[PATCH] D82706: [ASTMatchers] Enhanced support for matchers taking Regex arguments

[Nathan James via Phabricator via llvm-commits]

njames93 marked 7 inline comments as done.
njames93 added inline comments.


================
Comment at: clang/include/clang/ASTMatchers/ASTMatchersInternal.h:1952
 
+std::shared_ptr<llvm::Regex> createAndVerifyRegex(StringRef Regex,
+                                                  llvm::Regex::RegexFlags Flags,
----------------
aaron.ballman wrote:
> Is there a reason this needs to be a shared pointer as opposed to a unique pointer that gets moved into the class owning it?
It has to be shared ownership for the polymorphic matchers to enable reuse of `PolymorphicMatcherWithParam1`.  When a matcher or type `T` is required from a `PolymorphicMatcherWithParam1` it creates a copy of the Param. As you can't copy from a `unique_ptr`, a `shared_ptr` makes sense.


================
Comment at: clang/lib/ASTMatchers/Dynamic/Marshallers.cpp:126
+      Flag |= llvm::Regex::IgnoreCase;
+    else if (OrFlag == "NewLine")
+      Flag |= llvm::Regex::Newline;
----------------
aaron.ballman wrote:
> This makes me think we want to do a case insensitive comparison rather than an exact case match. Personally, I would always write `NewLine` as the string literal, but I could easily see someone writing `Newline` to match the spelling of the RegEx flag and being surprised it doesn't work. WDYT?
That could work, or I implement the `ArgTypeTraits::getBestGuess` to handle small issues like that. Would be a little trickier than the other cases as It would need to support `|`. WDYT?


================
Comment at: clang/lib/ASTMatchers/Dynamic/Marshallers.h:745
+  bool isVariadic() const override { return true; }
+  unsigned getNumArgs() const override { return 0; }
+
----------------
aaron.ballman wrote:
> I may be misunderstanding this, but it seems suspicious that `getNumArgs()` returns 0 but `getArgKinds()` suggests at least one argument. Is that intentional?
It was my understanding that when `isVariadic()` returns true, `getNumArgs()` is not used. I could change it to return 1, but that wouldn't be any more correct than it is now, nor would it have any functional change.


================
Comment at: clang/lib/ASTMatchers/Dynamic/Marshallers.h:770
+          << Args[0].Value.getTypeAsString();
+    }
+    if (Args.size() == 1) {
----------------
aaron.ballman wrote:
> Are you missing a `return` here, after issuing the error?
Yes I am


================
Comment at: clang/unittests/ASTMatchers/Dynamic/ParserTest.cpp:379
+      ParseMatcherWithError(
+          R"query(namedDecl(matchesName("[ABC]*", "IgnoreCase | Basicregex")))query"));
 }
----------------
aaron.ballman wrote:
> Should we have a failure for something like `IgnoreCase | NoFlags` ?
I'm against that, its perfectly legal in non dynamic matchers land to write `matchesName("[ABC]*", IgnoreCase | NoFlags)` Even if it is a little redundant.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82706/new/

https://reviews.llvm.org/D82706