Slidr works by adding its own view in the hierarchy and listening to touch
events in `onInterceptTouchEvent`. Once it detects movement in the correct
direction, it returns `true` and handles all the events itself.
Adding scrollable view detection to Slidr would solve the problem, but it is
not possible and would probably have performance impact.
Fortunately Slidr does not intercept the very first event, which is
ACTION_DOWN, and it reaches scrollable view. So the scrollable view itself can
decide if it should disallow the swipe.
This also has a performance benefit over `OnScrollChangedListener` because
the listener is triggered for every scroll of every view even if the child we
are interested in did not scroll. On the other hand `on(Intercept)TouchEvent`
is triggered only when the view is touched.
There is a possibility that swipe won't be unlocked if view never receives
ACTION_UP or ACTION_CANCEL. However the docs say nothing about the probability
of this happening. Anyways, one possible solution is to post a runnable that
will unlock swipe soon after locking.
Child comment expanding was broken because it did not take into account children
of children of children and deeper levels of comments when calculating new
comment's position.
Replaced with a simple tree to list conversion in pre-order.
* Persist read permission
ACTION_OPEN_DOCUMENT_TREE grants both read and write permissions, but they are
granted only until device reboot unless app persists them.
Once read permission is lost, app cannot check if folders exist in DownloadMediaService.
But it can still create new folders because it has write permission. This results
in duplicate folders.
* Remove unnecessary FLAG_GRANT_WRITE_URI_PERMISSION
This flag is ignored when used with ACTION_OPEN_DOCUMENT_TREE
Implementation is inspired by already existing in Markwon image and link processing
but has to work around some limitations of writing an external plugin.
The first one is storing brackets ourselves. Stored brackets need to be cleared
when a new block starts. Markwon does it in MarkwonInlineProcessor but there is
no callback that we could use. Clearing storage from our own block parser is
unreliable as it is not guaranteed to be called. Instead, every time we need to
access the storage we compare current block with the last used block and clear
storage if necessary.
The second problem is actually a feature of Markwon that it applies spans in
reverse order of calls to MarkwonVisitor#setSpansForNode. This causes other spans
like links and code to be drawn over spoilers making them visible. Adding spans
with a different priority doesn't help as it would require negative priority.
Instead we just remove all the SpoilerSpans from the final string and add them
again, so they are applied last as we want.
* Add todos to places that need more markdown fixes
* Parse spoilers and headings in sidebar
* Assign anonymous MarkwonPlugin to a variable
Prepare code for a future refactoring
* Assign click listener lambda to a variable
Prepare code for a future refactoring
* Add function for creating Markwon with full markdown
All the builders had the same plugins applied to them, except for
BetterLinkMovement. But it is safe to add the plugin as it just adjusts
link interactions.
Also some plugins are now applied in a different order but it doesn't
change anything in this case.
* Add function for creating Markwon with only links support
* Extract UrlMenuBottomSheetFragment creation
* Add functions for creating MarkwonAdapters
* Replace linkify with newInstance for BetterLinkMovementMethod
Because varargs weren't used, the two methods are identical
When the class name is relative, Android tries to resolve it against applicationId.
However it does not match the package because of `.debug` suffix so it tries
to load the wrong class. This results in ClassNotFoundException and a crash.
Using fully qualified class name fixes it as the system can use the class name
as is.
* Prioritize clicks on hidden spoilers over links
Extend BetterLinkMovementMethod to override selection of span that will be
touched and give spoilers and links following priorities:
1. Hidden spoiler
2. Non-spoiler span (i.e. link)
3. Shown spoiler
#609
* Ignore long clicks on spoilers
Ignore long clicks if it is a SpoilerSpan. Also don't apply highlight
because it breaks spoiler effect.
#529
* Rewrite spoiler parsing to properly support nested spoilers and code blocks
Parse all the spoilers, ignoring spoiler brackets that intersect with code
spans. Detect all the spoilers that are nested and mark them accordingly.
Delete all spoiler brackets that were matched. Add SpoilerSpans for non-nested
ranges.
* Simplify offset calculation
* Display comment body the same way as post body when replying
There are two views that support markdown, one of them was used to display post
titles and comments, the other - post bodies. The views are configured differently
even though post and comment bodies should be displayed the same way. Now post
and comment bodies are displayed by the same view.
* Rename extra keys from TEXT to TITLE
Now these extra keys are used only by post title, reflect this in the name.
