AvoidCommonTypeNames false positives when using --release

[jeffrey-easyesi]

Using Error Prone 2.50 and compiling with the --release option, the AvoidCommonTypeNames checker is sometimes including java.lang.X.Y classes, when really there is no reason to warn about clashes with those since they are not imported automatically with their undotted name.

Example (using Java 25 with --release 25):

Info.java:1: warning: [AvoidCommonTypeNames] Info clashes with java.lang.ProcessHandle.Info
class Info {}
^
    (see https://errorprone.info/bugpattern/AvoidCommonTypeNames)

I think this was introduced by https://github.com/google/error-prone/commit/924cb63735c9e6e44b0362b8ed9fbcfdf1dc8d9d/. The getSymbols call sometimes includes inner classes that getSymbolsByName does not, varying on both the Java version and the --release version:

Java version	--release version	Classes
21	8	Character.UnicodeScript
25	8	Character.UnicodeScript
21	11	StackWalker.StackFrame
25	11	none
21	17	StackWalker.StackFrame, Thread.State, Character.UnicodeScript
25	17	System.LoggerFinder, Character.UnicodeScript
21	21	none
25	21	System.LoggerFinder, ProcessHandle.Info
25	25	ProcessHandle.Info

[cpovirk]

Interesting, thanks. I wonder if Symtab is not quite the right tool for this job, similar to how we found that it might not have been the right tool for the job in #5791. I think we've worried before that alternatives can be expensive, but—assuming that there is a proper alternative here, as seems especially likely for a Java 9+ (modular) compilation—I'd hope we don't have to worry about expense because of the caching of the result. Or we just hard-code the list of java.lang classes and write a test to nudge us to update it when new ones are added....

[AzazelSensei]

Opened #6023