[clang] [libclang/python] Add CompletionChunkKind enum and deprecate old CompletionChunk.Kind (PR #176631)

[Jannick Kremer via cfe-commits]

================
@@ -3053,46 +3053,27 @@ class _CXUnsavedFile(Structure):
 
 
 class CompletionChunk:
-    class Kind:
-        def __init__(self, name: str):
-            self.name = name
-
-        def __str__(self) -> str:
-            return self.name
-
-        def __repr__(self) -> str:
-            return "<ChunkKind: %s>" % self
+    __kind_id: int
 
     def __init__(self, completionString: CObjP, key: int):
         self.cs = completionString
         self.key = key
-        self.__kindNumberCache = -1
+        self.__kind_id = conf.lib.clang_getCompletionChunkKind(self.cs, self.key)
----------------
DeinAlptraum wrote:

I've tested this by getting all CompletionChunks for all CodeCompletionResults on the first 2000 cursors of `<iostream>`. That's 184595 CCRs and 774559 code completion chunks.

Running this 5 times, I found the `CachedProperty` variant to actually be 1.5% faster, though I guess that's within margin of error anyway. 
That said, note that generating the CCRs and CompletionStrings accounts for a bit more than 90% of execution time, getting the Chunks themselves is fast in comparison so it is a bit difficult to measure the effect here.

https://github.com/llvm/llvm-project/pull/176631