/*
* SPDXVersion: SPDX-1.1
* Person: Person: Nuno Brito (nuno.brito@triplecheck.de)
* Person: Organization: TripleCheck (contact@triplecheck.de)
* Created: 2013-08-29T00:00:00Z
* LicenseName: NOASSERTION
* FileName: FileInfo.java
* FileType: SOURCE
* FileCopyrightText: <text> Copyright 2013 Nuno Brito, TripleCheck </text>
* FileComment: <text> Holds all info collected about each individual
* file of the package </text>
*/
package spdxlib;
import java.util.ArrayList;
public class FileInfo {
public TagValue
tagFileName, // full path and name of the file, mandatory, one
tagFiletype, // kind of file that is referenced here
tagFileChecksumSHA1,
tagLicenseConcluded; // result of audit, mandatory, one
public ArrayList<TagValue>
// list all the licenses collected from the files included
licenseInfoInFile = new ArrayList(); // mandatory, one or many
public TagValue
tagLicenseComments, // other info about the origin, optional, one
tagFileCopyrightText, // copyright text found in file, mandatory, one
tagArtifactOfProjectName, // identify origin of file, optional, one
tagArtifactOfProjectHomePage, // identify origin of file, optional, one
tagArtifactOfProjectURI, // direct link to file, optional, one
tagComment; // generic comment about file, optional, one
// non-standard tags
public TagValue
tagFilePath,
tagFileSize,
tagFileLOC,
tagFileChecksumSHA256,
tagFileChecksumSSDEEP,
tagFileChecksumMD5;
/**
* Prints the data from this object for debugging the result
*/
public void print(){
System.out.println(""
+ "---" + add("Filename", tagFileName)
+ add("LOC", tagFileLOC)
+ add("Size", tagFileSize)
+ "-----------------"
);
}
String add(String title, TagValue tag){
String result = "";
if(tag !=null){
result = title + ":" + tag.toString() + "\n";
}
return result;
}
/**
* Each file can have one or more license applicable.
* This method adds them up as needed.
* @param tag
*/
public void addLicense(TagValue tag){
licenseInfoInFile.add(tag);
}
@Override
public String toString(){
String fileName = getName();
String licenseText = getLicense();
if(licenseText.isEmpty() == false){
fileName += " (" + licenseText + ")";
}
// output the final result
return fileName;
}
/**
* Returns the file name without unwanted chars
* @return the file name of this object
*/
public String getName(){
// get the name in un-processed state (still with slashes)
String fileName = tagFileName.toString();
// remove slashes from the file name
if(fileName.contains("/")){
return fileName.substring(fileName.lastIndexOf("/")+1);
}
// all done
return fileName;
}
/**
* This method will pick on a given FileInfo object and extract the relevant
* licensing details.
* @return A string text with a list of licenses when available. If no
* license was found, it will return null
*/
public String getLicense(){
// where we store the results from this analysis
String licenseOutput = "";
// priority is the case where a human concludes a license
if((this.tagLicenseConcluded != null)
&&(tagLicenseConcluded.toString().equals("NOASSERTION")!=true)){
// no need to proceed
return tagLicenseConcluded.toString();
}
// second priority are licenses detected inside the code
if(licenseInfoInFile.size()>0){
// we need to count the valid licenses
int counterLicense = 0;
// go through all licenses listed
for(TagValue tag : licenseInfoInFile){
// should this license be ignored?
if(tag.toString().equalsIgnoreCase("none")){
continue;
}
// valid license, increase the counter
counterLicense++;
// add the license to the printable text viewed by user
licenseOutput = licenseOutput.concat(", " + tag.toString());
}
// don't display text if no license was accounted
if(counterLicense == 0){
licenseOutput = "";
}else{
// remove the first comma added throught our loop
licenseOutput = licenseOutput.substring(2);
}
}
return licenseOutput;
}
// String licenseOutput = "";
// // priority is the case where a human concludes a license
// if((tagLicenseConcluded != null)
// &&(tagLicenseConcluded.toString().equals("NOASSERTION")!=true)){
// // no need to proceed
// return
// //file.tagFileName.toString()
// //+ " ("+
// tagLicenseConcluded.toString()
// //+")"
// ;
// }
//
// // second priority are licenses detected inside the code
// if(licenseInfoInFile.size()>0){
//
// for(TagValue tag : licenseInfoInFile){
// licenseOutput = licenseOutput.concat(", " + tag.toString());
// }
// // remove the first comma added throught our loop
// licenseOutput = " (" + licenseOutput.substring(2) + ")";
// }
// return licenseOutput;
// }
/**
* When given a file, this method checks if it has either a reported or
* concluded license. The method is necessary since sometimes one of these
* fields is present using infomation like "NOASSERTION" which really
* becomes a problem to normalize.
* @param file A FileInfo object with the information that we will analyze
* @return
*/
public Boolean hasLicense(){
// do we have a license concluded for this file?
if(tagLicenseConcluded != null){
// get the text from the license that was concluded
String licenseText = tagLicenseConcluded.toString();
// a lot of trouble, but necessary to find out what is happening
if( (licenseText.equals("NONE")== false)
&& (licenseText.equals("NOASSERTION")== false)){
// we have at least one license concluded. Good enough
return true;
}
}
// are there any licenses inside this file?
for(TagValue licenseTag : licenseInfoInFile){
String licenseText = licenseTag.toString();
// excluse the cases that shouldn't count for this metric
if(licenseText.equals("NONE")){
continue;
}
if(licenseText.equals("NOASSERTION")){
continue;
}
// we just need to find one license to make our day happy! :-)
return true;
}
// no licenses, let's leave
return false;
}
}