[PATCH] Re-design the convenience interfaces on MatchFinder.

[Manuel Klimek]

================
Comment at: include/clang/ASTMatchers/ASTMatchFinder.h:137
@@ -136,2 +136,3 @@
 
-  /// \brief Finds all matches on the given \c Node.
+  /// \brief Triggers on all matches on the given \c Node.
+  ///
----------------
Daniel Jasper wrote:
> Triggers what? I presume it is the registered callbacks?
Done.

================
Comment at: include/clang/ASTMatchers/ASTMatchFinder.h:146
@@ +145,3 @@
+  }
+  void match(const clang::ast_type_traits::DynTypedNode &Node,
+             ASTContext &Context);
----------------
Daniel Jasper wrote:
> Should this be in the public interface?
Yes, as with the parent map in ASTContext having a DynTypedNode is actually something that happens...

================
Comment at: include/clang/ASTMatchers/ASTMatchFinder.h:167
@@ -160,1 +166,3 @@
 
+/// \brief Returns the results of matching \c Matcher on \c Node.
+///
----------------
Daniel Jasper wrote:
> How about: "Returns the set of bound nodes for each match of \c Matcher under \c Node."
> 
> The "under" is also used below, but I don't know whether it makes perfectly clear that the matcher is matched on Node and all of its descendants.
It is not. 'match' only matches the node. This is a strict superset of the findAll use case, but an important one - without it, it's impossible to run a matcher against a single node (for example inside a match callback), which is very useful (and the reason I started on this patch).

================
Comment at: include/clang/ASTMatchers/ASTMatchFinder.h:184
@@ +183,3 @@
+template <typename T>
+SmallVector<const T *, 1> findAll(internal::BindableMatcher<T> Matcher,
+                                  const Stmt &Node, ASTContext &Context);
----------------
Daniel Jasper wrote:
> Are you intentionally reusing the name of the methods you deleted above?
> 
> Could this be templated and then work with other node types?
I was, thinking that we might want to keep them around, but after some arguing on IRC with you I think we should leave out all other convenience stuff apart from the match() function, which provides significant value as it allows a different type of control flow for the code using the interface without the need for implementing a callback.

================
Comment at: include/clang/ASTMatchers/ASTMatchFinder.h:219
@@ +218,3 @@
+template <typename T>
+SmallVector<const T *, 1> findAll(internal::BindableMatcher<T> Matcher,
+                                  const Stmt &Node, ASTContext &Context) {
----------------
Daniel Jasper wrote:
> During the experiments I have conducted, it was really convenient to pass in a matcher as well as a string to which stuff is bound in that matcher. I am aware that it is a slightly worse interface, but it gives significantly more freedom. And I think nothing can fail horribly, if nothing gets bound, then nothing is returned.
> 
> So my suggestion is: Add a second method that additionally takes a string "BoundTo" and implement this function here by just calling the other one.
As mentioned, I deleted those methods now.

================
Comment at: unittests/ASTMatchers/ASTMatchersTest.cpp:3482
@@ +3481,3 @@
+public:
+  explicit VerifyMatchOnNode(StringRef Id,
+                             const internal::Matcher<T> &InnerMatcher)
----------------
Daniel Jasper wrote:
> nit: no need for explicit ..
Done.

================
Comment at: unittests/ASTMatchers/ASTMatchersTest.cpp:3500
@@ +3499,3 @@
+TEST(MatchFinder, CanMatchSingleNodesRecursively) {
+  EXPECT_TRUE(matchAndVerifyResultTrue(
+      "class X { class Y {}; };", recordDecl(hasName("::X")).bind("X"),
----------------
Daniel Jasper wrote:
> Independent of whether you change the behavior or not, I'd like a test where the recursive match happens on the node itself. (Probably easiest to use a function declaration for that to not get confused with the generated class name thing).
Now that we're only testing the match() interface, this test already exists ;)


http://llvm-reviews.chandlerc.com/D359