All Downloads are FREE. Search and download functionalities are using the official Maven repository.

com.ibm.ta.sdk.spi.assess.RecommendationJson Maven / Gradle / Ivy

/*
 * (C) Copyright IBM Corp. 2019,2020
 *
 * SPDX-License-Identifier: Apache-2.0
 */

package com.ibm.ta.sdk.spi.assess;

import com.google.gson.annotations.Expose;
import com.ibm.ta.sdk.spi.plugin.TAException;
import com.ibm.ta.sdk.spi.collect.AssessmentUnit;
import com.ibm.ta.sdk.spi.collect.Environment;
import com.ibm.ta.sdk.spi.recommendation.*;
import org.tinylog.Logger;

import java.util.*;

public class RecommendationJson {
  private static final String REC_ATTR_QM_NAME = "name";
  private static final String ASS_ATTR_TARGET = "targets";
  private static final String ASS_ATTR_TARGET_ID = "target";
  private static final String ASS_ATTR_ISSUES = "issues";
  private static final String ASS_ATTR_SUMMARY = "summary";
  private static final String ASS_SUMMARY_COMPLEXITY = "complexity";
  private static final String ASS_SUMMARY_ISSUES = "issues";
  private static final String ASS_SUMMARY_EFFORT = "effort";
  private static final String ASS_SUMMARY_COMPLEXITY_SCORE = "score";

  @Expose
  protected String domain;

  @Expose
  protected String collectionUnitType;

  @Expose
  protected String collectionUnitName;

  @Expose
  protected String middleware;

  @Expose
  protected String version;

  @Expose
  protected List complexityRules;

  @Expose
  protected Map issueCategories;

  @Expose
  protected List> assessmentUnits = new ArrayList>();
  private Recommendation recommendation;

  public RecommendationJson() {
    // Read from Json file
  }

  /**
   * Builds recommendation.json for output
   * @param recommendation Recommendation object to get issues, issue categories and complexity contributions
   * @param environment Environment object for additional attributes to include in the recommendations.json
   * @param auList List of assessment units to include in the recommendations.json
   * @param filterTargets List of target IDs to include in the recommendations.json. If null or empty list, all targets are included.
   * @throws TAException Error building recommendations.json
   */
  public RecommendationJson(Recommendation recommendation, Environment environment, List auList,
                            List filterTargets) throws TAException {
    this.recommendation = recommendation;
    domain = environment.getDomain();
    middleware = environment.getMiddlewareName();
    collectionUnitType = environment.getCollectionUnitType();
    collectionUnitName = environment.getCollectionUnitName();
    version = environment.getMiddlewareVersion();
    complexityRules = ComplexityContributionJson.getComplexityContributionJsonList(recommendation.getComplexityContributions());
    issueCategories = IssueCategoryJson.getIssueCategoryJsonMap(recommendation.getIssueCategories());

    for (AssessmentUnit au : auList) {
      Map auMap = null;
      for (Target target : recommendation.getTargets()) {
        // Check if target is to be included
        if (!isIncludeTarget(target, filterTargets)) {
          Logger.debug("Skipping target ID:" + target.getTargetId());
          continue;
        }

        List auIssues = recommendation.getIssues(target, au);

        // Build map of issues by category
        Map> issuesMap = new LinkedHashMap>();
        for (Issue issue : auIssues) {
          IssueCategory iCat = issue.getCategory();
          if (iCat == null) {
            throw new TAException("No matching issue category found for issue with ID:" + issue.getId());
          }
          String issueCat = iCat.getId();

          List issuesList = issuesMap.get(issueCat);
          if (issuesList == null) {
            issuesList = new ArrayList();
            issuesMap.put(issueCat, issuesList);
          }
          issuesList.add(issue);
        }

        Map auTargetMap = getAssessmentUnit(target, au, issuesMap);
        if (auMap == null) {
          auMap = auTargetMap;
        } else {
          // Add target to existing assessment unit
          List> auTargetList = (List>) auMap.get(ASS_ATTR_TARGET);
          auTargetList.addAll((List>) auTargetMap.get(ASS_ATTR_TARGET));
        }
      }

      // Put minimal info in the assessment unit
      if (auMap == null) {
        auMap = new HashMap<>();
        auMap.put(REC_ATTR_QM_NAME, au.getName());
      }

      // add additional info in recommendation
      Map additionalInfo = au.getAdditionalInfo();
      if (additionalInfo!=null && !additionalInfo.isEmpty()) {
        for (String infoKey : additionalInfo.keySet()) {
          auMap.put(infoKey, additionalInfo.get(infoKey));
        }
      }
      assessmentUnits.add(auMap);
    }
  }

