On 12 November 2010 11:14, Simon Marlow
Right, my main objection has always been that since GHC makes it impractical to do the modify/relink that the LGPL requires, LGPL Haskell code isn't really complying with the spirit of the LGPL here.
Right, the letter but not the spirit. Of course the technical situation with ABIs and GHC may improve in future as you've often talked about.
I get the impression that people who use the LGPL for Haskell don't care deeply about this - they're using the LGPL because they want modifications contributed back. In which case, wouldn't it make sense to add an exception for the modify/relink too? Are there Haskell authors out there using the LGPL who really do care about modify/relink?
Two points: Firstly, supposing for the sake of argument we switched tomorrow to using shared libs by default, would people consider the problem to have gone away? We have exactly the same ABI issues with shared libs as with static ones. Do we still need an exception to apologise that our shared lib implementation is not really a "suitable shared library mechanism": Quoth the LGPL-2.1 A suitable mechanism is one that (1) uses at run time a copy of the library already present on the user's computer system, rather than copying library functions into the executable, and (2) will operate properly with a modified version of the library, if the user installs one, as long as the modified version is interface-compatible with the version that the work was made with. Again, we comply with the letter but not the spirit of the license, because it's pretty clear this has to mean ABI compatible (even in C). Secondly, yes I agree that people tend not to care too deeply about this. To play devil's advocate, if people don't really care then instead of saying that means we must have an exception, perhaps that means that we're ok to comply with the letter of the law but not the spirit. It is a big hassle to work out wording for an exception and get all license holders of each project to agree (I did it once with Gtk2Hs and LGPL-2.0->2.1). If people do not actually care, why bother? And again about ABIs, other compilers do have stable ABIs (e.g. nhc) and there's the future possibility to do the same with GHC (for people who care to define a stable ABI and follow the rules) and in this case it should work just as well with shared libs as with static relinking. Duncan