[mlir][docgen] Handle Windows line endings in doc generation.

[scotttodd]

Summary

The printReindented function searches for Unix style line endings (\n), but strings may have Windows style line endings (\r\n). Prior to this change, generated document sections could have extra indentation, which some markdown renderers interpret as code blocks rather than paragraphs.

[jpienaar]

mlir/include/mlir/Support/IndentedOstream.h

117

Could we use rtrim('\\r') on the StringRef of the line?

[scotttodd]

Switch from copy/remove/erase to rtrim.

[scotttodd]

Fix indent in second loop.

[scotttodd]

I'm thoroughly confused by arc and Phabricator, but I updated the change a few times...

mlir/include/mlir/Support/IndentedOstream.h

117

Yep! Here's the code flow on the test case now (note that the output now has "\r\n", but the indentation is correct):

// Unix
"\n\n\n         First line\n                 second line"

// split 1:
""
"\n\n         First line\n                 second line"

// split 2:
""
"\n         First line\n                 second line"

// -----------------------------

// Windows
"\r\n\r\n\r\n         First line\r\n                 second line"

// split 1:
"\r"  // <-- trim this so it's empty string
"\r\n\r\n         First line\r\n                 second line"

[scotttodd]

Squash rebase changes... maybe a single diff in Phabricator now?

[scotttodd]

I don't think I have write access to land this myself. If I did, I'd probably wait until Monday... but I wouldn't object to someone merging earlier than that.

[scotttodd]

Okay, we're back in typical business hours now. Could you / someone merge this for me? I can also see about getting commit access myself.

mlir/include/mlir/Support/IndentedOstream.h

private:
   /// The underlying raw_ostream.
   raw_ostream &os;
 };
 
 inline raw_indented_ostream &
 mlir::raw_indented_ostream::printReindented(StringRef str,
                                             StringRef extraPrefix) {
   StringRef output = str;
   // Skip empty lines.
   while (!output.empty()) {
     auto split = output.split('\n');
-    size_t indent = split.first.find_first_not_of(" \t");
+    // Trim Windows \r characters from \r\n line endings.
+    auto firstTrimmed = split.first.rtrim('\r');
+    size_t indent = firstTrimmed.find_first_not_of(" \t");
     if (indent != StringRef::npos) {
       // Set an initial value.
       leadingWs = indent;
       break;
     }
     output = split.second;
   }
   // Determine the maximum indent.
   StringRef remaining = output;
   while (!remaining.empty()) {
     auto split = remaining.split('\n');
-    size_t indent = split.first.find_first_not_of(" \t");
+    auto firstTrimmed = split.first.rtrim('\r');
+    size_t indent = firstTrimmed.find_first_not_of(" \t");
     if (indent != StringRef::npos)
       leadingWs = std::min(leadingWs, static_cast<int>(indent));
     remaining = split.second;
   }
   // Print, skipping the empty lines.
   std::swap(currentExtraPrefix, extraPrefix);
   *this << output;
   std::swap(currentExtraPrefix, extraPrefix);

mlir/unittests/Support/IndentedOstreamTest.cpp

TEST(FormatTest, Reindent) {
   const auto *expected =
       R"(First line
         second line
 
 
 )";
   EXPECT_THAT(os.str(), StrEq(expected));
 }
+
+TEST(FormatTest, ReindentLineEndings) {
+  std::string str;
+  llvm::raw_string_ostream os(str);
+  raw_indented_ostream ros(os);
+
+  // Similar string as the previous test, but with \r\n (Windows style) line
+  // breaks. Note that C++'s internal string representation uses \n, so just
+  // running the previous test as-is on Windows is not sufficient.
+  const auto *desc =
+      "\r\n\r\n\r\n         First line\r\n                 second line";
+  ros.printReindented(desc);
+  ros.flush();
+  const auto *expected = "First line\r\n        second line";
+  EXPECT_THAT(os.str(), StrEq(expected));
+}