Hi Julius, I understand your concerns. In this case however, we can give very precise guidelines. That's mainly because there would only be the 1.4 NIO implementation to choose from, at least for a while :-) After that, it's: use the 1.3 IO implementation only if you have to. And hopefully, we would have the performance statistics to back that up.
I don't think the end users will have to make a choice here, it's the application developers that decide. I also imagine some kind of autoconfig class that tries to load the NIO implementation and falls back to IO if that fails. There would be a section about choosing NIO or IO in the documentation on the web site. Of course some people won't read that and will post a question on the mailing list, but they are easily directed to the web site. No, you are not annoying at all. The more input we get for this discussion, the better the choices that will be made. Thank you for addressing a real-life concern which at least I would have overlooked otherwise. cheers, Roland Julius Davies <[EMAIL PROTECTED]> 25.01.2005 20:28 Please respond to "HttpClient Project" To HttpClient Project <[email protected]> cc Subject Re: [POLL] Minimal JRE requirement for HttpClient 4.0 Hi, I haven't really been following this discussion, but this dependency table caught my eye: > http-common-api depends on 1.3 > http-common-implIO depends on 1.3 > http-common-implNIO depends on 1.4 Personally, I don't like it when applications ask me to choose an implementation (Apache2's thread-model, for example!). I never feel like I have adequete information to make a choice. The implementations might as well be called "Door Number 1" and "Door Number 2". In this specific case one might argue that java developers understand the difference between IO and NIO, and so we're informed enough to make the choice. But even if I know that I'm supposed to prefer NIO, I still prefer the code that is running in 99% of installations. What will the "default" out-of-the-box implementation will be? Would the end-user have to make some configuration changes? (How much traffic would these choices generate on this mailing list?) Hope I'm not being annoying! HTTPClient is fantastic, and I'm really excited to see so much momentum behind it. yours, Julius On Tue, 2005-25-01 at 20:15 +0100, Oleg Kalnichevski wrote: > On Tue, 2005-01-25 at 14:33 +0100, Roland Weber wrote: > > Hi Oleg, > > > > Still, I wouldn't expect java.nio to show up there. Sockets neither, > > in some of the smaller profiles. And even then, 1.4 features could > > only be found in high-end, early-adopter devices for quite a while. > > If we want to keep the J2ME option open for widespread use, we should > > have streams instead of channels in the API, and create wrappers around > > channels if an implementation uses java.nio. Which would make Java 1.3 > > support an option: > > > > http-common-api depends on 1.3 > > http-common-implIO depends on 1.3 > > http-common-implNIO depends on 1.4 > > > > and one needs at least the -api and one -impl component to get going. > > > > What do you (all) think? > > > > Roland, > > I am +1 to try to keep http-common API Java 1.3 compatible. I would not > worry about http-common-implIO and concentrate exclusively on NIO impl > until some real J2ME demand materializes backed by a number of > contributers. > > Oleg > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
