[PATCH] Lex and ignore Microsoft's #pragma warning(...)

[Reid Kleckner]

================
Comment at: include/clang/Basic/DiagnosticLexKinds.td:417
@@ +416,3 @@
+  ExtWarn<"#pragma warning expected 'push', 'pop', 'default', 'disable',"
+          " 'error', 'once', 'supress', 1, 2, 3, or 4">,
+  InGroup<UnknownPragmas>;
----------------
Richard Smith wrote:
> Typo 'supress'... I hope?
Yup.

================
Comment at: lib/Lex/Pragma.cpp:1017-1019
@@ +1016,5 @@
+    return 0;
+  NumericLiteralParser Literal(Spelling, Tok.getLocation(), PP);
+  if (Literal.hadError)
+    return 0;
+  llvm::APInt APVal(32, 0);
----------------
Richard Smith wrote:
> Does MSVC support octal/hex/binary literals?
> 
> I think you should also check for !Literal.hasUDSuffix() and Literal.isIntegerLiteral() here.
Yes, it supports all the builtin ways of writing integer literals.  Floats are out, and user-defined literals are out.

================
Comment at: lib/Lex/Pragma.cpp:1045
@@ +1044,3 @@
+    }
+    PPCallbacks *Callbacks = PP.getPPCallbacks();
+
----------------
Richard Smith wrote:
> Move this earlier or later? It's a bit strange for this to be between lexing the '(' and the next token.
Done.

================
Comment at: lib/Lex/Pragma.cpp:1085-1094
@@ +1084,12 @@
+        // Figure out which warning specifier this is.
+        const char *Specifiers[] = { "1",     "2",       "3",
+                                     "4",     "default", "disable",
+                                     "error", "once",    "suppress" };
+        StringRef Specifier;
+        for (int I = 0, E = llvm::array_lengthof(Specifiers); I != E; ++I) {
+          if (II->getName().equals(Specifiers[I])) {
+            Specifier = Specifiers[I];
+            break;
+          }
+        }
+        if (Specifier.empty()) {
----------------
Richard Smith wrote:
> Maybe use a StringSwitch for this?
Sure, that's easier.

================
Comment at: lib/Lex/Pragma.cpp:1106
@@ +1105,3 @@
+        // Collect the warning ids.
+        SmallVector<int, 1> Ids;
+        PP.Lex(Tok);
----------------
Richard Smith wrote:
> Only 1 inline ID? Cheapskate! =) Maybe 10? 16?
These are usually issued one by one.  :)

================
Comment at: lib/Lex/Pragma.cpp:1126-1129
@@ +1125,6 @@
+
+    if (Tok.isNot(tok::r_paren)) {
+      PP.Diag(Tok, diag::warn_pragma_warning_expected) << ")";
+      return;
+    }
+  }
----------------
Richard Smith wrote:
> Do you need to also check for extra junk after the ')' here?
Done.


http://llvm-reviews.chandlerc.com/D1652