Feature/fix change set applier

[diba1013]

Before submission, please check that ...

• the code follows the Clean Code guidelines.
• the necessary tests are either created or updated.
• all status checks (Travis CI, SonarCloud, etc.) pass.
• your commit history is cleaned up.
• you updated the changelog.
• you updated necessary documentation within retest/docs.

Description

TLDR; Fix that "equal" changes are not accepted together. This brings back the 1-click maintenance.

Bug

This especially noticeable, when testing the same website multiple times (different browsers, screen resolutions, etc.). Because the same site is tested multiple times, the same differences will occur within the different tests. The example below shows a test performed with two different screen resolutions, executed twice to account for the Golden Master creation.

Suite 'TemplateTest' has 6 difference(s) in 4 test(s):
	Test 'large' has 1 difference(s) in 1 state(s):
	page resulted in:
		No Golden Master found. First time test was run? Created new Golden Master, so don't forget to commit...
	Test 'small' has 1 difference(s) in 1 state(s):
	page resulted in:
		No Golden Master found. First time test was run? Created new Golden Master, so don't forget to commit...
	Test 'large' has 3 difference(s) in 1 state(s):
	page resulted in:
		div (div) at 'html[1]/body[1]/div[1]':
			outline:
			  expected="java.awt.Rectangle[x=0,y=0,width=300,height=200]",
			    actual="java.awt.Rectangle[x=0,y=0,width=297,height=198]"
			align-self:
			  expected="",
			    actual="flex-start"
		span (span) at 'html[1]/body[1]/span[1]':
			was deleted
	Test 'small' has 3 difference(s) in 1 state(s):
	page resulted in:
		div (div) at 'html[1]/body[1]/div[1]':
			outline:
			  expected="java.awt.Rectangle[x=0,y=0,width=200,height=100]",
			    actual="java.awt.Rectangle[x=0,y=0,width=187,height=97]"
			align-self:
			  expected="",
			    actual="flex-start"
		span (span) at 'html[1]/body[1]/span[1]':
			was deleted

These changes are present twice and thus are expected to be accepted with a single method call:

div (div) at 'html[1]/body[1]/div[1]':
    align-self:
        expected="",
        actual="flex-start"
span (span) at 'html[1]/body[1]/span[1]':
    was deleted

However, when accepting this within review, each change has to be accepted individually to be counted for the model and accepted into the Golden Master.

Note that review (develop) already uses a similar technique to identify similar changes, which are not propagated properly to the GlobalChangeSetApplier because of this bug.

Workaround

Instead of using the hashCode, use the reduced identifier methods of both IdentifyingAttributes and AttributeDifferences to only take into account important attributes. The former therefore will only take the path, type and suffix attribute, while the latter will only take actual and expected attributes. We might want to change this in the future, but it seems to work for now.

I would consider this more of a workaround, since per definition all IdentifyingAttributes should only hold stable attributes. Therefore I would consider implementing retest/recheck-web#200, but that might be a major breaking change.

Test

To have a test as close to the actual behavior, I introduced an integration test. Since manually putting together a report is cumbersome and we do not have a migration strategy for file based reports to load from, the integration test performs a full recheck lifecycle using the example above to create a report.

Since we are collecting the report within a separate class SuiteAggregator it has to be reset before the test begins to prevent other tests to contribute differences to the report. I am aware that this is not a clean solution as it relies on side effects due to the usage of static singletons. However, alternative solutions would be either introducing a new constructor that takes such a SuiteAggregator or exposing the current test report from RecheckImpl which both then would be public API. If the current solution does not suit you, I would be in favor of exposing the current report as that would be the lowest impact.

Due to the broken down identifier methods, the contents of the various lookup methods cannot be compared anymore. Therefore I decided to only use simple hasSize-checks to test for the validity of the generated lookup tables, as all other possible checks (e.g. accepting changes) seemed to be very cumbersome to properly check.

This has been verified manually that it works with review