Another fix for a memory leak. This one is not as big as #4150, but
still worth fixing for memory constrained devices imo.
The `DrawerImageLoader` implementation (a global singleton) references a
member of the `MainActivity`, causing the whole activity to leak.
This weird construct was introduced in #1989 to fix a crash, but I think
since we migrated to coroutines it is no longer necessary because all
calls get correctly cancelled. I tried reproducing the crash but could
not, so I'm pretty sure it is fine. I would appreciate it if someone
else could try it as well though.
(The crash could be reproduced on slow internet, when
`onFetchUserInfoSuccess` was called while the activity was being
destroyed, causing Glide to crash the app because it can't use destroyed
activities. `onFetchUserInfoSuccess` is now no longer called in this
case because it is inside a `lifecycleScope.launch` block.)
Found with Leak canary: The transformation ends up in Glide's memory
cache and leaks whole Activities through the view -> context reference.
This fixes the problem by removing the background detection logic (so
the view reference is no longer needed) and setting the background
directly instead. Looks exactly as before.
Steps to reproduce: Open the dialog to set a catption on an image.
Rotate the screen.
<details>
<summary>Stacktrace</summary>
```
Exception java.lang.NullPointerException: Attempt to invoke virtual method 'void android.widget.TextView.setText(java.lang.CharSequence)' on a null object reference
at com.keylesspalace.tusky.components.compose.dialog.CaptionDialog.onCreateView (CaptionDialog.kt:61)
at androidx.fragment.app.Fragment.performCreateView (Fragment.java:3114)
at androidx.fragment.app.DialogFragment.performCreateView (DialogFragment.java:775)
at androidx.fragment.app.FragmentStateManager.createView (FragmentStateManager.java:557)
at androidx.fragment.app.FragmentStateManager.moveToExpectedState (FragmentStateManager.java:272)
at androidx.fragment.app.FragmentStore.moveToExpectedState (FragmentStore.java:114)
at androidx.fragment.app.FragmentManager.moveToState (FragmentManager.java:1455)
at androidx.fragment.app.FragmentManager.dispatchStateChange (FragmentManager.java:3034)
at androidx.fragment.app.FragmentManager.dispatchActivityCreated (FragmentManager.java:2952)
at androidx.fragment.app.FragmentController.dispatchActivityCreated (FragmentController.java:263)
at androidx.fragment.app.FragmentActivity.onStart (FragmentActivity.java:350)
at androidx.appcompat.app.AppCompatActivity.onStart (AppCompatActivity.java:251)
at android.app.Instrumentation.callActivityOnStart (Instrumentation.java:1543)
at android.app.Activity.performStart (Activity.java:8682)
at android.app.ActivityThread.handleStartActivity (ActivityThread.java:4219)
at android.app.servertransaction.TransactionExecutor.performLifecycleSequence (TransactionExecutor.java:221)
at android.app.servertransaction.TransactionExecutor.cycleToPath (TransactionExecutor.java:201)
at android.app.servertransaction.TransactionExecutor.executeLifecycleState (TransactionExecutor.java:173)
at android.app.servertransaction.TransactionExecutor.execute (TransactionExecutor.java:97)
at android.app.ActivityThread$H.handleMessage (ActivityThread.java:2584)
at android.os.Handler.dispatchMessage (Handler.java:106)
at android.os.Looper.loopOnce (Looper.java:226)
at android.os.Looper.loop (Looper.java:313)
at android.app.ActivityThread.main (ActivityThread.java:8810)
at java.lang.reflect.Method.invoke
at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run (RuntimeInit.java:604)
at com.android.internal.os.ZygoteInit.main (ZygoteInit.java:1067)
```
</details>
Restoring the saved caption after the view was created fixes the
problem.
