This test seems to be hacking at the binding between Fenix and the
BrowserToolbar to simulate toolbar events passing to the Fenix
interactor.
This is rather clumsy test that relies on the magic working of mockk
instead of following a general unit testing strategy that would commonly
require the class to be re-written to allow for better testing instead.
It is far safer to remove this test since we are not guaranteeing
anything in it and instead we see intermittent failures that make us
lose more time.
So therefore.. 🔥
The expectation is that replacing `return` with `answers` will compute
the return value for the extension function again in order to avoid the
error, "no answer found for: Settings".
Theoretically, this should marginally decrease the duration of our unit
test suite. In my testing, for 1 iteration each (i.e. noise is very
possible), the duration changed from 9m 32s to 8m 21s – a 71s
improvement.
---
To identify tests that were running with robolectric that didn't need to
be, I removed the @RunWith(FenixRobo... from all relevant files:
sed -i '' "/@RunWith(FenixRobolectric/d" app/src/test/**/*.kt
I ran the tests and discovered which ones failed from the Classes tab of
the index.html test result file. Something like:
tests = document.querySelectorAll('table')[3].querySelectorAll('tr');
failureElements = tests.querySelectorAll('.failures');
// TODO: extract the test names
Then I copied these results to a text file and compared them to all the
files that had robolectric test runners to figure out which ones still
pass:
comm -1 -2 failures.txt changed_files.txt > robolectric_not_needed.txt
And undid the changes to the failing files:
for i in $(cat robolectric_not_needed.txt); do git checkout $i; done
Then I removed the import statements on those files:
for i in $(cut changed_files.txt); do sed -i '' "/import.*RunWith/d" $i; done
for i in $(cat changed_files.txt); do sed -i '' "/import.*RobolectricTestRunner/d" $i; done
* Navigate to home on toolbar click. Handle back press from search dialog
Update tests to show home behind search dialog. Remove unused test.
Jump back in show all button is clickable behind search dialog
Recently saved bookmarks show all button is clickable behind search dialog
* Add feature flag
* Past explorations show all button is clickable behind search dialog
Handle keyboard in controllers instead of viewholders. Update tests.
Allow collections to be visible behind search dialog
Dismiss keyboard and search dialog with navigateUp instead of just dismissing the keyboard
Verify navigateUp in tests
Adding ignore for flaky UI test
Only resize home behind search dialog
Add ignore for collection intermittent test
Cleanup
We disabled the allowaccessmodification proguard option because it broke
functionality or crashed the app (I can't rememeber). As far as we know,
the R8 bug was fixed in the R8 bundled with the Android Gradle Plugin
v4.1. We're now on AGP v7.0.0-rc1 so we should be able to revert this
now.
This commit reverts the following commits:
Revert "Proguard/r8: Do not allow access modification."
This reverts commit d2ec7c648856664c27b31831959fd2e83a580968.
Revert "Dump `proguard-android-optimize.txt` into local configuration for later modification"
This reverts commit c543ae338e0bce4a6e2395f3e72742d9c0d65042.
This differs from `tab_view_setting` which tells us what the user's tab
setting is at startup. It does not tell us if the user explicitly
changed it instead of just using the default (which was recently
changed in https://github.com/mozilla-mobile/fenix/issues/19809).
We use make the inactive tabs section of the tabstray collapsible in
this change, with a technical quirk: we want to make the "isExpanded"
state of the tabs stay for the lifetime of the app and not the tabs
tray, but this functionality does not exist.
In this patch, we're storing the UI state in a singleton class that
exists for the lifetime of the app, but a more concrete solution is to
use an AppStore that holds content like this, which we can land in a
future patch.
