You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@felix.apache.org by Peter Kriens <Pe...@aQute.se> on 2005/12/05 08:15:35 UTC

Re[2]: IDE Lockin? (was Re: Repository/package structure)

JM> and some tooling even has graphical editors and wizards for it.  If you
JM> pick some other format then you start duplicating the info in the manifest
JM> and have to start talking about tooling for it etc etc.
That is very restrictive implying that you can never generate the
import statements. I have been making bundles for 8 years now and the
greatest improvement I ever had was when I could leave the imports up
to btool. I generate the manifest from a template, so duplication is
not an issue.

I really think (ab)using the manifest as an input instead of the
traditional output is a wrong idea because it was just not intended to
be an input. For the R4 build I tried to do it but it became too
artificial.

Also, the classpath is a wider issue. I frequently created multiple
bundles from a single project. Our test suites use many bundles per
test suite sometimes up to 20 per testcase, it would be dramatic if
they all had to become their own project (which I think is implied by
your manifest model).

And about the graphic editors ... I wish there was a way to disable
them :-( They seem to be always in the way and too focused on Eclipse
features instead of simple bundles.

If the .classpath does not support this model, I sure hope Apache will
specify a format that they will support :-(

Kind regards,

     Peter Kriens









JM> Peter Kriens <Pe...@aQute.se> wrote on 12/03/2005 06:41:02 AM:

>> One thing that might solve the dichotomy between tool lock in and
>> flexibility to change IDEs is to have standard file format for the
>> dependencies.

JM> Errr, the OSGi bundle manifest is standard and hey its even there already
JM> in bundles.  It's relatively simple, there is excellent documentation :-)
JM> and some tooling even has graphical editors and wizards for it.  If you
JM> pick some other format then you start duplicating the info in the manifest
JM> and have to start talking about tooling for it etc etc.

>> I am using the Eclipse .classpath file in btool, but I made the parser
>> pluggable so I can support other dependency files as well. If
>> Felix/Apache just selects a standard format for declaring the 
>> dependencies, then
>> it should not be that hard for Eclipse to support this format with a
>> plugin.

JM> The Eclipse .classpath file is very brittle for use in this way.  Eclipse
JM> PDE just reads the bundle manifest file.  It completely generates the
JM> classpath.  The .classpath file really isn't used in normal bundle
JM> development with PDE.

JM> Jeff


-- 
Peter Kriens                              Tel +33870447986
9C, Avenue St. Drézéry                    AOL,Yahoo: pkriens
34160 Beaulieu, France                    ICQ 255570717
Skype pkriens                             Fax +1 8153772599


Re[4]: IDE Lockin? (was Re: Repository/package structure)

Posted by Peter Kriens <Pe...@aQute.se>.
I took this offline because the scope of this discussion is not felix
anymore :-)

Kind regards,

     Peter Kriens

JM> Peter Kriens <Pe...@aQute.se> wrote on 12/05/2005 02:15:35 AM:

>> JM> and some tooling even has graphical editors and wizards for it.  If
JM> you
>> JM> pick some other format then you start duplicating the info in the
JM> manifest
>> JM> and have to start talking about tooling for it etc etc.
>> That is very restrictive implying that you can never generate the
>> import statements. I have been making bundles for 8 years now and the
>> greatest improvement I ever had was when I could leave the imports up
>> to btool. I generate the manifest from a template, so duplication is
>> not an issue.

JM> How does btool manage the import version ranges and export version
JM> numbers?  I can see where it would be able to generate the list of
JM> packages needed but not their versions.  The btool function is indeed
JM> useful (we are in the process of adding similar function to PDE).  The use
JM> of manifests is not a limitation in doing this.

JM> The point here is that developers really do have to manage this 
JM> information.  To them the underlying representation is not important.
JM> Implicit in what you propose is a duplicate form of metadata that has to
JM> capture the same kind of info that is already in the manifest.  Tooling
JM> needs to understand both then because some things it manipulates has
JM> manifests (they are already built) and some don't (they are being built).
JM> This new metadata format has to somehow facilitate all the nuances of the
JM> manifest.mf but the syntax/form will be different. This is painful.  Also,
JM> it implies that things like editors have to be duplicated as well.

JM> In the end, I don't see a real benefit here.  Information that can be
JM> generated can be generated into a manifest.  Information that has to be
JM> entered by users can be entered into that same manifest.  The runtime and
JM> the tooling can all use that manifest.  One syntax. One file. Simple.

>> I really think (ab)using the manifest as an input instead of the
>> traditional output is a wrong idea because it was just not intended to
>> be an input. For the R4 build I tried to do it but it became too
>> artificial.

JM> Here in lies our disconnect.  There is a whole mess of information in the
JM> manifest that has to be crafted/specified by a developer.  Things like the
JM> version ranges and export versions, the choice between Import-Package and
JM> Require-Bundle, fragment hosting, ....  Setting things up so that you can
JM> spec this in a different file in a different syntax seems 
JM> counterproductive.

>> Also, the classpath is a wider issue. I frequently created multiple
>> bundles from a single project. Our test suites use many bundles per
>> test suite sometimes up to 20 per testcase, it would be dramatic if
>> they all had to become their own project (which I think is implied by
>> your manifest model).