* Remove markdown support from post title view
* Fix reply styling
Co-authored-by: Docile-Alligator <25734209+Docile-Alligator@users.noreply.github.com>
* Allow parallel installation of debug and release versions
...by adding '.debug' suffix to application id
* add separate app name resource for debug build
Leak found using LeakCanary. Steps:
1. Enable the LeakCanary dependency.
1. Open the app.
1. Go to the "All" tab.
1. Open any post, and go back to the post list.
Leak trace:
```
2022-09-04 17:56:05.904 32018-32018/ml.docilealligator.infinityforreddit.debug D/LeakCanary:
┬───
│ GC Root: System class
│
├─ android.net.ConnectivityManager class
│ Leaking: NO (a class is never leaking)
│ ↓ static ConnectivityManager.sInstance
│ ~~~~~~~~~
├─ android.net.ConnectivityManager instance
│ Leaking: UNKNOWN
│ Retaining 114 B in 5 objects
│ mContext instance of ml.docilealligator.infinityforreddit.activities.ViewPostDetailActivity with mDestroyed = true
│ ↓ ConnectivityManager.mContext
│ ~~~~~~~~
╰→ ml.docilealligator.infinityforreddit.activities.ViewPostDetailActivity instance
Leaking: YES (ObjectWatcher was watching this because ml.docilealligator.infinityforreddit.activities.
ViewPostDetailActivity received Activity#onDestroy() callback and Activity#mDestroyed is true)
Retaining 1.8 MB in 27752 objects
key = 22e99901-9689-4f70-b88c-092a4a7efad9
watchDurationMillis = 5518
retainedDurationMillis = 517
mApplication instance of ml.docilealligator.infinityforreddit.Infinity
mBase instance of androidx.appcompat.view.ContextThemeWrapper
```
Solution based on [this StackOverflow answer](https://stackoverflow.com/a/41431693)
* Update multireddit apply filter text
A "/" is needed at the end of the path in order for the filter to be applied correctly to the multireddit. Also, any characters that are not lowercase will also make it not apply properly.
Leak found using LeakCanary. Steps:
1. Enable the LeakCanary dependency.
1. Open the app.
1. Open any post image, and go back.
According to `Piasy/BigImageViewer` documentation,
[Initialize section](9cc045e814/README.md (initialize)),
the app context must be used to avoid memory leaks.
Leak trace:
```
2022-09-04 19:51:38.154 13332-13332/ml.docilealligator.infinityforreddit.debug D/LeakCanary:
┬───
│ GC Root: Thread object
│
├─ android.os.HandlerThread instance
│ Leaking: NO (PathClassLoader↓ is not leaking)
│ Thread name: 'LeakCanary-Heap-Dump'
│ ↓ Thread.contextClassLoader
├─ dalvik.system.PathClassLoader instance
│ Leaking: NO (BigImageViewer↓ is not leaking and A ClassLoader is never leaking)
│ ↓ ClassLoader.runtimeInternalObjects
├─ java.lang.Object[] array
│ Leaking: NO (BigImageViewer↓ is not leaking)
│ ↓ Object[257]
├─ com.github.piasy.biv.BigImageViewer class
│ Leaking: NO (a class is never leaking)
│ ↓ static BigImageViewer.sInstance
│ ~~~~~~~~~
├─ com.github.piasy.biv.BigImageViewer instance
│ Leaking: UNKNOWN
│ Retaining 969.9 kB in 14812 objects
│ ↓ BigImageViewer.mImageLoader
│ ~~~~~~~~~~~~
├─ com.github.piasy.biv.loader.glide.GlideImageLoader instance
│ Leaking: UNKNOWN
│ Retaining 969.9 kB in 14811 objects
│ ↓ GlideImageLoader.mRequestManager
│ ~~~~~~~~~~~~~~~
├─ com.bumptech.glide.RequestManager instance
│ Leaking: UNKNOWN
│ Retaining 969.9 kB in 14808 objects
│ context instance of ml.docilealligator.infinityforreddit.activities.ViewPostDetailActivity with mDestroyed = true
│ ↓ RequestManager.context
│ ~~~~~~~
╰→ ml.docilealligator.infinityforreddit.activities.ViewPostDetailActivity instance
Leaking: YES (ObjectWatcher was watching this because ml.docilealligator.infinityforreddit.activities.
ViewPostDetailActivity received Activity#onDestroy() callback and Activity#mDestroyed is true)
Retaining 966.2 kB in 14703 objects
key = f69c74cc-521e-4f6c-b5c8-8f787e27df75
watchDurationMillis = 5547
retainedDurationMillis = 541
mApplication instance of ml.docilealligator.infinityforreddit.Infinity
mBase instance of androidx.appcompat.view.ContextThemeWrapper
```
* Copy heading parser and adjust it to match Reddit behavior
Unlike CommonMark, Reddit does not require space after #. This behavior is
coded in a private static function, so the only way to override it is to
copy everything and use the modified copy instead of the default parser.
* Use RedditHeadingPlugin instead of regexes
* Apply plugins to post body when writing a comment
This fixes display when writing comment to a post
that contains spoilers or headings without space
* Apply plugins to parent comment body when writing a comment
This fixes display when replying to a comment that contains strikethrough text
* In anonymous mode, fixed bug in CheckIsFollowingUser.java that does not check if accountName is null or not, which prevents room from retrieving SubscribedUserData to check if the user is followed or not.
Also updated Room version to 2.4.2 to be compatible with M1 macbook.
* Simplified SubscribedUserData accountName null check code
