Fixed
Status Update
Comments
ky...@bytedance.com <ky...@bytedance.com> #2
If you need to add a custom Navigator but don't want to subclass NavHost (which we expect will be needed in most cases), the correct approach is to add it before you set your graph:
1) Don't use app:navGraph in your XML
2) Get a reference to your NavController
3) Add your Navigator to the NavController
4) Call setGraph on the NavController
This relies on the fact that NavController waits until the graph is set before restoring state / doing anything at all.
Technically, the createFragmentNavigator() call in NavHostFragment is already called at the same time as your proposed onNavControllerCreated() method so as a workaround, you could override that method, add your Navigator, then return super.createFragmentNavigator().
1) Don't use app:navGraph in your XML
2) Get a reference to your NavController
3) Add your Navigator to the NavController
4) Call setGraph on the NavController
This relies on the fact that NavController waits until the graph is set before restoring state / doing anything at all.
Technically, the createFragmentNavigator() call in NavHostFragment is already called at the same time as your proposed onNavControllerCreated() method so as a workaround, you could override that method, add your Navigator, then return super.createFragmentNavigator().
rk...@google.com <rk...@google.com>
rk...@google.com <rk...@google.com> #3
The first solution does not work, because after screen rotation, the graph (probably a restored graph) is set when Activity.setContentView() is called.
The workaround with createFragmentNavigator() works, though it's not very elegant. Maybe the method could be renamed to createNavigators() and return a list or an array of Navigators?
Anyway, it is a good enough solution for now, so it is not very urgent.
In conclusion:
* I think that a better support for custom destinations should be provided before final release.
* I would suggest describing the "Add support for new destination types" section in the documentation in more details.
Thank you
The workaround with createFragmentNavigator() works, though it's not very elegant. Maybe the method could be renamed to createNavigators() and return a list or an array of Navigators?
Anyway, it is a good enough solution for now, so it is not very urgent.
In conclusion:
* I think that a better support for custom destinations should be provided before final release.
* I would suggest describing the "Add support for new destination types" section in the documentation in more details.
Thank you
rk...@google.com <rk...@google.com> #4
I suspect the issue you're running into for the first solution is that you're using the setGraph(R.navigation.graph) method, which does indeed cause it to automatically inflate the graph by id on rotation. Using setGraph(navGraph) where navGraph is a navGraph object you've inflated with navController.getNavInflater().inflate(R.navigation.graph) will never automatically be recreated.
Definitely something we can do better on though :)
Definitely something we can do better on though :)
Description
AutoLock
appears to be a custom implementation ofstd::unique_lock/lock_guard
. It does, however, a method namedisLocked()
. There is only one use of the method at the moment:I believe, that has room for improvement. Firstly, even if we checked
isLock()
in general, asisLock()
is not thread-safe, there is no guarantee (in general) that the same thread can then lock it. It is only useful when there is only one thread that locks/unlocks the givenAutoLock
object and the thread wants know whether thelock
is acquired or not by itself. I don't think that's useful. I think the method being public, it has a good chance to confuse external developers.My belief is that the
AutoLock
should be simply replaced withstd::unique_lock
and/orstd::lock_guard
, as many uses of the class do not calllock
orunlock
. And,isLocked()
should be removed. The only call site should be implemented without that.Aside from all mentioned above, I am trying to compile the code with
--enable-thread-safety-checks
, and the first code that blocked the compilation on Linux was the use ofisLocked()
.