Better lazy resolution #91
Conversation
|
I have come to the conclusion that this approach is simply not good enough in all scenarios. We need to add an I will work on that as soon as I can... |
|
I reworked the branch to add a It is working fine, but is a breaking change from the SJC 2.x.x API, so bumps the snapshot version to 3.0.0. It should be merged only in concert with other breaking SJC branches, particularly |
Specifically, we test that lazy object resolution works in conjunction with a call to getAll(), to ensure that pending objects get resolved.
The LazyObjects instance will definitely return a collection of objects of type PT, not just Object. This distinction is important now that the ObjectIndex is more selective about which pending objects it resolves.
This breaks backwards compatibility, so bumps the development snapshot version to 3.0.0. The rationale is to make it possible for the ObjectIndex to better discriminate which objects should be resolved for a particular call to ObjectIndex#get(Class). The current behavior is to resolve _all_ pending objects with _any_ call to get(Class), which is generally suboptimal.
If a pending object will not be of a type compatible with the requested type, then skip resolution of that pending object. This improves performance of the get(Class) and getAll() methods in cases where there are a large number of pending objects in the index. It also helps to eliminate difficulties like those in issue #90.
This tests that the addLater method works in conjunction with calls to get(Class). The logic is nontrivial because we only want to resolve pending objects when the types are compatible with the requested one.
The filterInstances method should not alter the return type of the input list. The purpose of that method is to potentially remove some elements from the input list -- nothing more -- so it should still return a List<PT>, not something else.
Instead of maintaining our own list of singletons, we should be using the ObjectService.
ctrueden
force-pushed
the
better-lazy-resolution
branch
from
cbc827c to
d6a7eba
Compare
|
While rebasing this branch, I noticed a problem with |
ctrueden
force-pushed
the
master
branch
3 times, most recently
from
53b6733 to
3dc99c9
Compare
MERGE ONLY IN CONCERT WITH THE OTHER 3.0.0 BRANCHES.
This improves the
ObjectIndexlazy object resolution feature to only resolve objects as needed based on the type. Previously, as soon asget(Class)orgetAll()was called, all pending objects would be resolved. Now we call the newLazyObjects#getType()method to see if the type of theCollectionis even worth bothering about.See also issue #90.