[infer] fix typo in the computation of the bug hash

Summary: The Nullsafe checker integration is filtering out the pre-existing warnings based on the bug hash only. However, there was a typo in the regexp and the bug hash for methods in anonymous classes was then depending on the name (in the bytecode) of the anonymous class, i.e. depending on the `N` in  `ClassName$N.methodName()` where `N` is the occurrence of the anonymous class in `ClassName`. As a consequence, introducing a new anonymous class in a file was leading to all the reports in the subsequent anonymous classes to be marked as introduced.

Reviewed By: jberdine

Differential Revision: D10186651

fbshipit-source-id: 42e27c132

infer/src/IR/Typ.ml

@@ -1074,7 +1074,7 @@ module Procname = struct
     | Java pname ->
         (* Strip autogenerated anonymous inner class numbers in order to keep the bug hash
            invariant when introducing new annonynous classes *)
-        Str.global_replace (Str.regexp "$[0-9]+") "$_"
+        Str.global_replace (Str.regexp "\\$[0-9]+") "$_"
           (Java.to_string ~withclass:true pname Simple)
     | ObjC_Cpp m when ObjC_Cpp.is_objc_method m ->
         (* In Objective C, the list of parameters is part of the method name. To prevent the bug