forked from google/pinject
-
Notifications
You must be signed in to change notification settings - Fork 0
/
TODO
35 lines (33 loc) · 1.69 KB
/
TODO
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
TODO:
- ensure that memoization works properly with partial injection
- improve DirectlyPassingInjectedArgsError
- auto-require provider method args
- allow testing that bindings exist to allow object to be provided
- move obj_provider into injection_context, so that provisers only have one arg?
- describe features specifically omitted, e.g.,
- circular injection
- scope annotations on classes
- figure out whether/how Pinject works with inheritance
- make provider bindings inject a provider that has the correct signature (difficult!)
Maybe TODO:
- make @inject complain if not applied to initializer or binding spec method
(doesn't seem possible: method passed to decorator has type 'function' and
seemingly no info about the class in which it's defined)
- allow object_graph.provide('arg name')?
- standard tests for scopes (reentrant? thread-safe?), annotations (eq?
hash?), etc.
- change default scope back to prototype?
- eager singletons
- find modules on PYTHONPATH instead of having to import them
- automatically instantiate the concrete subclass of an interface?
(use abc module)
- visual graph of created objects
Questions:
- How should I deal with someone wanting to instantiate a class in a scope,
without using an arg name in between? Scopes apply to binding keys,
which are arg names, not classes. Is it going to be harmful if you
can't instantiate a class directly in a scope? I don't think so, but
I'm not sure. It makes some things awkward, like a special scope from
which it's OK to inject objects from any scope. Maybe the main
ObjectGraph method should be wrap() instead of provide_class()??
- Why should @inject apply to __init__() rather than the class?