[MSHARED-307] Performance: The implementation of DefaultClassAnalyzer unnecessarily scans jar files bodies

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: maven-dependency-analyzer-1.4
• Fix Version/s: maven-dependency-analyzer-1.5
• Component/s: maven-dependency-analyzer
• Assignee: Olivier Lamy
• Reporter: Victor Bronstein
• Created: 24/Nov/13 06:20
• Updated: 29/Oct/14 11:56
• Resolved: 28/Apr/14 18:40

Description

The contract of the method is to provide a list of classes given a URL. In case of JAR files it would be enough to use JarFile#entries. However, since the implementation delegates to ClassFileVisitorUtils using an implementation of the visitor, the JAR files are being read completely, including the streams of the individual classes.
This is a considerable performance hit. Specifically, in our project we have vast amount of Maven modules and do a dependency validation for them. By replacing the current implementation with one that uses JarFile#getEntries we were able to reduce the build time by 4 minutes.

Issue Links

• is related to: MSHARED-361 DefaultProjectDependencyAnalyzer.buildArtifactClassMap assumes dependencies are jar files (regression)
• is related to: MSHARED-382 JarFile object is not being closed