fixed, but 
perceptions have a tendency to linger, especially in
a conservative setting, like doing maintenance on a 
"5-nines" system.

Regards,
Ulf W

> -----Original Message-----
> From: owner-erlang-questions@REDACTED 
> [mailto:owner-erlang-questions@REDACTED] On Behalf Of Mats Cronqvist
> Sent: den 9 mars 2006 15:01
> To: erlang-questions@REDACTED
> Subject: Re: optimization of list comprehensions
> 
> Richard A. O'Keefe wrote:
> 
>  > If you have evidence to back this up, it's publishable.
>  > Please publish it.  Some sort of survey analysing what 
> kind of  > programmers use what kind of language features 
> would be most illuminating.
> 
> Ulf Wiger (AL/EAB) wrote:
> 
> > I think it's safe to say that even "average" industrial programmers 
> > rather quickly learn to exploit the virtues of higher-order 
> functions 
> > and iterators.
> 
>    not to be outdone by wiger (and since o'keefe said 
> please), i ran a few greps on 1,748,162 lines of erlang 
> sources (comments included). i estimate that some 3-400 
> people have contributed to the code sample.
>    i counted how many times each of the patterns below 
> appeared per module.
> 
>          "[ (]fun[ (]"  "lists:fold" "lists:map" "lists:foreach"
> never          64%            89%          84%          85%
> <3 times       81%            97%          94%          94%
> 
>    36% of the modules defined at least one fun. however, 15% 
> of all modules had at least one call to mnesia:transaction. 
> so i feel pretty confident stating that
>   ~80% of all modules have no (non-mnesia related) funs.
>    another observation is that although there were ~2 fun 
> definition per module, 4% of the modules accounts for 50% of 
> the fun definitions.
>    one possible interpretation is that most of this code was 
> produced by programmers that rarely use funs if they don't 
> have to. obviously, the methodology is much too weak to 
> *prove* this. i'm not sure it can be proven, since there's no 
> way to establish who wrote what. most modules have been 
> touched by at least a handful of people, and everyone has 
> worked on more than one module.
>    note also that most of this code sample has been used for 
> years in a very challenging environment, and is part of one 
> of the most stable products in its market.
>    i does seem safe to say that one can write well working 
> Erlang code while rarely, if ever, using funs (excluding 
> mnesia use, as always).
> 
>    mats
>