[clang] [clang-format][NFC] Make formatting Verilog faster (PR #121139)

[via cfe-commits]

https://github.com/sstwcw created https://github.com/llvm/llvm-project/pull/121139

A regular expression was used in the lexing process.  It made the program take more than linear time with regards to the length of the input.  It looked like the entire buffer could be scanned for every token lexed.  Now the regular expression is replaced with code. Previously it took 20 minutes for the program to format 125 000 lines of code on my computer.  Now it takes 315 milliseconds.

>From e8af860ea0f877862469af5658fa6ad86abc0557 Mon Sep 17 00:00:00 2001
From: sstwcw <su3e8a96kzlver at posteo.net>
Date: Thu, 26 Dec 2024 05:14:42 +0000
Subject: [PATCH] [clang-format][NFC] Make formatting Verilog faster

A regular expression was used in the lexing process.  It made the
program take more than linear time with regards to the length of the
input.  It looked like the entire buffer could be scanned for every
token lexed.  Now the regular expression is replaced with code.
Previously it took 20 minutes for the program to format 125 000 lines of
code on my computer.  Now it takes 315 milliseconds.
---
 clang/lib/Format/FormatTokenLexer.cpp         | 57 ++++++++++++-------
 clang/unittests/Format/TokenAnnotatorTest.cpp | 34 +++++++++++
 2 files changed, 71 insertions(+), 20 deletions(-)

diff --git a/clang/lib/Format/FormatTokenLexer.cpp b/clang/lib/Format/FormatTokenLexer.cpp
index 7a264bddcdfe19..381b8b9451cbba 100644
--- a/clang/lib/Format/FormatTokenLexer.cpp
+++ b/clang/lib/Format/FormatTokenLexer.cpp
@@ -1389,34 +1389,51 @@ FormatToken *FormatTokenLexer::getNextToken() {
 }
 
 bool FormatTokenLexer::readRawTokenVerilogSpecific(Token &Tok) {
+  const char *Start = Lex->getBufferLocation();
+  size_t Len;
+  switch (Start[0]) {
   // In Verilog the quote is not a character literal.
-  //
+  case '\'':
+    Len = 1;
+    break;
   // Make the backtick and double backtick identifiers to match against them
   // more easily.
-  //
-  // In Verilog an escaped identifier starts with backslash and ends with
-  // whitespace. Unless that whitespace is an escaped newline. A backslash can
-  // also begin an escaped newline outside of an escaped identifier. We check
-  // for that outside of the Regex since we can't use negative lookhead
-  // assertions. Simply changing the '*' to '+' breaks stuff as the escaped
-  // identifier may have a length of 0 according to Section A.9.3.
+  case '`':
+    if (Start[1] == '`')
+      Len = 2;
+    else
+      Len = 1;
+    break;
+  // In Verilog an escaped identifier starts with a backslash and ends with
+  // whitespace. Unless that whitespace is an escaped newline.
   // FIXME: If there is an escaped newline in the middle of an escaped
   // identifier, allow for pasting the two lines together, But escaped
   // identifiers usually occur only in generated code anyway.
-  static const llvm::Regex VerilogToken(R"re(^('|``?|\\(\\)re"
-                                        "(\r?\n|\r)|[^[:space:]])*)");
-
-  SmallVector<StringRef, 4> Matches;
-  const char *Start = Lex->getBufferLocation();
-  if (!VerilogToken.match(StringRef(Start, Lex->getBuffer().end() - Start),
-                          &Matches)) {
+  case '\\':
+    // A backslash can also begin an escaped newline outside of an escaped
+    // identifier.
+    if (Start[1] == '\r' || Start[1] == '\n')
+      return false;
+    Len = 1;
+    while (Start[Len] != '\0' && Start[Len] != '\f' && Start[Len] != '\n' &&
+           Start[Len] != '\r' && Start[Len] != '\t' && Start[Len] != '\v' &&
+           Start[Len] != ' ') {
+      // There is a null byte at the end of the buffer, so we don't have to
+      // check whether the next byte is within the buffer.
+      if (Start[Len] == '\\' && Start[Len + 1] == '\r' &&
+          Start[Len + 2] == '\n') {
+        Len += 3;
+      } else if (Start[Len] == '\\' &&
+                 (Start[Len + 1] == '\r' || Start[Len + 1] == '\n')) {
+        Len += 2;
+      } else {
+        Len += 1;
+      }
+    }
+    break;
+  default:
     return false;
   }
-  // There is a null byte at the end of the buffer, so we don't have to check
-  // Start[1] is within the buffer.
-  if (Start[0] == '\\' && (Start[1] == '\r' || Start[1] == '\n'))
-    return false;
-  size_t Len = Matches[0].size();
 
