Indexer vs DotName.equals() for a nested class with simple name that ends with $

[mkouba]

If there is a nested class with simple name Test$, e.g. org.jboss.jandex.test.IndexerTestCase$Test$ (note that although it's not recommended it's legal to use $ in a class name) then Indexer decodes the DotName as:

componentized	true	
innerClass	true	
local	"" (id=108)	
prefix	DotName  (id=111)	
   componentized	true	
   innerClass	true	
   local	"Test" (id=117)
   prefix	DotName  (id=86)	
      componentized	true	
      innerClass	false	
      local	"test" (id=130)	
      prefix	DotName  (id=131)	
         ...

The toString() of the DotName is org.jboss.jandex.test.IndexerTestCase$Test$ however the equals() method does not match the following:

DotName org = DotName.createComponentized(null, "org");
DotName jboss = DotName.createComponentized(org, "jboss");
DotName jandex = DotName.createComponentized(jboss, "jandex");
DotName test = DotName.createComponentized(jandex, "test");
DotName indexerTestCase = DotName.createComponentized(test, "IndexerTestCase");
DotName testName = DotName.createComponentized(indexerTestCase, Test$.class.getSimpleName(), true);

whose toString() is also org.jboss.jandex.test.IndexerTestCase$Test$.

Maybe I'm missing something obvious.

[mkouba]

CC @Sanne @n1hility

[mkouba]

A test case is available in my branch: https://github.com/mkouba/jandex/tree/issue-97

[n1hility]

thanks will check it out

[ungerts]

At least the test execution is now successful.

[ungerts]

I think a general solution should take into account the InnerClasses Attribute of the class file and iterate through the outer classes which are already parsed.

[msogrin]

Any chance of a fix?
This bug breaks Quarkus compatibility with Scala (at least with Scala 2.13) which has several classes ending with $ in its standard library.

[Ladicek]

I'm looking at the idea of using the InnerClasses attribute and while I agree it would be the best way to do it, I also believe doing that would require a major overhaul of the indexing algorithm. The ClassFile attributes are stored at the very end of the class file, and we [currently] [have to] parse names way before reaching that point. So it's currently not practical, unfortunately.

[mkouba]

Do we really want to close this issue? IIUIC that PR does not solve the problem, or?

[Ladicek]

It does solve the problem -- for a class name that ends with a single $ :-)

[mkouba]

It does solve the problem -- for a class name that ends with a single $ :-)

Ok, how about multiple $? :D

No, I'm fine if you say it does solve the problem. Let's keep the issue closed.

[Ladicek]

It's been a few days, so I don't have it fresh in memory, but if I recall correctly, I couldn't find a proper solution that wouldn't use the InnerClasses attribute, and that would require a huge refactoring (or parsing the class file twice).