RunnerClassLoader getResources returns duplicate results

[Postremus]

Describe the bug

I was updating from 1.12.2 to 1.13.2, and noticed that drools is not initializing anymore.

The root cause seems to be this getResources call, which returns the same file paths multiple times:
https://github.com/kiegroup/droolsjbpm-knowledge/blob/master/kie-api/src/main/java/org/kie/api/internal/utils/ServiceDiscoveryImpl.java#L190

When searching for files, the same file path is listed multiple times.

Enumeration<URL> resources = this.getClass().getClassLoader().getResources("META-INF/kie.conf");

This returns 8 results on 1.13.2, but returned only 4 results on 1.12.2.
For example, the first 2 results are:

jar:file:/C:/workspaces/aviator/drools-113/target/quarkus-app/lib/main/org.kie.kie-internal-7.51.0.Final.jar!/META-INF/kie.conf
jar:file:/C:/workspaces/aviator/drools-113/target/quarkus-app/lib/main/org.kie.kie-internal-7.51.0.Final.jar!/META-INF/kie.conf

Which indicates that a duplication is happening somewhere inside the RunnerClassLoader.

This bug only occurs when running as fast-jar. When starting in dev mode or as legacy-jar, only the expected 4 paths are returned.

Expected behavior

getResources should not return duplicate results.

To Reproduce

Steps to reproduce the behavior:

1. Download the reproducer: drools-113.zip
2. mvn clean install
3. cd target/quarkus-app
4. java -jar quarkus-run.jar
5. Visit http://localhost:8080/hello-resteasy. The same files are listed twice.

Environment (please complete the following information):

Output of uname -a or ver

MSYS_NT-10.0 NANB7NLNVP2 2.10.0(0.325/5/3) 2018-06-13 23:34 x86_64 Msys

Output of java -version

GraalVM version (if different from Java)

Quarkus version or git rev:

1.13.0.CR1, 1.13.0.Final, 1.13.1.Final, 1.13.2.Final, 2.0.0.Alpha1

Build tool (ie. output of mvnw --version or gradlew --version)

Apache Maven 3.6.3 (cecedd343002696d0abb50b32b541b8a6ba2883f)
Maven home: C:\eclipse\tools\apache-maven\bin..
Java version: 11.0.7, vendor: Azul Systems, Inc., runtime: C:\eclipse\tools\zulu11.39.15-ca-jdk11.0.7-win_x64
Default locale: de_DE, platform encoding: Cp1252
OS name: "windows 10", version: "10.0", arch: "amd64", family: "windows"

[famod]

/cc @stuartwdouglas @geoand

[geoand]

Interesting, I'll take a look next week

[Postremus]

I tested a bit further, and I think this getResources duplication is caused every time a directory has a non empty sub directory.

This observation also holds true for the drools artifacts. META-INF contains the non empty maven/org.drools/drools-core/ directory, causing the duplication for META-INF/kie.conf.

The following file layout forces the same file path to be returned three times.

this.getClass().getClassLoader().getResources("FIRST/FIRST.FILE")

->
[jar:file:/home/martin/IdeaProjects/drools-113/target/quarkus-app/app/drools-113-1.0.0-SNAPSHOT.jar!/FIRST/FIRST.FILE, 
jar:file:/home/martin/IdeaProjects/drools-113/target/quarkus-app/app/drools-113-1.0.0-SNAPSHOT.jar!/FIRST/FIRST.FILE, 
jar:file:/home/martin/IdeaProjects/drools-113/target/quarkus-app/app/drools-113-1.0.0-SNAPSHOT.jar!/FIRST/FIRST.FILE]

https://github.com/quarkusio/quarkus/blob/main/independent-projects/bootstrap/runner/src/main/java/io/quarkus/bootstrap/runner/SerializedApplication.java#L130

[geoand]

#16782 should take care of it