   // The kind has to be an identifier so we can match it against those defined
   // in Keywords. The kind has to be set before the length because the setLength
diff --git a/clang/unittests/Format/TokenAnnotatorTest.cpp b/clang/unittests/Format/TokenAnnotatorTest.cpp
index b2fb5227993c3f..923f6a100d5b39 100644
--- a/clang/unittests/Format/TokenAnnotatorTest.cpp
+++ b/clang/unittests/Format/TokenAnnotatorTest.cpp
@@ -2623,6 +2623,40 @@ TEST_F(TokenAnnotatorTest, UnderstandsVerilogOperators) {
                     "endmodule");
   ASSERT_EQ(Tokens.size(), 11u) << Tokens;
   EXPECT_TOKEN(Tokens[6], tok::l_paren, TT_VerilogMultiLineListLParen);
+
+  // Escaped identifiers.
+  Tokens = Annotate(R"(\busa+index)");
+  ASSERT_EQ(Tokens.size(), 2u) << Tokens;
+  EXPECT_TOKEN(Tokens[0], tok::identifier, TT_Unknown);
+  Tokens = Annotate(R"(\busa+index ;)");
+  ASSERT_EQ(Tokens.size(), 3u) << Tokens;
+  EXPECT_TOKEN(Tokens[0], tok::identifier, TT_Unknown);
+  EXPECT_EQ(Tokens[0]->TokenText, R"(\busa+index)");
+  EXPECT_TOKEN(Tokens[1], tok::semi, TT_Unknown);
+  Tokens = Annotate(R"(\busa+index
+;)");
+  ASSERT_EQ(Tokens.size(), 3u) << Tokens;
+  EXPECT_TOKEN(Tokens[0], tok::identifier, TT_Unknown);
+  EXPECT_TOKEN(Tokens[1], tok::semi, TT_Unknown);
+  // The escaped identifier can be broken by an escaped newline. The result is
+  // still 1 identifier.
+  Tokens = Annotate(R"(\busa+index\
++
+;)");
+  ASSERT_EQ(Tokens.size(), 3u) << Tokens;
+  EXPECT_TOKEN(Tokens[0], tok::identifier, TT_Unknown);
+  EXPECT_EQ(Tokens[0]->TokenText, R"(\busa+index\
++)");
+  EXPECT_TOKEN(Tokens[1], tok::semi, TT_Unknown);
+  // An escaped newline should not be treated as an escaped identifier.
+  Tokens = Annotate("\\\n");
+  ASSERT_EQ(Tokens.size(), 1u) << Tokens;
+  EXPECT_TOKEN(Tokens[0], tok::eof, TT_Unknown);
+  // Macros.
+  Tokens = Annotate("`define x x``x");
+  ASSERT_EQ(Tokens.size(), 7u) << Tokens;
+  EXPECT_TOKEN(Tokens[0], tok::hash, TT_Unknown);
+  EXPECT_TOKEN(Tokens[4], tok::hashhash, TT_Unknown);
 }
 
 TEST_F(TokenAnnotatorTest, UnderstandTableGenTokens) {