Issue #400: Prevent side effects / glitches in Observable implementations.

[pocmo]

This is a first version of a helper to fix the issue described in #400.

It's not used in any Observable yet. We may want to use it in SessionManager and Session but probably not EngineSession implementations because observers do not really modify the internal state.

[sputnikci]

[ktlint] WARNING: [no-unit-return] Unnecessary "Unit" return type in column 74

[codecov]

Codecov Report

Merging #453 into master will decrease coverage by 0.88%.
The diff coverage is 25%.

@@             Coverage Diff              @@
##             master     #453      +/-   ##
============================================
- Coverage     76.54%   75.65%   -0.89%     
+ Complexity      915      898      -17     
============================================
  Files           143      137       -6     
  Lines          3376     3282      -94     
  Branches        484      462      -22     
============================================
- Hits           2584     2483     -101     
- Misses          544      560      +16     
+ Partials        248      239       -9

Impacted Files	Coverage Δ	Complexity Δ
...a/components/support/utils/observer/SideEffects.kt	25% <25%> (ø)	4 <4> (?)
...omponents/service/fretboard/ExperimentEvaluator.kt	73.17% <0%> (-12.88%)	26% <0%> (ø)
.../mozilla/components/service/fretboard/Fretboard.kt	61.11% <0%> (-8.89%)	4% <0%> (-6%)
...illa/telemetry/ping/TelemetryEventPingBuilder.java	92.3% <0%> (-7.7%)	4% <0%> (-1%)
...omponents/browser/session/engine/EngineObserver.kt	63.63% <0%> (-5.6%)	8% <0%> (-1%)
...ce/fretboard/source/kinto/KintoExperimentSource.kt	76.47% <0%> (-3.53%)	17% <0%> (+5%)
...retboard/storage/flatfile/ExperimentsSerializer.kt	93.33% <0%> (-1.67%)	5% <0%> (-1%)
...ents/ui/autocomplete/InlineAutocompleteEditText.kt	72.47% <0%> (-1.26%)	43% <0%> (ø)
...src/main/java/org/mozilla/telemetry/Telemetry.java	68.83% <0%> (-0.59%)	14% <0%> (-1%)
.../mozilla/components/browser/search/SearchEngine.kt	100% <0%> (ø)	7% <0%> (-5%)	⬇️
... and 16 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f885ae3...d89db2d. Read the comment docs.

[pocmo]

@csadilek What do you think? If you agree I'll go ahead with the patch and apply it to SessionManager and Session.

[pocmo]

This may cause a concurrent modification exception if we are adding something to the queue while processing the queue. I may add a test for that and fix it. We may want to use an actual queue here.

[csadilek]

hm yes, or we take the recursive approach: #400 (comment) ?

[pocmo]

I'm afraid that a return statement inside the lambda will jump out of this method because it's getting inlined. I thought crossinline would prevent that. But I seem to be able to still use a return@label:

fun someMethod(bar: Foo?) = sideEffects.modifyWithoutSideEffects {
    if (bar == null) {
        // This doesn't compile because of 'crossinline'
        return
    }

    if (bar == null) {
        // This compiles... but isn't it the same as the one above?
        return@modifyWithoutSideEffects
    }

    // ...
}

[pocmo]

I wonder if every action() should actually be wrapped in:

try {
    action()
} finally {
   processQueue()
}

... so that we will always continue with the next action if one of them returns after being inlined?

[pocmo]

This try catch doesn't really work... Every action we process in the queue would need to be wrapped in the same try-catch. We can do that with recursion - but we can't inline a method that is called recursively..

[csadilek]

Interesting. Shouldn't we just use noinline for the lambda here? It shouldn't have the power to jump out of the executing method :).

[pocmo]

I wanted to inline aggressively so that using SideEffects has no penalty. Otherwise Java (6) will need to create an object (+ the class) for every lambda :(

[csadilek]

@pocmo yes, I like the solution. This way the observable can "decide" whether or not it allows side-effects. So, let's get in hardened and landed ;).

[pocmo]

Updated the PR.

This is working nicely with SessionManager now. I just wanted to apply it to Session and then realized that we are using Delegates.observable here. Queuing up the change here isn't as straight-forward.

[pocmo]

Unable to instrument file with Jacoco: /home/travis/build/mozilla-mobile/android-components/components/browser/session/build/tmp/kotlin-classes/debug/mozilla/components/browser/session/SessionManager$removeSessions$$inlined$modifyWithoutSideEffects$1.class

🔥

[pocmo]

Closing: I moved this PR out of the sprint. I want to think a bit more about this.