Use annotations to configure rules in rules-complexity

[marschwar]

This belongs to #3670 and replaces all configuration kdoc tags in rules-complexity with annotations. There are a few decisions to make:

• Named arguments vs. positional arguments
• Method references vs. lambda

simple properties

private val prop: Boolean by config(defaultValue = false)
// or
private val prop: Boolean by config(false)

fallback properties

private val prop: Int by configWithFallback(fallbackPropertyName = "threshold", defaultValue = 1)
// or 
private val prop: Int by configWithFallback("threshold", 1)

transformed properties

private val prop: Regex by config("$^", String::toRegex)
// or
private val prop: Regex by config("$^") { it.toRegex() }

I am fairly ambivalent about this and I don't know if that is something we even have to agree upon. I am curious about your thoughts.

[codecov]

Codecov Report

Merging #3768 (e866330) into main (21b7d7c) will decrease coverage by 0.06%.
The diff coverage is 98.07%.

@@             Coverage Diff              @@
##               main    #3768      +/-   ##
============================================
- Coverage     83.52%   83.46%   -0.07%     
+ Complexity     2915     2900      -15     
============================================
  Files           452      452              
  Lines          8765     8749      -16     
  Branches       1665     1665              
============================================
- Hits           7321     7302      -19     
- Misses          542      545       +3     
  Partials        902      902              

Impacted Files	Coverage Δ	Complexity Δ
...bosch/detekt/rules/complexity/LabeledExpression.kt	87.50% <66.66%> (ø)	14.00 <0.00> (ø)
...rbosch/detekt/rules/complexity/ComplexCondition.kt	87.80% <100.00%> (ø)	10.00 <1.00> (ø)
...rbosch/detekt/rules/complexity/ComplexInterface.kt	90.62% <100.00%> (-3.13%)	13.00 <0.00> (-1.00)
...rturbosch/detekt/rules/complexity/ComplexMethod.kt	95.23% <100.00%> (ø)	13.00 <1.00> (ø)
...b/arturbosch/detekt/rules/complexity/LargeClass.kt	96.96% <100.00%> (-0.18%)	10.00 <0.00> (-2.00)
...b/arturbosch/detekt/rules/complexity/LongMethod.kt	88.37% <100.00%> (-0.52%)	15.00 <1.00> (-1.00)
...bosch/detekt/rules/complexity/LongParameterList.kt	89.36% <100.00%> (+0.72%)	22.00 <1.00> (-1.00)	⬆️
...bosch/detekt/rules/complexity/MethodOverloading.kt	91.66% <100.00%> (-0.44%)	3.00 <0.00> (-2.00)
...turbosch/detekt/rules/complexity/NamedArguments.kt	83.33% <100.00%> (-1.67%)	13.00 <0.00> (-2.00)
...rbosch/detekt/rules/complexity/NestedBlockDepth.kt	91.66% <100.00%> (-0.34%)	3.00 <0.00> (-2.00)
... and 10 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 21b7d7c...e866330. Read the comment docs.

[schalkms]

Thanks for pushing this initiative forward. This really improves detekt's code base. No misuse of the comments anymore.

I have a few open questions.

• I guess the generator module also needs an update in order to output the correct website sources. Am I right?
• Do we need any other updates to the documentation? I'm thinking about the Contributing.md in order to not mislead new contributors or users. Detekt's docs become outdated quite often. It's sometimes forgotten when introducing changes.

[marschwar]

The generator module and the contribution guide have already been updated.

[schalkms]

The generator module and the contribution guide have already been updated.

Cool. I missed this, unfortunately.

[BraisGabin]

Probably out of scope but I think that inline this values will improve the readability of this code.

[marschwar]

Great point. I was afraid to trigger magic number violations but it seems to work

[BraisGabin]

Can't we use the threshold? The description that you set here is not correct.

[marschwar]

That is a great catch. Unfortunately that is not possible. When threshold is evaluated, the ruleId is used to lookup the config key. In order to get the ruleId, the issue is created. -> infinite loop

How about: "Prefer named arguments for functions to improve readability." This is semi-consistent with rules like ComplexInterface or ComplexCondition where only the general rule is described without mentioning the exact threshold.

[chao2zhang]

This is actually a bug today: Even if the threshold in configuration is 4, the issue.description will always be Function invocation with more than 3 parameters must all be named because the existing description is reading value from the threshold constructor argument.