On 11/26/09 20:55, Ivan wrote:
Thanks, Richard, please help to check my comments.
2009/11/26 Richard S. Hall<[email protected]>
On 11/25/09 21:49, Ivan wrote:
Hi,
While trying the Felix as the OSGI runtime, I have some questions
about
the resolve policy, please help to give some comments, thanks !
1. Can an import connection be wired to an installed (not resolved)
bundle ? I found there is description in the spec 3.7, which says "A
resolved exporter must be preferred over an unresolved exporter.". So does
it mean that the answer is yes ?
All bundles must be resolved to have wires and to have wires to them. The
spec is just saying you should prefer already resolved bundles over bundles
that you must yet resolve.
2. I have a bunlde A and bundle B, bundle A needs to import some
packages
from bundle B. But my installation order is that first A, then B. From the
log messages, I could see after installing A, FelixDispatchQueue said that
bunldle A is in the installed state. Then I install B, the messages showed
that " Bundle B resolved, Bundle A resloved". I wonder when the resolve
action will occured ? Will Felix check whether all the installed bundle
could be resolved once a new bundle is installed or resolved ?
Felix does not try to resolve bundles automatically, it only resolves them
on demand. Typically this is when you try to start a bundle, which will
resolve that bundle plus the bundles on which it depends transitively.
Let's say we have three bundles A, B, C.
A depends on both B and C.
B depends on C.
The installation order is A, B, C. After installing them, their status
should be :
A : installed
B : installed
C: resolved
C would not be resolved, since the Felix framework doesn't automatically
resolve bundles until someone tries to use them. They would all be in
the INSTALLED state.
Then I start them in the same order.
While starting bundle A, Felix will first try to resolve B, or A could be
resolved as its import requirements can not be fulfilled. In other words,
although a "real" wire connection had not been established between A and B,
as B is in _installed_ status, but there may be a "candidate" connection
existed between A and B, so Felix knows it has to first resolve B ?
Yes. If you try to start A, the framework tries to resolve A first. It
looks for all candidates to solve A's dependency, which is only B. The
framework checks to see if B is resolved, since it is not it tries to
resolve B first. Since B depends on C and C is not resolved, the
framework then tries to resolve C. Assuming there are no errors, then C
is resolved, then B is resolved with a wire to C, then A is resolved
with a wire to B. Then A can be started.
-> richard
3. Related to question 2, if I have ten bundles and I am sure that each
bundle could find export packages from other nine bundles. But I do not
install them according to their dependency relations. Then, will the logic
below work correctly ? In my tests, it seems that they could work
correctly.
String[] locations = {......};
Bundle[] bundles = new Bundles[locations.length];
for(int i=0;i<locations.length;i++) {
bundles[i] = bundleContext.installBundle(locations[i]);
}
for(Bundle b : bundles) {
b.start();
}
Yes. For example, if all of your bundles are installed and started, on a
subsequent restart of the framework, it will first reinstall all bundles in
one pass, then start them in another pass, which ensures dependencies can be
satisfied.
The above code does the same thing.
-> richard
Ivan
---------------------------------------------------------------------
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]
