Can Faulty Modules Be Predicted by Warning Messages of Static Code Analyzer?

Algorithm 1

Procedure of TOE experiment.

𝑡_FP∶ Threshold of probability to determine FP and NFP

𝑠^𝑝_𝑖∶ Predicted fault status (FP or NFP) of 𝑀_𝑖

for each 𝑀_𝑖 in list of modules sorted by 𝑑_𝑖’s

 prob = fpclassify(𝑀_𝑖)

 if prob > 𝑡_FP then 𝑠^𝑝_𝑖=FP

    else 𝑠^𝑝_𝑖=NFP

 endif

 if 𝑠^𝑎_𝑖≠𝑠^𝑝_𝑖 then fptrain(𝑀_𝑖,𝑠^𝑎_𝑖)

 endif

endfor

fpclassify(𝑀)∶

 if Experiment 1 then

  Generate a set of tokens 𝑇_𝑀 from source code M.

  Calculate probability 𝑃(𝑇_FP∣𝑇_𝑀)

   using corpuses 𝑇_FP and 𝑇_NFP.

  Return 𝑃(𝑇_FP∣𝑇_𝑀).

 if Experiment 2 then

  Generate a set of tokens 𝑇_𝑊

   from warning messages W

   by applying PMD to the source code M.

  Calculate probability 𝑃(𝑇_FP∣𝑇_𝑊)

   using corpuses 𝑇_FP and 𝑇_NFP.

  Return 𝑃(𝑇_FP∣𝑇_𝑊)

fptrain(𝑀,𝑠^𝑎)∶

  if Experiment 1 then

  Generate a set of tokens 𝑇_𝑀 from 𝑀.

  Store tokens 𝑇_𝑀 to the corpus 𝑇𝑠^𝑎.

  if Experiment 2 then

  Generate a set of tokens 𝑇_𝑊 from 𝑊

   by applying PMD to 𝑀.

  Store tokens 𝑇_𝑊 to the corpus 𝑇𝑠^𝑎.