nerzid / autocomment

Generating automated comments using AST and Callgraphs

Automated Comment Generation for JAVA

Type git clone https://github.com/nerzid/autocomment.git in your favorite terminal to download the project.

Feel free to Fork and send Pull Requests 👍

📢 To Do List

• Use a logging library instead of a custom one.
• Gather at least 10.000 java files for training process. Current count Approximately: 2.200.
• Finish all of the javadocs.
• Prepare Wiki section for newcomers.

[Checkout License] (https://github.com/nerzid/autocomment/blob/master/LICENSE)

autocomment's Issues

Prepare Meaningful Sentence Structure and Grammar Rules for Title

This is the continuation of this work #15 . This issue will stay open until the autocomment can handle almost every postag sentence .

Integrating call graph extractor with main app

Current call graph extractor doesn't have a db connection, all info stored locally. But the same info is already stored in db.

So make use of existing db to find the invoked methods.

Interface of call graph extractor:

getListOfInvokedMethods ( classOfCallerMethod, callerMethodSignature)

Outputs:

list of (classOfCalleeMethod, calleeMethodSignature)

Create Custom Exception for isOrdinaryGet and isOrdinarySet Methods

We want to ignore ordinary get/set methods (see #3 ). For logs after training process, we may want to see how many get/set methods we ignored.

Need Custom Exception Class for Tokenizer.simplifyDataType

It may return null or empty string which we don't want to store in db.

Exclude common methods from DB

e.g. equals() , run() , toString(), init() ...

They don't need templates to be explained. All developers know their meanings already.

Hardcode some of the known identifiers' postags

Core NLP Stanford Toolkit recognize those verbs as NP which is incorrect. Im going to hardcode some of the identifiers. Below is the current list that we'll be converted to VB .

• sort
• fire
• copy
• swap
• check
• process
• append
• dump
• print
• println
• register
• resolve
• start
• end
• visit
• fill
• search
• use
• clone

Ignore get/set methods for training process

Get and Set methods (Accessors and Mutators) must be exluded from train proces. Their comment template are already easy to do. Therefore they'r not needed in database.

📢 IMPORTANT

This kind of get() method must be excluded

public int getCount(){
    return count;
}

but following one has some operations in it, therefore it shouldn't be excluded

public int getAverageGradeOfStudents(){
    double sum = 0;
    for(Student s: student_list){
        sum += s.getGrade();
    }
    return (sum / student_list.size());
}

Method Comment Template - Title

The following template can be used to synthesize descriptive comments based on method signature:

VP -->V + NP + [PP]
PP -->Prep + NP
NP --> Noun | Pronoun

VP: Verb phrase
PP: Prepositional Phrase
NP: Noun phrase

IdentifierSplitter splits numbers as well

getInt32() becomes get, int, 32 after splitting operation. It must be get, int32.

Training process ignores punctuations always

Example:

Collection of Listboolean

Where it should be like following:
Collection of List< boolean >

Problem comes from
com.nerzid.autocomment.nlp.Tokenizer.simplifyDataType(String data_type)

Simplify the output of 'data_type'

Simplify the data data_type in db and new data must be stored in a new simplified way.

Example 1

Current output:

java.nio.charset.Charset

Simplified output:

Charset

Example 2

Current output:

java.io.File[]

Simplified output:

Array of File or Collection of File

Example 3

Current output:

java.util.SortedMap<java.lang.String, java.nio.charset.Charset>

Simplified output:

Collection of Charset

Example 4

Current output:

java.util.Set<java.util.Map.Entry<java.lang.String, java.lang.Object>>

Simplified output:

?

Resolve user defined class names

Split the class names into components

For example a class named as
SoftwareEngineering

should be splitted as
Softwareand Engineering.

The results should be stored into another table called UserDefinedType.

Prepare new Database Table Models

As stated in #10 we'll need two tables called Parameter and UserDefinedType. We also need to change existing table WordGroup ( Word renamed to WordGroup 0b6a3c8 ) 's structure.

Trainer makes non-collection datatypes 0D collection

Expectation
int a = 5; to int

Current result
int a = 5; to 0D Collection of int

This bug is easy to solve but im writing here to remember it.

isOrdinaryGet/SetMethod can check convention of get/set methods

Those methods are already doing excellent job to check if get/set methods are ordinary or not #3 . We can use that to implement a brand new convention checker tool for JAVA.

Add .gitignore for spooned files

They shouldn't be on remote repo.

Should we exclude test methods ?

identifier -> testReadSwapped()
splitted_identifier -> test read swapped
postag -> NN VBD VBD

I think test methods are not needed since their identifiers explain the situtation mostly and it would be better if the developer itself writes its javadoc.

Prepare GUI for Autocomment

It may be easier to use GUI instead of running files one by one.