Previously, it was not explicit that the `onCreate` happened to be invoked
in the UI thread. Now it is, due to passing `new Handler(Looper.getMainLooper())`.
Also, the categories are now loaded in a background task, and then the UI is
updated on the UI thread.
Previously, they were registered, then forgotten. This means that each time
the start swap view was run, another receiver was registered. As a result,
they were being invoked multiple times.
It doesn't appear that this had any specific side effects which were terrible,
but they definitely have the potential to going forward.
Note that because we are not using `Fragments` with their convoluted, but at
least well documented API, I'm not 100% certain that I've unregistered the
receivers at the right location.
Previously, something like this would happen:
* Swap service is cancelled
* WiFi swap is asked to stop
* Event is broadcast when done
* UI listens to this event
* Upon receiving the event, it updates the UI
* Updating the UI triggers an event, causing the process to happen again
An alternative solution to this would have been for the UI to stop listening
to listeners before WiFi swap is shut down, but that is then only specific
to the case when the swap view is being destroyed/removed. This could also
happen in other situations however, such as when the swap service times out.
When the view is detached, then the listeners will be unregistered.
This will also help in the future so that they can be temporarily
unregistered when manually changing the state of the switches.
Thread runs at normal priority by default. AsyncTasks are integrated into
Android for handling things running in the background while keeping the UI
responsive.
This reverts most of commit 828cc272ee5235f868104b009349cc7e835e144f.
Not sure if we should be here or not in this situation, so this is
a little bit defensive. Can't bind to an IP address of `null`, so
don't bother starting LocalHTTP server unless we have an IP.
`Thread.run()` is not the correct call, changed to the correct `Thread.start()`.
Also, explicitly indicate that we want the stopping of wifi to happen in the
background.
On some devices this can take some time (i.e. a second) and the UI needs
to be disabled for that time. This should stop users quickly stopping and
starting regularly, queuing up many "start jmdns, stop jmdns, start jmdns"
calls.
We still allow them in single-line statements, like:
if (foo) bar;
for (int i : ints) bar;
Everything else should use braces to help readability and avoid silly
human mistakes that might result in bugs.
These changes were completely automated via a python script.
Work around dead activity issue in AppDetails
It seems like install() sometimes runs when the AppDetails activity is
finished or finishing. This results in the windows (dialogs) failing to
show, and a BadTokenException to fire:
android.view.WindowManager$BadTokenException: Unable to add window -- token android.os.BinderProxy@d6e3570 is not valid; is your activity running?
This seems to be the culprit:
at org.fdroid.fdroid.AppDetails.install(AppDetails.java:840)
at org.fdroid.fdroid.AppDetails$AppDetailsListFragment.install(AppDetails.java:1657)
at org.fdroid.fdroid.AppDetails$AppDetailsListFragment.onListItemClick(AppDetails.java:1721)
Apparently, you can check whether an activity/context is being finished:
https://stackoverflow.com/questions/7811993/error-binderproxy45d459c0-is-not-valid-is-your-activity-running
I cannot reproduce this issue, thus can't say whether this fixes it or
not. Either way, it can't hurt to try. This can be reverted if we see
ACRA reports of this in the future, and the issue reopened.
Fixes#565.
See merge request !204
It seems like install() sometimes runs when the AppDetails activity is
finished or finishing. This results in the windows (dialogs) failing to
show, and a BadTokenException to fire:
android.view.WindowManager$BadTokenException: Unable to add window -- token android.os.BinderProxy@d6e3570 is not valid; is your activity running?
This seems to be the culprit:
at org.fdroid.fdroid.AppDetails.install(AppDetails.java:840)
at org.fdroid.fdroid.AppDetails$AppDetailsListFragment.install(AppDetails.java:1657)
at org.fdroid.fdroid.AppDetails$AppDetailsListFragment.onListItemClick(AppDetails.java:1721)
Apparently, you can check whether an activity/context is being finished:
https://stackoverflow.com/questions/7811993/error-binderproxy45d459c0-is-not-valid-is-your-activity-running
I cannot reproduce this issue, thus can't say whether this fixes it or
not. Either way, it can't hurt to try. This can be reverted if we see
ACRA reports of this in the future, and the issue reopened.
Fixes#565.
Fix 555 content provider invalid uri
Was not correctly encoding "/" characters when searching. This caused the Uri used by the Content Providers to include a slash, which makes it look like a separate segment of the path which was wrong. Now correctly encodes "/" characters. Also noticed one other place incorrectly encoding characters, where they would've been double encoded when added as query parameters to a Uri.
