Respect order of META-INF/ and META-INF/MANIFEST.MF entries in a JAR file

[michael-o]

The insertion order did not take into account that META-INF/ and/or
META-INF/MANIFEST.MF can or should come first. The code has added it
META-INF/ to directories, thus the actual order of a JAR file was:

META-INF/MANIFEST.MF
META-INF/
META-INF/maven

which does not make sense.

[michael-o]

Found in https://issues.apache.org/jira/projects/MNG/issues/MNG-7357.

[hboutemy]

we need a MJAR issue, because the issue in MNG is triggered by MJAR, with its root cause here in plexus-archiver...

I don't know if we can add a unit test

[hboutemy]

https://issues.apache.org/jira/browse/MJAR-281 created

[michael-o]

I think this issue is everywhere, WAR, shade, assembly, compiler, javadoc, sources and maybe others.

[michael-o]

I think a test is possible since you can use the commons-compress to open a JAR and check for order. I wouldn't be able to create the test next week. Maybe end of. Maybe someone else can provide it?

[plamentotev]

There is existing test that verifies whether the archive contains META-INF/MANIFEST.MF. I've modified it to verify that the order is respected. @michael-o feel free to squash it in single commitif you think it makes more sense.

[michael-o]

There is existing test that verifies whether the archive contains META-INF/MANIFEST.MF. I've modified it to verify that the order is respected. @michael-o feel free to squash it in single commitif you think it makes more sense.

Thank you, the test is simple and does exactly what we need.

[jorsol]

Related: Zlika/reproducible-build-maven-plugin#16

https://github.com/openjdk/jdk/blob/03a8d342b86e720d3cba08d540182b4ab161fba3/src/java.base/share/classes/java/util/jar/JarInputStream.java#L78-L81

The jar tool also adds module-info.class as the third entry but I guess plexus-archiver is not affected since the JarToolModularJarArchiver class calls the jar tool directly.

[michael-o]

It is unfortunate that Oracle failed to officially specify that.