Almost daily engineers write, mainly from scratch, ad hoc programs suitable
to solve very specific problems. In order to provide assistance to enginee
rs willing to develop their programs through white-bon reuse, it is mandato
ry to offer them repositories of software components populated with stand-a
lone routines selected according to software metrics able to measure the co
de complexity. Our selection approach is inspired to that previously adopte
d by Caldiera and Basili in a well-known paper, indeed, we use the same cla
ssical metrics, namely: volume, cyclomatic complexity and regularity. In es
sence, this paper represents both a refinement and a replication of the exp
eriment performed by Caldiera and Basili, being the domain of the software
investigated and the structure of the software itself the two major differe
nces. In order to carry out the experiment, it was necessary to adapt the e
xtremes of the values of the metrics mentioned above to our reference conte
xt which is very different from that Caldiera and Basili referred to. The a
daptation was not trivial because in the case of routines being part of sof
tware libraries it is not possible to proceed as they did. In this paper, a
general strategy suitable to "calibrate" their original reusability model
is given. To validate the proposal, the strategy is instantiated to the dom
ain of mathematical software. By referring to the calibrated reusability mo
del, we carried out a large scale empirical investigation in order to selec
t candidate reusable stand-alone (FORTRAN) routines from two commercial lib
raries belonging to the mathematical domain totally adding up to 2500 routi
nes. The second part of the paper reports about such an experiment. (C) 200
0 Elsevier Science Inc. All rights reserved.