Talk:PropertyFiles

FWIW, The Eclipse Plug-in Development Environment (PDE) has tools for configuring OSGi manifests. And it works with Felix.

--Wayne Beaton 14:54, 16 December 2008 (CET)

Actually, the real reason we had to introduce Bundle-ManifestVersion is because we changed the semantics of the existing Export-Package header (in R3 it also implied an Import-Package, but in R4 it does not). Otherwise, I agree it would have been smarter to include the manifest version from the beginning. :-)

--Richard S. Hall 15:25, 16 December 2008 (CET)

Also, Wayne makes a good point. BND will also "verify" bundles, but I am not sure if it would have pointed out the issue in this case. Of course, the moral to this story is, don't use Require-Bundle, use Import-Package instead. ;-)

--Richard S. Hall 15:27, 16 December 2008 (CET)

If you used any OSGi-related tools like Eclipse PDE... this would catch the problem immediately.

Your point about including version identifiers is very valid and people should learn to take it in heart. It's a common mistake people forget about.

--Chris Aniszczyk 16:03, 16 December 2008 (CET)

Somewhere in TheAPIBook, I guess it is in Chapter 9, I mentioned that one of the best APIs is wizard. Good wizard can turn any API, regardless how bad, into perfectly shining, beautiful star. I am sure that if I used PDE, I would avoid falling into the ManifestVersion trap.

--JaroslavTulach 14:39, 19 December 2008 (CET)

Hi Jaroslav,

what if a developer accidentally specifies a wrong Bundle-ManifestVersion value? The Require-Bundle tag would still be ignored!

To me, merely making the Bundle-ManifestVersion header mandatory from version 1 on does not help very much in practice.

Instead, the user of the API should be *informed* if the service providing this API detects "inconsistent use" of different versions of the API. In this case: the OSGi container or bundling tool should warn at packaging or deployment time if one tries to create/use a bundle without Bundle-ManifestVersion specified, but containing tags from "non-1.0" API versions inside the bundle manifest.

This is similar to a "syntax check" done by programming language compilers. If our OSGi tooling does not support this syntax check (yet?), at least the runtime container should do.

The point I'm trying to make is: let's not blame the API *specification* for a *missing mechanism* which can ensure that the API is used correctly.

Best regards

Ansgar

--Ansgar Konermann 17:34, 25 November 2010 (CET)

Hello Ansgar, I think you are right. The most important thing is to keep the user informed. Probably the OSGi spec, tools, impls could do better job. However my point is that having the version present (and required) since first version makes your life easier. This is not that visible in case of PropertyFiles, where it is relatively easy to add new attribute in (almost ~ 99%) compatible way, but in case of binary formats, it is essential to get the version right as soon as possible, otherwise one can completely loose offset and interpret the content completely incorrectly.

Btw. your fix is not 100% backwardcompatible: In old version I might accidently use one of the added attributes for completely unrelated purposes. If I then upgrade the OSGi container to new version, it will complain and refuse my bundle which used to run in previous version. I guess this is the reason why new OSGi specification requires the Bundle-Version attribute to be present. Guys wanted to be 100% backwardcompatible.