[wasm][asm.js] Adding custom asm.js lexer.

src/asmjs/asm-lexer.cc

+// Copyright 2017 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/asmjs/asm-lexer.h"
+
+#include <stdlib.h>
+
+#include "src/objects.h"

[marja, 2017/03/15 12:34:49]

Why is objects.h needed?

[bradn, 2017/03/16 00:21:46]

There was a Handle<String> used in scanner.h inline.
Moved into the .cc file.

+#include "src/parsing/scanner-character-streams.h"

[vogelheim, 2017/03/15 12:07:41]

I don't see scanner-character-streams.h being used. Am I missing something?

[bradn, 2017/03/16 00:21:47]

Dropped.

+#include "src/parsing/scanner.h"

[marja, 2017/03/15 12:34:49]

Hmm, you're still including scanner.h even though my previous comment about it
was marked done.

[bradn, 2017/03/16 00:21:47]

That was in the header.
This is needed here because the Utf16CharacterStream's declaration is in
scanner.h

[marja, 2017/03/16 17:05:33]

My orig. comment suggested moving the streams out of scanner.h. It's weird that
a complete separate scanner needs to include scanner.h. But as I can see this CL
is close to landing, would doing a follow-up be OK?

+
+namespace v8 {
+namespace internal {
+
+namespace {
+// Cap number of identifiers to ensure we can assign both global and
+// local ones a token id in the range of an int32_t.
+static const int kMaxIdentifierCount = 0xf000000;
+};
+
+AsmJsLexer::AsmJsLexer()
+    : token_(0),
+      preceding_token_(0),
+      next_token_(0),
+      rewind_(false),
+      in_local_scope_(false),
+      global_count_(0),
+      double_value_(0.0),
+      unsigned_value_(0),
+      preceded_by_newline_(false) {
+#define V(name, _junk1, _junk2, _junk3) property_names_[#name] = kToken_##name;
+  STDLIB_MATH_FUNCTION_LIST(V)
+  STDLIB_ARRAY_TYPE_LIST(V)
+#undef V
+#define V(name) property_names_[#name] = kToken_##name;
+  STDLIB_MATH_VALUE_LIST(V)
+  STDLIB_OTHER_LIST(V)
+#undef V
+#define V(name) global_names_[#name] = kToken_##name;
+  KEYWORD_NAME_LIST(V)
+#undef V
+}
+
+void AsmJsLexer::SetStream(std::unique_ptr<Utf16CharacterStream> stream) {
+  stream_ = std::move(stream);
+  Next();
+}
+
+void AsmJsLexer::Next() {

[vogelheim, 2017/03/15 12:07:40]

I find this method much nicer to read now. Thanks.

[bradn, 2017/03/16 00:21:47]

:-)

+  if (rewind_) {
+    preceding_token_ = token_;
+    token_ = next_token_;
+    next_token_ = 0;
+    rewind_ = false;
+    return;
+  }
+
+  if (token_ == kEndOfInput || token_ == kParseError) {
+    return;
+  }
+
+#if DEBUG
+  if (FLAG_trace_asm_lexer) {
+    if (Token() != 0) {

[vogelheim, 2017/03/15 12:07:41]

nitpick: No real problem here, but this logic is a little weird. I think I'd
either just handle a 0 token in Name, or just have the last else being an "else
if (Token() != 0)".

(Maybe Name(0) should be "{uninitialized}". I don't think 0 is a valid unicode
code point.)

[bradn, 2017/03/16 00:21:47]

Done.

+      if (Token() == kDouble) {
+        PrintF("%lf ", AsDouble());
+      } else if (Token() == kUnsigned) {
+        PrintF("%" PRIu64 " ", AsUnsigned());
+      } else {
+        std::string name = Name(Token());
+        PrintF("%s ", name.c_str());
+      }
+    }
+  }
+#endif
+
+  preceded_by_newline_ = false;
+  preceding_token_ = token_;
+  for (;;) {
+    uc32 ch = stream_->Advance();
+    switch (ch) {
+      case ' ':
+      case '\t':
+      case '\n':
+      case '\r':
+        // Ignore whitespace, track when we've passed a newline for optional
+        // semicolon support.
+        if (ch == '\n') {

[vogelheim, 2017/03/15 12:07:41]

nitpick: This is weird. If you have a switch-case anyhow, and if you want to
treat one branch separately, why not just have a separate case for it?

  case '\n':
    preceded_...
    break;
  case ' ':
  case '\t':
  ....

