Skip to content

Key ideas

Everything in lumioguard CC comes down to a few ideas. This page explains them without code.

Measurements, thresholds and findings

A measurement is a number the tool records, such as "this function has a cognitive complexity of 21".

A threshold is the highest number your team accepts for that measurement, such as 15. You choose it in the configuration file.

A finding is a measurement over its threshold, or a structural problem such as two files that depend on each other. Every finding names the file, and usually the function and line.

Thresholds are your team's choice

There is no universal right number for complexity or length. The defaults are a starting point. They only warn until you decide otherwise.

Blocking and advisory findings

Each rule is either:

  • advisory: it reports findings but never fails the check, or
  • blocking: a new or worse finding fails the check.

You switch a rule with "block": true or false. The label info, warning or error is only a label; it does not decide anything.

By default, the complexity, size, duplication and fan-out rules are advisory. Dependency cycles and architecture boundary violations block.

Comparing with an earlier version

A check on an existing codebase lists every problem it already has. If old problems failed every check, nobody could ever pass. So lumioguard CC compares with a reference:

  • a Git commit or branch, such as your last commit or main, or
  • a stored baseline: a snapshot of results saved in a file.

Each finding is then labelled:

Label Meaning Can fail the check
new Not there before Yes, if the rule blocks
worsened There before, and the number went up Yes, if the rule blocks
existing There before, and not worse No
resolved There before, gone now No 
flowchart LR
    BEFORE["Code before"] --> CMP{"Compare"}
    AFTER["Code now"] --> CMP
    CMP --> OLD["existing<br/>shown, never fails"]
    CMP --> NEW["new or worsened<br/>fails if the rule blocks"]
    CMP --> GONE["resolved<br/>counted as fixed"]

Passed, failed and incomplete

Every check ends with one of three results, and a matching exit code for scripts and CI:

Result Exit code Meaning
PASSED 0 Everything was analyzed, and nothing blocking is new or worse
FAILED 1 Something blocking is new or worse
INCOMPLETE 2 Part of the code could not be analyzed, or the input was invalid

Incomplete always wins. If a file cannot be read, the result is incomplete, even if everything else passed. Code the tool could not see must never look clean.

Where to go next