Function points per lines of code

[Alex Martelli]

"Stefan Schwarzer" <s.schwarzer at ndh.net> wrote in message
news:3A35791C.5930A7FA at ndh.net...
> Hello,
>
> for an effort/schedule estimation I look for values of lines of code per
> function point for Python. I have searched Deja.com and Google.com but to
> no avail.

Starting from Lutz Prechelt's study, e.g. at
http://ncstrl.ubka.uni-karlsruhe.de:8080/Dienst/Repository/2.0/Body/ncstrl.u
bka_cs%2firatr-2000-5/text

one sees Capers Jones' table for a few languages:

language LL LOC/FP
C     3.5 91
C++   6   53
Java  6   53
Perl  15  21
Python ?  ?
Rexx   7  46
Tcl    5  64

and Prechelt sensibly assumes that the value for Tcl is
a typo (it has much higher language-level, and lower
LOC/FP, than, say, Java!).

The rest of Prechelt's study appears to strongly indicate
semantic equivalence (on his example problems) between
Python and Perl; so, taking Capers Jones' estimate as
good for the latter, it seems that around 20 LOCs/FP is
a reasonable approximation.

Having worked extensively in all of these languages, I
do NOT intuitively concur that C++ or Java are only twice
as powerful (in terms of LOC/FP) as C, nor that Rexx
takes more LOC/FP than them, and the estimate for Perl
(and Python) appear 'compressed downwards' on this scale,
too.  But then, I have NOT conducted actual counts and
measurements!  So, if my intuition tells me that, taking
C at about 100, C++ and Java must be closer to 30 (about
1:3 ratio) than to 50 (about 1:2) LOC/FP; that Rexx and
Tcl may be, oh, 25 (lower LOC/FP than C++ or Java); and
that Perl and Python must be definitely below 20 (15
sounds better); then this is just my own intuition and
experience.  Maybe it reflects the way *I* code in these
various languages, rather than the way they're 'meant'
to be used, for example.


Alex