[bradn, 2017/03/16 00:21:46]

Hah, yeah good point (missed that in the refactor).
Done.

+          preceded_by_newline_ = true;
+        }
+        break;
+
+      case kEndOfInput:
+        token_ = kEndOfInput;
+        return;
+
+      case '\'':
+      case '"':
+        ConsumeString(ch);
+        return;
+
+      case '/':
+        ch = stream_->Advance();
+        if (ch == '/') {
+          ConsumeCComment();
+        } else if (ch == '*') {
+          ConsumeCPPComment();
+        } else {
+          stream_->Back();
+          token_ = '/';
+          return;
+        }
+        // Breaks out of switch, but loops again (i.e. the case when we parsed
+        // a comment, but need to continue to look for the next token).
+        break;
+
+      case '<':
+      case '>':
+      case '=':
+      case '!':
+        ConsumeCompareOrShift(ch);
+        return;
+
+      default:
+        if (IsIdentifierStart(ch)) {
+          ConsumeIdentifier(ch);
+        } else if (IsNumberStart(ch)) {
+          ConsumeNumber(ch);
+        } else if (ch >= 32 && ch < 127) {

[vogelheim, 2017/03/15 12:07:41]

[Not sure this is an issue, but... ]

How many of these are legitimate tokens anyhow? Alphabetic chars + numeric ones
are handles above, as are some operators. So these would be some legitimate
tokens (+, -, *, square + curly + round braces), but some others would just be
syntax errors (#)?

[bradn, 2017/03/16 00:21:47]

Listed out the single char ones.

+          // Use fixed token IDs for ASCII.
+          token_ = ch;
+        } else {
+          // TODO(bradnelson): Support unicode (probably via UnicodeCache).
+          token_ = kParseError;
+        }
+        return;
+    }
+  }
+}
+
+void AsmJsLexer::Rewind() {
+  DCHECK(!rewind_);
+  next_token_ = token_;
+  token_ = preceding_token_;
+  preceding_token_ = 0;
+  rewind_ = true;
+  preceded_by_newline_ = false;
+  identifier_string_.clear();
+}
+
+void AsmJsLexer::ResetLocals() { local_names_.clear(); }
+
+#if DEBUG
+// Only used for debugging.
+std::string AsmJsLexer::Name(token_t token) const {
+  // TODO(bradnelson): Make thread safe.
+  if (token >= 32 && token < 127) {
+    return std::string(1, static_cast<char>(token));
+  }
+  for (auto& i : local_names_) {
+    if (i.second == token) {
+      return i.first.c_str();

[Karl, 2017/03/15 15:04:13]

Why not just:

    return i.first;

[bradn, 2017/03/16 00:21:46]

Done.

+    }
+  }
+  for (auto& i : global_names_) {
+    if (i.second == token) {
+      return i.first.c_str();

[Karl, 2017/03/15 15:04:13]

Same here.

[bradn, 2017/03/16 00:21:47]

Done.

+    }
+  }
+  for (auto& i : property_names_) {
+    if (i.second == token) {
+      return i.first.c_str();

[Karl, 2017/03/15 15:04:13]

Same here.

[bradn, 2017/03/16 00:21:46]

Done.

+    }
+  }
+  switch (token) {
+#define V(rawname, name) \
+  case kToken_##name:    \
+    return rawname;
+    LONG_SYMBOL_NAME_LIST(V)
+#undef V
+    default:
+      break;
+  }
+  if (token == kUnsigned) {

[vogelheim, 2017/03/15 12:07:40]

Why not handle all of these inside the switch right above?