See merge request !203
Refactor swap "peer finders" to use ReactiveX
*NOTE: This includes the commit specified by !197.*
In the old code, there is a _lot_ of procedual style "Is this peer finder running, if so, do this". In addition, the choice to do things on background threads or not is a little ad-hoc. Finally, the `SwapService` needs to know about both bluetooth and wifi peer finders, whereas really they are both only there to emit "Peers", regardless of the type.
As such, some improvements in this change are:
* The choice to run peer finding on a background thread is made once, at a higher level when starting the peer finder.
* No longer does the UI code ask "Am I searching for peers". It instead waits to be told whether it is or isn't.
* The addition of new types of peers in the future is the job of the Peer finder itself. It quietly aggregates all of the Peer Finders it knows about into a single observable that emits different types of peers.
This code doesn't fix any particular issue, but rather it is about making the entire swap workflow easier to reason about. I plan on migrating more of this workflow to this functional style in the future, and hopefully that will have benefits in terms of stability and code understanding.
See merge request !198
I misread the documentation when first using the `appendEncodedPath` method,
because it expects the path to already be encoded. This causes a bug because
if you search for a '/'. The result is a malformed URI that has the path
'/search//' instead of '/search/%2F'.
Using `appendPath` will always encode the string given to it, which is desirable.
Also check for empty strings, and return a URI that gives all apps. This was
not strictly neccesary, because the code which invokes it checks for empty
strings, but if somewhere else in the future starts to use this code, they
would've had to know to check for empty strings first.
Fixes#555.
Put null check around access of `R.id.header` fragment.
Please note I haven't reproduced the specific problem. Also, the stack
traces being reported are only marginally informative, because they are
in response to a content providers firing events, and thus don't have
any context about when or where the event was fired from.
However, my looking at the code seems to indicate that this will prevent
NPE when the Activity is no longer visible but an app is finished
installing. Also, the view should still update correctly on resuming the
Activity because the `onResumeFragments()` methods will be invoked
which invokes the `refreshHeaders()` method.
Fixes#286.
See merge request !202
Please note I haven't reproduced the specific problem. Also, the stack
traces being reported are only marginally informative, because they are
in response to a content providers firing events, and thus don't have
any context about when or where the event was fired from.
However, my looking at the code seems to indicate that this will prevent
NPE when the Activity is no longer visible but an app is finished
installing. Also, the view should still update correctly on resuming the
Activity because the `onResumeFragments()` methods will be invoked
which invokes the `refreshHeaders()` method.
Fixes#286.
The benefits of this are as follows:
No longer need to worry about how many types of `Peer`s exist.
There is a single publicly accessible `PeerFinder` which aggregates
the results of both the Bluetooth and Bonjour peer finders. In the
future if another is added, the consumer of the peer finder
(i.e. `StartSwapView`) doesn't need to be aware of this. Neither does
the `SwapService` or `SwapActivity` or any other code.
Never ask "Are we searching" but instead receive push notifications
from the peer finder when it stops searching.
Don't worry about receiving the same peer multiple times, it will
automatically get filtered out.
Less concern about doing things in `AsyncTasks` (and knowing what to
do in an `AsyncTask`). The RXJava + RXAndroid libraries deal with this
by allowing the client consuming the `PeerFinder` to specify which
thread to perform the background task on, and also that the found
`Peer`s should be emitted on the UI thread.
In the future, can play with caching the results of a particular
sequence of found peers. However right now using the `Observable.cache()`
method means we can no longer unsubscribe from the peer finders
and thus they run longer than they need to when we move on from
the initial swap screen.
Changing the search query is quite an expensive operation, so this does some rudimentary
checking to see if the two queries are meaningfully different. At present, it trims the
strings and does a case insensitive comparison.
The query is eventually exploded based on whitespace, so leading and trailing white
space is not important. Also, sqlite `LIKE` clauses are case insensitive, so case
is unimportant. Having said that, I'm not sure how someone will be able to change
the queries case without first deleting and then adding characters (thus inducing
meaningfull changse).
This means that instead of building invalid SQL such as `WHERE (() OR ())` it
will build `WHERE((1) OR (1))` which, while non-optimal, is at least valid.
Fixes issue #560.
The fix for the database locking bug was to have a singleton
`DBHelper` instance. This breaks tests because multiple tests
share the same database. The solution is to:
* Hack together a static method which clears the singleton,
then invoke it in the `setUp()` method for relevant test cases.
* Ensure the mock context provided to the providers during
the tests is able to provide a context via `getApplicationContext()`.
Without this, the mock context throws an `UnsupportedOperationException`
when invoking this method.