When viewing a video in Tusky, there is a top toolbar where the
description is shown and the bottom toolbar where play, forward,
backward, and scrub controls are found. In both Tusky 23 and the new
media3 video player code, the logic for showing these toolbars is
*unrelated*; Tusky catches tap events and shows and hides the
description, and the Android media library separately catches tap events
and shows and hides the bottom toolbar. Meanwhile, Tusky and the Android
media library each separately manage a set of logic for auto-hiding
their respective toolbars after a certain number of seconds has passed.
This all results in several problems:
- The top and bottom toolbars can desync, so that one is visible and the
other is not, and tapping to show/hide after this will only swap which
one is visible. This happens *every* time you switch to another
application then back to Tusky while the video player is up.
- You can also desync the top and bottom toolbars in this way by simply
tapping very rapidly.
- The autohide logic was difficult for us to control or customize,
because it was partially hidden inside the Android libraries (relevant
because under media3, the autohide delay increased from 3 to something
like 5 or 6 seconds).
In this patch, I disabled all auto- and tap-based show/hide logic in
media3 and set the Tusky-side show/hide to directly control the media3
toolbar. I then audited the code with printfs until I understood the
state machine of show/hide, and removed anything irrational (some code
was either unreachable, or redundant; either these lines were broken in
the media3 transition, or they never worked).¹
While doing this, I made two policy changes:
- As discussed on Matrix, the autohide delay is now 4 seconds. (In
discussions with users on Mastodon, some complained the previous 3
seconds was too short; but in my opinion and [I think?] charlag's, the
new 5 seconds is too long).
- In the pre-existing code, if the user has hidden the controls, and
they switch to another app and back, the controls display for 4 seconds
then re-hide themselves, just like if the video had been presented for
the first time. I think this is good and kept it— *however* I made a
decision if the user intentionally taps to display the controls, *then*
switches to another app and back, the controls should *not* auto-hide,
because the user most recently requested those controls be shown.
Tests I performed on the final PR (successfully):
- Start video. Expect: toolbar+description hides after 4 seconds.
- Start video. Pause. Resume. Expect: t+d hides after 4 seconds.
- Start video. Wait 4 seconds until t+d hide. Switch to other app.
Switch back. Expect: t+d reappears, then hides after 4 seconds.
- Start video. Wait 4 seconds until t+d hide. Tap to show t+d. Switch to
other app. Switch back. Expect: t+d appear, do NOT autohide.
- Start video. Before 4 seconds up, switch to other app. Switch back.
Expect: t+d reappears, then hides after 4 seconds.
- Start video. Pause. Resume. Before 4 seconds up, switch to other app.
Switch back. Expect: t+d reappears, then hides after 4 seconds.
- Start video. Wait 4 seconds until t+d hide. Tap rapidly over and over
for many seconds. Expect: Nothing weird
- Start *audio*. Expect: At no point does controller autohide, not even
if I switch to another app and back, but I can hide it by manually
tapping
These tests were performed on Android 13. There is an entirely separate
`Build.VERSION.SDK_INT <= 23` path I did not test, but Android Studio
says this is dead code (I think it thinks our minimum SDK is higher than
that?)
---
<small>¹ Incidentally, the underlying cause of #4073 (the show/resume
part of it anyway) turned out to be that the STATE_READY event was being
received not just on video load but also a second time on app resume,
causing certain parts of the initialization code to run a second time
although the fragment had already been fully initialized.</small>
Steps to reproduce:
1. install Gboard
(https://play.google.com/store/apps/details?id=com.google.android.inputmethod.latin)
2. open a direct link to any image in Firefox
3. long-press the image to get a "Copy Image" dialogue (and copy the
image)
4. compose a new post in Tusky
5. Gboard will suggest to paste the image from clipboard
6. paste image, see that when opening alt text editor, it is filled out
with this garbage string
Why is this bad? It's not when I just fix the alt text. But it breaks
every mechanism that is supposed to remind me of adding alt text.
It's hard to argue that this is within scope of Tusky but I also don't
see it getting fixed in Gboard, so here we go.
Fixes: #4063
Switching from an AlertDialog to only a DialogFragment.
I didn't get the AlertDialog to be sized correctly.