[bradn, 2017/03/16 00:21:46]

Done.

+    return "{unsigned value}";
+  } else if (token == kDouble) {
+    return "{double value}";
+  } else if (token == kParseError) {
+    return "{parse error}";
+  } else if (token == kEndOfInput) {
+    return "{end of input}";
+  }
+  UNREACHABLE();
+  return "{unreachable}";
+}
+#endif
+
+int AsmJsLexer::GetPosition() const { return static_cast<int>(stream_->pos()); }

[vogelheim, 2017/03/15 12:07:40]

Does this work if rewind_ is set? If not, maybe add DCHECK(!rewind_).

[bradn, 2017/03/16 00:21:46]

Done.

+
+void AsmJsLexer::Seek(int pos) {
+  stream_->Seek(pos);
+  preceding_token_ = 0;
+  token_ = 0;
+  next_token_ = 0;
+  rewind_ = false;
+  Next();
+}
+
+void AsmJsLexer::ConsumeIdentifier(uc32 ch) {
+  // Consume characters while still part of the identifier.
+  identifier_string_ = "";

[vogelheim, 2017/03/15 12:07:41]

identifier_string_.clear();

(STL is bizarre, but.. The way you've written it is very clear to read, but
unfortunately will actually call a copy operation, and will iterate over the
empty input string, only to figure out that it's empty. Not sure if compilers
are being clever about it, but .clear() always does the right thing.)

[bradn, 2017/03/16 00:21:46]

Yep. Done.

+  while (IsIdentifierPart(ch)) {
+    identifier_string_ += ch;
+    ch = stream_->Advance();
+  }
+  // Go back one for next time.
+  stream_->Back();
+
+  // Decode what the identifier means.
+  if (preceding_token_ == '.') {
+    auto i = property_names_.find(identifier_string_);
+    if (i != property_names_.end()) {
+      token_ = i->second;
+      return;
+    }
+  } else {
+    {
+      auto i = local_names_.find(identifier_string_);
+      if (i != local_names_.end()) {
+        token_ = i->second;
+        return;
+      }
+    }
+    if (!in_local_scope_) {
+      auto i = global_names_.find(identifier_string_);
+      if (i != global_names_.end()) {
+        token_ = i->second;
+        return;
+      }
+    }
+  }
+  if (preceding_token_ == '.') {
+    CHECK(global_count_ < kMaxIdentifierCount);
+    token_ = kGlobalsStart + global_count_++;
+    property_names_[identifier_string_] = token_;
+  } else if (in_local_scope_) {
+    CHECK(local_names_.size() < kMaxIdentifierCount);
+    token_ = kLocalsStart - static_cast<token_t>(local_names_.size());
+    local_names_[identifier_string_] = token_;
+  } else {
+    CHECK(global_count_ < kMaxIdentifierCount);
+    token_ = kGlobalsStart + global_count_++;
+    global_names_[identifier_string_] = token_;
+  }
+}
+
+void AsmJsLexer::ConsumeNumber(uc32 ch) {
+  std::string number;
+  number = ch;
+  bool has_dot = ch == '.';
+  for (;;) {
+    ch = stream_->Advance();
+    if ((ch >= '0' && ch <= '9') || (ch >= 'a' && ch <= 'f') ||
+        (ch >= 'A' && ch <= 'F') || ch == '.' || ch == 'x' ||
+        ((ch == '-' || ch == '+') && (number[number.size() - 1] == 'e' ||
+                                      number[number.size() - 1] == 'E'))) {
+      // TODO(bradnelson): Test weird cases ending in -.
+      if (ch == '.') {
+        has_dot = true;
+      }
+      number += ch;
+    } else {
+      break;
+    }
+  }
+  stream_->Back();
+  // Special case the most common number.
+  if (number == "0") {
+    unsigned_value_ = 0;
+    token_ = kUnsigned;
+    return;
+  }
+  // Pick out dot.
+  if (number == ".") {
+    token_ = '.';
+    return;
+  }
+  // Decode numbers.
+  // TODO(bradnelson): Replace strto* with shared code with scanner.cc
+  char* end;
+  if (has_dot) {
+    double_value_ = strtod(number.c_str(), &end);
+    token_ = kDouble;
+  } else {
+    if (number.size() > 2 && number[0] == '0' && number[1] == 'x') {
+      // Decode 0x* as hex.
+      unsigned_value_ = strtoul(number.c_str() + 2, &end, 16);
+    } else if (number.size() > 1 && number[0] == '0') {
+      // Decode 0* as octal.
+      unsigned_value_ = strtoul(number.c_str() + 1, &end, 8);
+    } else {
+      // Decode the rest as double.
+      // This can come up in asm.js as for example 1e2 is used to encode 100.
+      double_value_ = strtod(number.c_str(), &end);
+      unsigned_value_ = static_cast<uint32_t>(double_value_);
+    }
+    token_ = kUnsigned;
+  }
+  // Check if string to number conversion didn't consume all the characters.
+  // This happens if the character filter let through something invalid
+  // like: 0123ef for example.
+  // TODO(bradnelson): Check if this happens often enough to be a perf problem.
+  if (end != number.c_str() + number.size()) {
+    // If things didn't parse fully, but start with a '.', back out the other
+    // characters and emit the '.' token.
+    if (number[0] == '.') {
+      for (size_t k = 1; k < number.size(); ++k) {
+        stream_->Back();
+      }
+      token_ = '.';
+      return;
+    }
+    // Anything else that doesn't parse is an error.
+    token_ = kParseError;
+    return;
+  }
+}
+
+void AsmJsLexer::ConsumeCComment() {
+  for (;;) {
+    uc32 ch = stream_->Advance();
+    if (ch == '\n' || ch == kEndOfInput) {
+      break;
+    }
+  }
+}
+
+void AsmJsLexer::ConsumeCPPComment() {
+  for (;;) {
+    uc32 ch = stream_->Advance();
+    if (ch == '*') {

