On 12/16/09 12:15, [email protected] wrote:
Trying to summarize what I was trying to say, I believe both ways are
valid and have there own advantages and at the end it is more a matter of
taste (or doing it the same way as Equinox to ensure sort of consistency
across frameworks).
Given the differences in our resolver impls, I am not sure it will be
possible for us to do it the same way, but my guess is that the
intersection approach would work in most cases where people would expect it.
-> richard
Statically using the host bundle versions leads to a more predictable
behaviour at build time, whereas dynamically adjusting to fragment imports
leads to better runtime resolution.
We could very well go for the dynamic solution, but encourage the "good
way of doing fragments" by not restricting the host import versions...
Thanks
Andreas
"Richard S. Hall"<[email protected]> wrote on 12/16/2009 09:00:24 AM:
p.s. I should point out, I am not against your approach. It is a little
more restrictive than what I am proposing, but more lenient than
currently implemented. Rather than requiring floor and ceiling of the
fragment to be == to the host range, like the current approach, you
require the floor and ceiling of the fragment to be<=/>= the host
range.
On 12/16/09 11:48, [email protected] wrote:
Richard,
this could also work, but will add more complexity to it. You may run
into
situations were you need to randomly abandon one fragment:
Host imports version [1.0.0,2.0.0), fragment 1 imports [1.0.0,1.5.0),
fragment 2 imports [1.5.0,2.0.0)
In my approach, both fragments would be exluded, in your approach you
have
to find a tie breaker to decide on which fragment to include.
With my approach, statically determining the valid version range by
looking at the host bundle, you'd keep the control over valid versions
with the host bundle and everybody using it can be sure that the host
and
all possibly existing fragments will run when obeying the host version
range (so, when using a consistent set of bundles you can be sure that
you
can add any fragment you want without updating other bundles). When
dynamically determining the version range by looking at fragments as
well,
you'd take away this guarantee. It may happen that a consistent set of
bundles gets inconsistent out of a sudden because of a fragment with
too
restrictive import versions (meaning the fragment does not work).
The dynamic approach obviously has the advantage to be more flexible,
but
I think it should rather be the responsibility of a fragment developer
to
adhere to the version ranges of the host bundle than the
responsibility of
the user of a host bundle to comply to (more restrictive) version
ranges
in fragments he'd like to use.
Maybe it would be interesting to look into how Equinox is doing it,
unfortunately I don't know. Can anybody help?
Thanks
Andreas
"Richard S. Hall"<[email protected]> wrote on 12/16/2009 08:30:10
AM:
What I was planning on doing is taking the intersection, which would
be
the highest floor and the lowest ceiling of each overlapping version
range...and of course, if there is no intersection, then they are in
conflict and the fragment would be thrown out.
-> richard
On 12/16/09 11:13, [email protected] wrote:
Guo,
I think your algorithm is not 100% correct. The host version
boundaries
must lie within the fragment version boundaries. So, looking at your
example:
Host version [2.0.0,3.0.0)
Fail fragment versions [1.0.0], [1.0.0,2.0.0), [3.0.0]
Fails, since version lies completely outside host version boundaries
Pass fragment versions
[1.0.0,5.0.0),
Passes, since host version lies within these boundaries
[2.5.0,2.9.0)
Fails, since host version lies outside boundaries. E.g., when host
is
importing 2.1.0 this would cause fragment to fail.
Thanks,
Andreas
Guo Du<[email protected]> wrote on 12/16/2009 02:25:50 AM:
On Wed, Dec 16, 2009 at 12:54 AM,<[email protected]>
wrote:
Hi,
I found that Felix is validating the compatibility of host vs.
fragment
imports by ensuring that in case host and fragment are importing
the
same
package, they should use the exactly same version (range). I
believe
that
this is a little too restrictive and Felix should also allow the
host
bundle to be more restrictive on the version range than the
fragment.
This
way, it is still guaranteed that the fragment will run with using
the
version range from the host bundle (which is a subset of the
fragment
version range in this case).
I just ran into this problem when trying to use the Hibernate +
Annotations bundles packaged by SpringSource. Hibernate
Annotations
is
a
fragment bundle, hosted by the Hibernate bundle. Hibernate imports
org.dom4j;version="[1.6.1, 1.7.0)" whereas the fragment imports
org.dom4j;version="[1.6.1, 2.0.0)" and the current implementation
does
not
allow this fragment to be linked to its host.
What do you think?
+1
I have this problem with spring-osgi-extender as well. For
fragment,
we may enable the fragment when there are common set between host
version and fragment version.
Host version [2.0.0,3.0.0)
Fail fragment versions [1.0.0], [1.0.0,2.0.0), [3.0.0]
Pass fragment versions [1.0.0,5.0.0), [2.5.0,2.9.0)
Any drawback to this approach?
-Guo
