Florian:
On most systems, readline is GPLed. There is a non-copyleft reimplementation somewhere, but I don't think it's widely used.
The non-copyleft implementation is an upgrade from BSD's lineedit, called libedit http://www.thrysoee.dk/editline/. Apple's OS X includes it as a version of libreadline by a adding "/usr/lib/ libreadline.dylib" as a symbolic link to libedit.dylib. (I had forgotten about this because I have a version of Darwin Ports's libreadline I had installed so I could build GHC on OS 10.4: libedit does not fully implement libreadline's API so GHC requires libreadline). That GHC includes libreadline is not a problem for developers compiling programs using GHC because libreadline is used for GHC's interactive bytecode interpreter, GHCi, which is part of the compiler system but is not linked into the RTS of compiled programs. (This might be a problem with GHC's BSD license; I don't know.) Clearly the best solution would be to write to the FSF. The whole purpose of licensing GHC as BSD3 is to promote the broad use of Haskell as a programming language and developer involvement in the support, maintenance and research of Haskell and the GHC compiler. Those goals realistically require commercial developer support, otherwise small company or individual developers who lack the market recognition to sell software as a service could not write applications in Haskell and compile them with GHC. There are a few companies that do profitably sell Haskell as a service, notably Galois Connections, Aetion Technologies, Bluespec, Erlang Training and Consulting and OM Consulting.
the "no further restriction" clause is so broad
It has to be very broad, otherwise developers could bypass the copyleft concept.
True.
the same clause (verbatim) in section 10 of the LGPL means the GPL license is incompatible with the terms of the LGPL!
The LGPL permits relicensing of covered works under the GPL. Without that provision, it would indeed be incompatible with the GPL.
Are you referring to section 3 of the LGPL? The drafters should have referenced section 10 there to avoid confusion: section 3 and section 10 are clearly conflicting provisions with the only indicator of precedence being the order (section 3 comes before 10). I should note that this would not be a problem for licenses assigned directly by the FSF since they could refer to other documentation that would clarify the discrepancy. It might be a problem for an individual programmer who simply adopted the LGPL license and then had to enforce it. Best regards, Peter