JM> This is actually not a manifest issue but a classpath management issue. To
JM> get a highly accurate classpath you have to compile each bundle on its own
JM> (so that you can control visibility correctly).  This implies separate
JM> build structures.  In the Eclipse case that means separate projects. Don't
JM> know what it means in Maven or any other setup.

JM> Project sets and working sets in Eclipse help you manage collections of
JM> projects but yes, having lots can still be an issue depending on what you
JM> do.  Note that this does not particularly impact you if you are not using
JM> Eclipse since you don;t have projects.  There is relatively little that is
JM> magic about being a project.  You just have to have a .project file and
JM> .classpath file that points to the source folders for the project.  You
JM> can inject these into many different repo/source structures and does not
JM> impact your ability to use other tooling (IDEA, javac, ...).

>> And about the graphic editors ... I wish there was a way to disable
>> them :-( They seem to be always in the way and too focused on Eclipse
>> features instead of simple bundles.

JM> I think you should be looking at the latest Eclipse builds.  You will find
JM> very few references to Eclipse-specific pieces (other than the use of the
JM> term "plugin" vs. "bundle").  We are actively and aggressively seeking to
JM> make PDE a first-class bundle development environment so for those who
JM> care, please get Eclipse 3.2 M4 (end of next week) and try the tooling.
JM> Let us know what you don't like.

JM> Jeff


-- 
Peter Kriens                              Tel +33870447986
9C, Avenue St. Drézéry                    AOL,Yahoo: pkriens
34160 Beaulieu, France                    ICQ 255570717
Skype pkriens                             Fax +1 8153772599


Re: Re[2]: IDE Lockin? (was Re: Repository/package structure)

Posted by Jeff McAffer <Je...@ca.ibm.com>.
Peter Kriens <Pe...@aQute.se> wrote on 12/05/2005 02:15:35 AM:

> JM> and some tooling even has graphical editors and wizards for it.  If 
you
> JM> pick some other format then you start duplicating the info in the 
manifest
> JM> and have to start talking about tooling for it etc etc.
> That is very restrictive implying that you can never generate the
> import statements. I have been making bundles for 8 years now and the
> greatest improvement I ever had was when I could leave the imports up
> to btool. I generate the manifest from a template, so duplication is
> not an issue.

How does btool manage the import version ranges and export version 
numbers?  I can see where it would be able to generate the list of 
packages needed but not their versions.  The btool function is indeed 
useful (we are in the process of adding similar function to PDE).  The use 
of manifests is not a limitation in doing this.

The point here is that developers really do have to manage this 
information.  To them the underlying representation is not important. 
Implicit in what you propose is a duplicate form of metadata that has to 
capture the same kind of info that is already in the manifest.  Tooling 
needs to understand both then because some things it manipulates has 
manifests (they are already built) and some don't (they are being built). 
This new metadata format has to somehow facilitate all the nuances of the 
manifest.mf but the syntax/form will be different. This is painful.  Also, 
it implies that things like editors have to be duplicated as well.

In the end, I don't see a real benefit here.  Information that can be 
generated can be generated into a manifest.  Information that has to be 
entered by users can be entered into that same manifest.  The runtime and 
the tooling can all use that manifest.  One syntax. One file.  Simple.

> I really think (ab)using the manifest as an input instead of the
> traditional output is a wrong idea because it was just not intended to
> be an input. For the R4 build I tried to do it but it became too
> artificial.

Here in lies our disconnect.  There is a whole mess of information in the 
manifest that has to be crafted/specified by a developer.  Things like the 
version ranges and export versions, the choice between Import-Package and 
Require-Bundle, fragment hosting, ....  Setting things up so that you can 
spec this in a different file in a different syntax seems 
counterproductive.

> Also, the classpath is a wider issue. I frequently created multiple
> bundles from a single project. Our test suites use many bundles per
> test suite sometimes up to 20 per testcase, it would be dramatic if
> they all had to become their own project (which I think is implied by
> your manifest model).

This is actually not a manifest issue but a classpath management issue. To 
get a highly accurate classpath you have to compile each bundle on its own 
(so that you can control visibility correctly).  This implies separate 
build structures.  In the Eclipse case that means separate projects. Don't 
know what it means in Maven or any other setup.

Project sets and working sets in Eclipse help you manage collections of 
projects but yes, having lots can still be an issue depending on what you 
do.  Note that this does not particularly impact you if you are not using 
Eclipse since you don;t have projects.  There is relatively little that is 
magic about being a project.  You just have to have a .project file and 
.classpath file that points to the source folders for the project.  You 
can inject these into many different repo/source structures and does not 
impact your ability to use other tooling (IDEA, javac, ...).

> And about the graphic editors ... I wish there was a way to disable
> them :-( They seem to be always in the way and too focused on Eclipse
> features instead of simple bundles.

I think you should be looking at the latest Eclipse builds.  You will find 
very few references to Eclipse-specific pieces (other than the use of the 
term "plugin" vs. "bundle").  We are actively and aggressively seeking to 
make PDE a first-class bundle development environment so for those who 
care, please get Eclipse 3.2 M4 (end of next week) and try the tooling. 
Let us know what you don't like.

Jeff