Add indented raw_ostream class

Class simplifies keeping track of the indentation while emitting. For every new line the current indentation is simply prefixed (if not at start of line, then it just emits as normal). Add a simple Region helper that makes it easy to have the C++ scope match the emitted scope.

Use this in op doc generator and rewrite generator.

This reverts revert commit be185b6 addresses shared lib failure by fixing up cmake files.

Differential Revision: https://reviews.llvm.org/D84107

Author: jpienaar

mlir/include/mlir/Support/IndentedOstream.h

@@ -0,0 +1,102 @@
+//===- IndentedOstream.h - raw ostream wrapper to indent --------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// raw_ostream subclass that keeps track of indentation for textual output
+// where indentation helps readability.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_SUPPORT_INDENTEDOSTREAM_H_
+#define MLIR_SUPPORT_INDENTEDOSTREAM_H_
+
+#include "mlir/Support/LLVM.h"
+#include "llvm/Support/raw_ostream.h"
+
+namespace mlir {
+
+/// raw_ostream subclass that simplifies indention a sequence of code.
+class raw_indented_ostream : public raw_ostream {
+public:
+  explicit raw_indented_ostream(llvm::raw_ostream &os) : os(os) {
+    SetUnbuffered();
+  }
+
+  /// Simple RAII struct to use to indentation around entering/exiting region.
+  struct DelimitedScope {
+    explicit DelimitedScope(raw_indented_ostream &os, StringRef open = "",
+                            StringRef close = "")
+        : os(os), open(open), close(close) {
+      os << open;
+      os.indent();
+    }
+    ~DelimitedScope() {
+      os.unindent();
+      os << close;
+    }
+
+    raw_indented_ostream &os;
+
+  private:
+    llvm::StringRef open, close;
+  };
+
+  /// Returns DelimitedScope.
+  DelimitedScope scope(StringRef open = "", StringRef close = "") {
+    return DelimitedScope(*this, open, close);
+  }
+
+  /// Re-indents by removing the leading whitespace from the first non-empty
+  /// line from every line of the the string, skipping over empty lines at the
+  /// start.
+  raw_indented_ostream &reindent(StringRef str);
+
+  /// Increases the indent and returning this raw_indented_ostream.
+  raw_indented_ostream &indent() {
+    currentIndent += indentSize;
+    return *this;
+  }
+
+  /// Decreases the indent and returning this raw_indented_ostream.
+  raw_indented_ostream &unindent() {
+    currentIndent = std::max(0, currentIndent - indentSize);
+    return *this;
+  }
+
+  /// Emits whitespace and sets the indendation for the stream.
+  raw_indented_ostream &indent(int with) {
+    os.indent(with);
+    atStartOfLine = false;
+    currentIndent = with;
+    return *this;
+  }
+
+private:
+  void write_impl(const char *ptr, size_t size) override;
+
+  /// Return the current position within the stream, not counting the bytes
+  /// currently in the buffer.
+  uint64_t current_pos() const override { return os.tell(); }
+
+  /// Constant indent added/removed.
+  static constexpr int indentSize = 2;
+
+  // Tracker for current indentation.
+  int currentIndent = 0;
+
+  // The leading whitespace of the string being printed, if reindent is used.
+  int leadingWs = 0;
+
+  // Tracks whether at start of line and so indent is required or not.
+  bool atStartOfLine = true;
+
+  // The underlying raw_ostream.
+  raw_ostream &os;
+};
+
+} // namespace mlir
+#endif // MLIR_SUPPORT_INDENTEDOSTREAM_H_

mlir/lib/Support/CMakeLists.txt

@@ -1,5 +1,6 @@
 set(LLVM_OPTIONAL_SOURCES
   FileUtilities.cpp
+  IndentedOstream.cpp
   MlirOptMain.cpp
   StorageUniquer.cpp
   ToolUtilities.cpp
@@ -27,3 +28,12 @@ add_mlir_library(MLIROptLib
   MLIRParser
   MLIRSupport
   )
+
+# This doesn't use add_mlir_library as it is used in mlir-tblgen and else
+# mlir-tblgen ends up depending on mlir-generic-headers.
+add_llvm_library(MLIRSupportIndentedOstream
+  IndentedOstream.cpp
+
+  LINK_LIBS PUBLIC
+  LLVMSupport
+  )

