IANAL, TINLA, speaking only for myself blah blah blah

While looking at what the GPL says is important, the real issue is
what copyright law says - you only need permission from the copyright
holder to exercise rights protected by copyright law.

Copyright protects the expression of ideas, not the ideas themselves.
So I see no problem with understanding algorithms and data structures
and creating a implementation of those structures, without copying any
of the code.  If one essentially translated the code by rewriting it
in a different style but with almost identical structure, then you'd
have an interesting legal question.

If the code requires patent licenses to function, that's a separate
issue.  But if that's the case, GPL'd code cannot be distributed
anyway.

As far as ethics (standards of acceptabile conduct within a group,
v.s. morals which have more of a religious/personal connotation) go,
my impression is that the Free software community does not believe
that algorithms should be patentable, or that user interfaces should
be copyrightable, and does believe in the sort of "reading other'scode
for understanding/inspiration" you discuss.  So I don't see any
problem with this.

The following book (pdf online) is helpful in understanding these
issues:

http://www.rosenlaw.com/oslbook.htm

Although written from a primarily US perspective, copyright law is
fairly similar world wide, at least among signatories to the Berne
Convention.

http://www.law.cornell.edu/treaties/berne/overview.html

http://en.wikipedia.org/wiki/Berne_Convention_for_the_Protection_of_Literary_and_Artistic_Works


-- 
        Greg Troxel <gdt@ir.bbn.com>