  /*
   * Returns true if the ID of the target is included in the targetIdList
   */
  private boolean isIncludeTarget(Target target, List targetIdList) {
    if (targetIdList == null || targetIdList.isEmpty()) {
      return true;
    }
    return targetIdList.contains(target.getTargetId());
  }

  private Map getAssessmentUnit(Target target, AssessmentUnit au, Map>  issuesMap) {
    Map auMap = new LinkedHashMap();
    auMap.put(REC_ATTR_QM_NAME, au.getName());
    List> targetList = new LinkedList>();
    auMap.put(ASS_ATTR_TARGET, targetList);
    Map targetMap = new LinkedHashMap();
    targetList.add(targetMap);

    targetMap.put(ASS_ATTR_TARGET_ID, target.getTargetId());

    targetMap.put(ASS_ATTR_ISSUES, issuesMap);
    targetMap.put(ASS_ATTR_SUMMARY, getAssessmentSummary(issuesMap));
    return auMap;
  }

  private Map getAssessmentSummary(Map> issues) {
    Map summaryMap = new LinkedHashMap();
    summaryMap.put(ASS_SUMMARY_COMPLEXITY, getSummaryComplexity(issues));
    summaryMap.put(ASS_SUMMARY_ISSUES, getSummaryIssues(issues));
    summaryMap.put(ASS_SUMMARY_EFFORT, getSummaryDevEffort(issues));

    return summaryMap;
  }

  private Map  getSummaryComplexity(Map> issues) {
    Map complexityMap = new LinkedHashMap();

    Iterator itIssueCategory = issues.keySet().iterator();
    while (itIssueCategory.hasNext()) {
      String issueCategory = itIssueCategory.next();
      List categoryIssues = issues.get(issueCategory);
      Logger.trace("Get complexity for category:" + issueCategory + " issues:" + categoryIssues);
      for (Issue catIs : categoryIssues) {
        ComplexityContribution matchedCC = getMatchedComplexityContribution(catIs);
        if (matchedCC != null) {
          catIs.setComplexityContribution(matchedCC);
          String issueComplexity = matchedCC.getComplexity().name();
          Map complexityScoreMap = (Map) complexityMap.get(issueComplexity);
          if (complexityScoreMap == null) {
            complexityScoreMap = new LinkedHashMap();
            complexityMap.put(issueComplexity, complexityScoreMap);
          }
          Integer ccount = complexityScoreMap.get(catIs.getCategory());
          if (ccount == null) {
            ccount = 0;
          }

          complexityScoreMap.put(catIs.getCategory().getId(), ++ccount);
        } else {
          Logger.warn("No complexity rule found for issue:" + catIs.getId());
        }
      }
    }

    addComplexityScore(complexityMap);
    return complexityMap;
  }

  private ComplexityContribution getMatchedComplexityContribution(Issue issue) {
    Logger.trace("Find matching complexity contribution for issue:" + issue);

    ComplexityContribution matchedCC = null;
    for (ComplexityContribution cc : recommendation.getComplexityContributions()) {
      List ruleIssues = cc.getIssues();
      if (ruleIssues != null) {
        for (String ruleIssue : ruleIssues) {
          if (ruleIssue.equals(issue.getId())) {
            Logger.debug("Matching complexity contribution found (matched by ID):" + cc);
            return cc;
          }
        }
      }

      List ruleCategories = cc.getIssuesCategory();
      if (ruleCategories != null) {
        for (String ruleCategory : ruleCategories) {
          if (ruleCategory.equals(issue.getCategory().getId())) {
            Logger.debug("Matching complexity contribution found (matched by category):" + cc);
            if (matchedCC == null) {
              matchedCC = cc;
            } else {
              if (cc.getComplexity().compareTo(matchedCC.getComplexity()) > 0) {
                matchedCC = cc;
              }
            }
            break; // Continue to the next rule
          }
        }
      }
    }
    return matchedCC;
  }

  private void addComplexityScore(Map complexityMap) {
    ComplexityRating maxComplexity = ComplexityRating.simple;

    Iterator itComplexity = complexityMap.keySet().iterator();
    while (itComplexity.hasNext()) {
      ComplexityRating complexity = ComplexityRating.valueOf(itComplexity.next());
      if (maxComplexity == null || complexity.compareTo(maxComplexity) > 0) {
        maxComplexity = complexity;
      }
    }

    complexityMap.put(ASS_SUMMARY_COMPLEXITY_SCORE, maxComplexity.toString());
  }

  private Map  getSummaryIssues(Map>  issues) {
    Map summaryIssuesMap = new LinkedHashMap();

    Iterator itIssueCategory = issues.keySet().iterator();
    while (itIssueCategory.hasNext()) {
      String issueCategory = itIssueCategory.next();
      List categoryIssues = issues.get(issueCategory);
      for (Issue catIs : categoryIssues) {
        String severity = catIs.getSeverity().name();
        Integer sevCount = summaryIssuesMap.get(severity);
        if (sevCount == null) {
          sevCount = 0;
        }
        summaryIssuesMap.put(severity, ++sevCount);
      }
    }

    return summaryIssuesMap;
  }

  private Map  getSummaryDevEffort(Map>  issues) {
    Map effortMap = new LinkedHashMap();
    float totalCost = 0f;

    Iterator itIssueCategory = issues.keySet().iterator();
    while (itIssueCategory.hasNext()) {
      float categoryTotalCost = 0f;

      String issueCategory = itIssueCategory.next();
      List categoryIssues = issues.get(issueCategory);
      for (Issue catIs : categoryIssues) {
        categoryTotalCost += catIs.getCost();
      }

      totalCost += categoryTotalCost;
      effortMap.put(issueCategory, categoryTotalCost);
    }
    effortMap.put("total", totalCost);

    return effortMap;
  }
}




© 2015 - 2024 Weber Informatics LLC | Privacy Policy