[C++20][Modules] Quote header unit name in preprocessor output (-E) (l…

…lvm#112883)

Summary:
Before this change clang produced output with header unit names that may
conaint path separators, dots and other non-identifier characters. This
diff prints header unit name in quotes and -E output can be compiled
again. Also remove unnecessary space between header unit name and semi.

Test Plan: check-clang

clang/lib/Frontend/PrintPreprocessedOutput.cpp

@@ -952,13 +952,15 @@ static void PrintPreprocessedTokens(Preprocessor &PP, Token &Tok,
       continue;
     } else if (Tok.is(tok::annot_header_unit)) {
       // This is a header-name that has been (effectively) converted into a
-      // module-name.
+      // module-name, print them inside quote.
       // FIXME: The module name could contain non-identifier module name
-      // components. We don't have a good way to round-trip those.
+      // components and OS specific file paths components. We don't have a good
+      // way to round-trip those.
       Module *M = reinterpret_cast<Module *>(Tok.getAnnotationValue());
       std::string Name = M->getFullModuleName();
-      Callbacks->OS->write(Name.data(), Name.size());
-      Callbacks->HandleNewlinesInToken(Name.data(), Name.size());
+      *Callbacks->OS << '"';
+      Callbacks->OS->write_escaped(Name);
+      *Callbacks->OS << '"';
     } else if (Tok.is(tok::annot_embed)) {
       // Manually explode the binary data out to a stream of comma-delimited
       // integer values. If the user passed -dE, that is handled by the

clang/lib/Lex/TokenConcatenation.cpp

@@ -160,6 +160,10 @@ static char GetFirstChar(const Preprocessor &PP, const Token &Tok) {
 bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok,
                                      const Token &PrevTok,
                                      const Token &Tok) const {
+  // No space is required between header unit name in quote and semi.
+  if (PrevTok.is(tok::annot_header_unit) && Tok.is(tok::semi))
+    return false;
+
   // Conservatively assume that every annotation token that has a printable
   // form requires whitespace.
   if (PrevTok.isAnnotation())

clang/test/Headers/header_unit_preprocessed_output.cpp

@@ -0,0 +1,20 @@
+// RUN: rm -fR %t
+// RUN: split-file %s %t
+// RUN: cd %t
+// RUN: %clang_cc1 -verify -std=c++20 -emit-header-unit -xc++-user-header bz0.h
+// RUN: %clang_cc1 -verify -std=c++20 -fmodule-file=bz0.pcm -xc++-user-header bz1.h -E -o bz1.output.h
+// RUN: FileCheck %s < bz1.output.h
+// RUN: %clang_cc1 -std=c++20 -fmodule-file=bz0.pcm -emit-header-unit -xc++-user-header bz1.output.h
+
+//--- bz0.h
+// expected-no-diagnostics
+#pragma once
+
+void foo();
+
+//--- bz1.h
+// expected-no-diagnostics
+import "bz0.h";
+
+// CHECK: # 1 ".{{/|\\\\?}}bz1.h"
+// CHECK: import ".{{/|\\\\?}}bz0.h";

clang/test/Modules/cxx20-include-translation.cpp

@@ -105,6 +105,6 @@ export void charlie() {
 }
 
 // CHECK: #pragma clang module import ".{{/|\\\\?}}h1.h"
-// CHECK: import .{{/|\\\\?}}h2.h
-// CHECK: import .{{/|\\\\?}}h3.h
+// CHECK: import ".{{/|\\\\?}}h2.h"
+// CHECK: import ".{{/|\\\\?}}h3.h"
 // CHECK-NOT: #pragma clang module import ".{{/|\\\\?}}h4.h"