Hi, On Saturday, 2008 June 28 at 11:54:14 +0200, Pablo Neira Ayuso wrote: > Eric Leblond wrote: > > On Friday, 2008 June 27 at 21:41:33 +0200, Eric Leblond wrote: > >> Hello, > >>> As I'm about to conclude a massive release of libraries and the tools, I > >>> think that forcing the use of a recent library version is better as it > >>> also contains several bugfixes. > > > > I did not really understand the meaning of the word "forcing" till I discover > > that you've forced the use of at least libnfnetlink 0.0.39 in configure.in. > > > > IMHO, this is too strict and not an easy step into ulogd2 acceptance: > > Instead of having a single software to compile, users will need to > > compile almost all libnf* library :( > > This latest library release-set includes several fixes. Having the > lastest version installed always suppose an extra effort for the users. > Moreover, bumping the dependencies is also a way to force packagers to > upgrade. > > >> Users prefer to use the library packaged in their system and it will > >> take some time before every distribution includes the required > >> libnfnetlink_log. This patch finishes to provide a clean backward > >> compatibility at a really low cost. > > The libraries are backward compatible. We did not break the API nor ABI > so that users with the own applications are not forced to upgrade. I know that. The point is really not here. Work done in this area is clearly good. > The thing changes if we discuss about the userspace tools. I think that > they must use the latest library release. Thus, we avoid having people > that report problems that has been already fixed. I really don't like this policy because it is too efficient: You will have any report before required version of library get into the distributions because people won't test it it there is too much work to build a test system. > >> Furthermore, I don't think a gcc warning about "unused variable" will > >> force a lot of people to upgrade ;) > > > > But a strict configure.in will :/ > > The gcc warning is anecdotic, it was not the main reason to bump the > dependencies. I know that, I was just talking about the discussed patch. But after looking at the changelog and after using ulogd2 without problem with older libraries, I don't think the dependencies bump was needed. If a bug in the library was preventing ulogd2 to work correctly, I will be the first to ask for a dependencies bump but it is far from being the case here. > Looking at the current state of the libraries, I think that it's the > best for now. Later on, once they get more stable and well-tested, we > may change this policy. Please revert commit 3178606785161296dc5a1bd4d42d965db8b3e2cd if you want to apply this strict policy. This code is now useless. BR, -- Eric Leblond INL: http://www.inl.fr/ NuFW: http://www.nufw.org/
Attachment:
signature.asc
Description: Digital signature
