sci_survey/4_function.tex

\section{Function Level}

\subsection{Code Summarization}

Maintaining documentation of source code is an expensive activities in practice, resulting in incomplete and 

\subsection{Code Traceability}
Documentation written in natural language and source code are two of the major artifacts of a software system. 
Tracking a variety of traceability links between software documentation and source code assists software developers in comprehension, efficient development, and effective management of a system.
Automated traceability systems to date have been faced with a major open research challenge: how to extract these links with both high precision and high recall.
% 写一段描述，一般用IR做，主要问题是如何避免假阳性率

Trevor et al.~\cite{savage2010flat3} describe FLAT, a feature location and textual tracing tool.
It is a Eclipse plug-in to help developers to locate source code of the desired features.
FLAT supports textual feature location and dynamic feature location.
The textual feature location via textually searching based on IR technique.
Dynamic feature location entails running the software and invoking the feature of interest to capture a trace of the source code that was executed.
Developers can select one or both location methods to search source code of desired feature.
Nasir et al.~\cite{ali2011trust} propose an approach, Trustrace, inspired by Web trust models to improve the precision and recall of traceability links.
They uses any traditional traceability recovery approach to obtain a set of traceability links, which rankings are then re-evaluated using a set of other traceability recovery approaches.
Then they propose their own approach, Histrace, through CVS/SVN change logs using a Vector Space Model (VSM).
Finally, they combine traditional approach with their own approach, in which they use Histrace as one expert adding knowledge from CVS/SVN change logs.
The results of their approach can significantly improve precision compared with VSM-based approach.
Xiaofan et al.~\cite{chen2011improving} introduce an approach that combines three supporting techniques, Regular Expression, Key Phrases, and Clustering, with a Vector Space Model (VSM) to improve the performance of automated traceability between documents and source code. 
They link code and documentation based on similarity of code and documentation, and combine RE, KP and Clustering to filter out noise links.
They run their approach on 4 projects (JDK1.5, ArgoUML, Freenet, Jmeter), and the results shows the approach combining RE, KP and clustering performs best.
Nasir et al.~\cite{ali2011requirements} present an approach, Coparvo, to complement existing traceability recovery approaches for object-oriented programs.
They use AST to extract identifiers (class name, method names, variable name and comments) and VSM to index requirements and source code.
Then they select high similarity code partitions as experts to vote and score each links, which are used to filter false link. 
The results show that their approach significantly improves the of the recovered links accuracy and also reduces up to 83\% effort required to manually remove false positive links.
Diana et al.~\cite{diaz2013using} propose an approach to recovery of traceability links between use cases and Java classes, TYRION, by leveraging code ownership information to help filter out noise.
They extract author of each source code component and identify the ``context'' for each author.
For each query from the external documentation, they compute the similarity with the context of the authors.
Then a standard IR-based approach are used to retrieving classes, which rewards all the classes developed by the authors having their context most similar to the query.
They evaluate their approach on two software and the results indicate that code ownership information is useful for improving the accuracy.
Dagenais et al.~\cite{dagenais2014using} apply traceability recovery approach to infer documentation patterns, which is helpful for detecting recently-added framework elements that needs to be documented and identifying references to deprecated or deleted code elements that need to be corrected.
They use the fully qualified name (FQN) to identify code-term in documentation, and count the number of types (classes and interfaces) from each package mentioned in the same documentation section to disambiguate false links.
Though identifying links they can recognize documentation patterns, a coherent set of code elements referenced by the documentation, which are used to recommend documentation changes.
They propose AdDoc to automatically discover documentation patterns.
Muslim et al.~\cite{chochlov2017historical} implement a toll, ACIR, which can leverage change-sets of source code for feature location.
They use version control systems (VCSs) to extract change-sets of source code, and use comments in VCSs to help to retrieve code.
They use this tool and study different configurations of the approach on eight subject systems and 600 features.
They results show the importance of rigorously characterizing proposed feature location techniques, to identify their optimal configuration.


\subsection{I/O Generation}

Recent studies by Duala-Ekoko and Robillard~\cite{duala2012asking} and by Sillito et al.~\cite{sillito2008asking} found that when learning API, programmers look for input/output examples, actual values of a function's input/output, in order to understand the function. In particular, the programmers asked ``How does this data structure look at runtime?''~\cite{sillito2008asking} and ``we have a newInstance(String) method that takes a String argument and I have no idea what this String is supposed to be''~\cite{duala2012asking}. 

Facing hard comprehending of extensive and complex computations existing in technical software systems, 
Josef et al.~\cite{pichler2013specification} propose an approach to automatically identify input and output data from source code based on symbolic execution, which is helpful to deduce specifications from given program code.
They use AST to parser programs, executes concrete input values and collect results of a selected function based on symbolic execution.
Hui et al.~\cite{liu2016nomen} presents an empirical study of the lexical similarity between arguments (values passed to a method at a call site) and parameters (the formal parameter in the method's definition) of methods to show names can provide otherwise missing information.
Furthermore, they present an anomaly detection technique that identifies 144 renaming opportunities and incorrect arguments in 14 programs, and a code recommendation system that suggests correct arguments with a precision of 83\%.
Siyuan et al.~\cite{jiang2017docio} implement a prototype toolset, Docio, to generate I/O examples.
Through running tests of the programs and logging the I/O values during execution, they can extract information and generate examples.
They combine their examples with a documentation tools, doxygen~\footnote{\url{http://www.doxygen.org}}.

\begin{table*}[]
\centering
\caption{List of Studies (function level)}
\label{fls}
\vspace{0.5em}
\begin{tabular}{p{0.35\columnwidth}p{0.25\columnwidth}p{0.37\columnwidth}p{0.43\columnwidth}p{0.45\columnwidth}}

\toprule[1.5pt]
Author & Domain & Method & Data Set & Contribution  \\
\midrule[1pt]

Trevor et al.~\cite{savage2010flat3} & Traceability & IR + Execution Trace & & feature location \\

Nasir et al.~\cite{ali2011trust} & Traceability & VSM & Pooka and SIP & traceability links between requirement and source code \\

Xiaofan et al.~\cite{chen2011improving} & Traceability & VSM + RE + KP + Clustering & JDK1.5, ArgoUML, Freenet, Jmeter & traceability links between documents and source code \\

Nasir et al.~\cite{ali2011requirements} & Traceability & AST + VSM & Pooka, SIP, iTrust & traceability links between requirement and source code \\
 
Diana et al.~\cite{diaz2013using} & Traceability & IR + User Information & eTour and SMOS & traceability links between use cases and Java classes \\

Dagenais et al.~\cite{dagenais2014using} & Traceability & FQN & Joda Time, HttpComponents, Hibernate, XStream & link code-term in documentation to source code, recommend documentation changes. \\

Muslim et al.~\cite{chochlov2017historical} & Traceability & VCSs + IR & 8 systems & feature location,  change-sets \\

Josef et al.~\cite{pichler2013specification} & I/O & AST + symbolic execution & & specifications deduction \\

Hui et al.~\cite{liu2016nomen} & I/O & lexical similarity & 14 programs & anomaly detection, arguments recommendation \\

Siyuan et al.~\cite{jiang2017docio} & I/O & I/O logging & 3 API libraries & I/O examples \\
\bottomrule[1.5pt]
\end{tabular}
\vspace{\baselineskip}
\end{table*}