It also opens now directly with the right (full screen) size. When the
imageView fails to load (i.e. with an audio file) it will be hidden.
This changes the button layout somewhat.
One observation: The placeholder text "... visually impaired..." is not
quite right as a description for an audio file is not intended for the
visually impaired. But I couldn't think of a better text just yet.
Fixes#2512
Can add an arbitrary number of tabs.
Graphical behavior is unchanged for small numbers: the whole space if
filled with the tabs - they are enlarged if needed.
If there are more the mode switches to "scrollable".
This does not, however, look very differently (see screenshot with the
current tab scrolled out).
---------
Co-authored-by: Konrad Pozniak <connyduck@users.noreply.github.com>
## Issue
Close#3967
# What I did
- Displayed the date of each announcement.
- Date is placed in the lower left corner of the Announcement
- Supported date format internationalization using
getBestDateTimePattern
# Screenshot
<image
src="https://github.com/tuskyapp/Tusky/assets/62137820/7c124183-1a13-4cae-8667-ff82ca99b60c"
width="500"/>
## Note
I am not good at English so I use machine translation a bit. So, you may
find my writing style a little strange...
The idea here is: Everytime we get hold of a new version of a post, we
update everything about that post everywhere.
This makes the distincion between different event types unnecessary, as
everythng is just a `StatusChangedEvent`.
The main benefit is that posts should be up-to-date more often, which is
important considering there is now editing and #3413
This adds support for the new Mastodon 4.2 role badges. Admins can
define if a role should be visible in the interface and then we get it
delivered by the Api on the `Account` object like this:
```
"roles": [
{
"id": "4",
"name": "TEST",
"color": "#ffee00"
}
]
```
- keeps compatibility with older Mastodon version and non Mastodon
servers
- Took me a while, but I figured out a way to use the color and have it
look ok on all backgrounds (Mastodon itself ignores the color and just
always uses its brand color)
- falls back to Tusky blue in case no color is configured
- I adjusted the "Follows you" and "Bot" badges so they match the new
badge style
- In case the "Follows you" and "Bot" badges are visible at the same
time, "Follows you" gets its own line and "Bot" goes into the same line
as the role badge.
- Will work even with a lot of role badges (right now users can only
have 1 role at once though)
- Will work even when the badges federate (right now they don't)
<img
src="https://github.com/tuskyapp/Tusky/assets/10157047/24cbe889-ae46-408e-bfa0-cf3fd3c24f74"
width="320" />
Not quite sure why/when this happens - every stack trace is not our
code, but I do get an ClassNotFoundException for Notification$Type with
the new reverted code.
The notification fetching (worker) then stops/crashes so I never get an
Android notification.
It might have something to do with
https://github.com/tuskyapp/Tusky/pull/3732 ?
Trailing commas on Kotlin sources [has many
advantages](https://pinterest.github.io/ktlint/0.49.0/rules/standard/#trailing-comma-on-call-site):
- It makes version-control diffs cleaner – as all the focus is on the
changed value.
- It makes it easy to add and reorder elements – there is no need to add
or delete the comma if you manipulate elements.
- It simplifies code generation, for example, for object initializers.
The last element can also have a comma.
This PR doesn't go as far as require it, but tweaks KtLint to at least
allow it.
The two `.kt` files prove that the KtLint rules have been properly
disabled.
While helping test an issue with
[Bookwyrm](https://github.com/bookwyrm-social/bookwyrm) I noticed that
the URL formats used by that project aren't checked as possible profile
or post links. They're quite close to a couple of others, so I just
copied close examples and edited a couple of terms.
It's pretty minor, I just used a previous commit as a reference. Let me
know if it needs anything more though. I've only quickly tested it on a
local build with a couple of links against a live Bookwyrm and it picks
them up as expected now.
Konrad Pozniak
mcclure
UlrichKu
Levi Bard
sanao
Markus Unterwaditzer
sanao
Conny Duck
Angelo Suzuki
Mylloon