Quantitatively Measuring Cleanliness of C++ Code

Metrics and Symbols

(Metric description : Symbol)

• Ratio of number of lines with more than 80 columns over all lines : R_80C
• Ratio of source files with more than 100 lines orver all files : R_100L
• Nesting depth for each of {}, (), [] and <> : N$ where $ is {}, (), [] or <>
  • Average nesting depth : Avg_N$
  • Maximum nesting depth : Max_N$
  • Ratio of nested blocks with Ndepth$ more than 4 over all nested blocks of the same type : R_4N$
• Length of a sequence that appears X times for X > 1, multiplied by X (within min. edit distance < one third of the length of it) : QX (Edit distance: add/remove/replace will count as 1 character edits)
  • Total QX for sequences with QX > 50 : T_QX50
  • Ratio of T_QX50 over total number of characters in source : R_T_QX50
• Number of member functions in a class : CF
  • Avg_CF
  • Max_CF
  • Ratio of classes with more than 6 member functions over all classes : R_CF6
• Number of list items that are spearated by commas in a list : I
  • Avg_I
  • Max_I
  • Ratio of lists with more than 3 items over all lists : R_I3

Clean Code Characteristics

• R_80C < 0.1%
• R_100L < 5%
• Avg_N$ < 4
• Max_N$ < 6
• R_4N$ < 10%
• R_T_QX50 < 1%
• Avg_CF < 6
• Max_CF < 10
• R_CF6 < 10%
• Avg_I < 3
• Max_I < 6
• R_I3 < 20%

Clean Code Policies

• Every single successful merge into the "main" branches MUST improve the metrics.
  • No merge can make any of the metrics worse.
• If characteristics do not hold (i.e. codebase is not clean), a refactoring merge that improves two of the metrics by at least 50% (only in refactoring diff) must be done before doing a new merge. The refactoring merge needs to modify at least the same number of lines as the new merge.