mlir/lib/Support/IndentedOstream.cpp

@@ -0,0 +1,65 @@
+//===- IndentedOstream.cpp - raw ostream wrapper to indent ----------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// raw_ostream subclass that keeps track of indentation for textual output
+// where indentation helps readability.
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Support/IndentedOstream.h"
+
+using namespace mlir;
+
+raw_indented_ostream &mlir::raw_indented_ostream::reindent(StringRef str) {
+  StringRef remaining = str;
+  // Find leading whitespace indent.
+  while (!remaining.empty()) {
+    auto split = remaining.split('\n');
+    size_t indent = split.first.find_first_not_of(" \t");
+    if (indent != StringRef::npos) {
+      leadingWs = indent;
+      break;
+    }
+    remaining = split.second;
+  }
+  // Print, skipping the empty lines.
+  *this << remaining;
+  leadingWs = 0;
+  return *this;
+}
+
+void mlir::raw_indented_ostream::write_impl(const char *ptr, size_t size) {
+  StringRef str(ptr, size);
+  // Print out indented.
+  auto print = [this](StringRef str) {
+    if (atStartOfLine)
+      os.indent(currentIndent) << str.substr(leadingWs);
+    else
+      os << str.substr(leadingWs);
+  };
+
+  while (!str.empty()) {
+    size_t idx = str.find('\n');
+    if (idx == StringRef::npos) {
+      if (!str.substr(leadingWs).empty()) {
+        print(str);
+        atStartOfLine = false;
+      }
+      break;
+    }
+
+    auto split =
+        std::make_pair(str.slice(0, idx), str.slice(idx + 1, StringRef::npos));
+    // Print empty new line without spaces if line only has spaces.
+    if (!split.first.ltrim().empty())
+      print(split.first);
+    os << '\n';
+    atStartOfLine = true;
+    str = split.second;
+  }
+}

mlir/tools/mlir-tblgen/CMakeLists.txt

@@ -25,6 +25,7 @@ add_tablegen(mlir-tblgen MLIR
 set_target_properties(mlir-tblgen PROPERTIES FOLDER "Tablegenning")
 target_link_libraries(mlir-tblgen
   PRIVATE
+  MLIRSupportIndentedOstream
   MLIRTableGen)
 
 mlir_check_all_link_libraries(mlir-tblgen)

mlir/tools/mlir-tblgen/OpDocGen.cpp

@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "DocGenUtilities.h"
+#include "mlir/Support/IndentedOstream.h"
 #include "mlir/TableGen/GenInfo.h"
 #include "mlir/TableGen/Operator.h"
 #include "llvm/ADT/DenseMap.h"
@@ -35,39 +36,8 @@ using mlir::tblgen::Operator;
 // in a way the user wanted but has some additional indenting due to being
 // nested in the op definition.
 void mlir::tblgen::emitDescription(StringRef description, raw_ostream &os) {
-  // Determine the minimum number of spaces in a line.
-  size_t min_indent = -1;
-  StringRef remaining = description;
-  while (!remaining.empty()) {
-    auto split = remaining.split('\n');
-    size_t indent = split.first.find_first_not_of(" \t");
-    if (indent != StringRef::npos)
-      min_indent = std::min(indent, min_indent);
-    remaining = split.second;
-  }
-
-  // Print out the description indented.
-  os << "\n";
-  remaining = description;
-  bool printed = false;
-  while (!remaining.empty()) {
-    auto split = remaining.split('\n');
-    if (split.second.empty()) {
-      // Skip last line with just spaces.
-      if (split.first.ltrim().empty())
-        break;
-    }
-    // Print empty new line without spaces if line only has spaces, unless no
-    // text has been emitted before.
-    if (split.first.ltrim().empty()) {
-      if (printed)
-        os << "\n";
-    } else {
-      os << split.first.substr(min_indent) << "\n";
-      printed = true;
-    }
-    remaining = split.second;
-  }
+  raw_indented_ostream ros(os);
+  ros.reindent(description.rtrim(" \t"));
 }
 
 // Emits `str` with trailing newline if not empty.
@@ -116,7 +86,7 @@ static void emitOpDoc(Operator op, raw_ostream &os) {
 
   // Emit the summary, syntax, and description if present.
   if (op.hasSummary())
-    os << "\n" << op.getSummary() << "\n";
+    os << "\n" << op.getSummary() << "\n\n";
   if (op.hasAssemblyFormat())
     emitAssemblyFormat(op.getOperationName(), op.getAssemblyFormat().trim(),
                        os);
@@ -228,7 +198,7 @@ static void emitDialectDoc(const RecordKeeper &recordKeeper, raw_ostream &os) {
   }
 
   os << "<!-- Autogenerated by mlir-tblgen; don't manually edit -->\n";
-  for (auto dialectWithOps : dialectOps)
+  for (const auto &dialectWithOps : dialectOps)
     emitDialectDoc(dialectWithOps.first, dialectWithOps.second,
                    dialectTypes[dialectWithOps.first], os);
 }