On 08/16/2013 11:06 AM, Les Mikesell wrote: > Exactly my point. Everything is about derived works. So binaries > cannot be exempt from the requirement that the work as a whole can > only be distributed under a license that permits free redistribution > and that additional restrictions cannot be added. If you want to > refute that, please quote the section stating what you think permits it. Les, binaries aren't derived works. They're machine-generated translations. A derived work would be a change in the source code; binaries are direct machine-readable translations of unmodified source code. And the GPL covers just the programs on the distribution that are, well, covered by the GPL at the source level. Mere aggregation doesn't mean the whole iso is under the GPL, only the binaries that are compiled from GPL source are. The copyright for the collection may prohibit distribution of the collection (in its aggregated form), but you might be able to distribute those individual binaries that are built from GPL sources; but you would violate your subscription agreement (a separate legal agreement and not part of the copyright license) if you did so. After all, the licensor of the GPL-covered program is in many cases not Red Hat; the subscription agreement is a contract with Red Hat and Red Hat alone. The GPL is all about source code availability, not binary availability. To wit, see this section in the GPL FAQ: https://www.gnu.org/licenses/gpl-faq.html#MustSourceBuildToMatchExactHashOfBinary And even https://www.gnu.org/licenses/gpl-faq.html#ExportWarranties applies, as ITAR would represent a 'restriction' on distribution, no? But again the GPL coverage doesn't extend to the aggregation in ISO form, only to the individual programs on the ISO. Nothing in the GPL says that if you distribute the source to the public you must distribute binaries to the public; all it says is that if you distribute binaries you must distribute or include a written offer to distribute the source to the people to whom you have distributed binaries. This is how SuSE (to use Johnny's example cross-thread) gets away with not having public distribution of the sources for SLES (if you find the publicly available sources for SLES with updates please let me know, and OpenSuSE is not the same thing).