[vogelheim, 2017/03/15 12:07:41]

Your choice, but I think this if-branch would be a bit more compact if replaced
with this:

while (ch == '*') {
  ch = stream_->Advance();
  if (ch == '/' || ch == kEndOfInput)
    return;
}

[bradn, 2017/03/16 00:21:46]

Ah, yeah, that's better.
Done.

+      ch = stream_->Advance();
+      if (ch == '/') {
+        break;
+      }
+      if (ch == '*') {
+        stream_->Back();
+      }
+    } else if (ch == kEndOfInput) {

[vogelheim, 2017/03/15 12:07:40]

I think this potentially swallows a syntax error w/ an unclosed comment at the
end of a file. (E.g. xxx /* yyy <EOF>)

[bradn, 2017/03/16 00:21:46]

Ah, yes.
Fixed and added a test.

+      break;
+    }
+  }
+}
+
+void AsmJsLexer::ConsumeString(uc32 quote) {
+  // Only string allowed is 'use asm' / "use asm".
+  const char* expected = "use asm";
+  for (; *expected != '\0'; ++expected) {
+    if (stream_->Advance() != *expected) {
+      token_ = kParseError;
+      return;
+    }
+  }
+  if (stream_->Advance() != quote) {
+    token_ = kParseError;
+    return;
+  }
+  token_ = kToken_UseAsm;
+}
+
+void AsmJsLexer::ConsumeCompareOrShift(uc32 ch) {
+  uc32 next_ch = stream_->Advance();
+  if (next_ch == '=') {
+    switch (ch) {
+      case '<':
+        token_ = kToken_LE;
+        break;
+      case '>':
+        token_ = kToken_GE;
+        break;
+      case '=':
+        token_ = kToken_EQ;
+        break;
+      case '!':
+        token_ = kToken_NE;
+        break;
+      default:
+        UNREACHABLE();
+    }
+  } else if (ch == '<' && next_ch == '<') {
+    token_ = kToken_SHL;
+  } else if (ch == '>' && next_ch == '>') {
+    if (stream_->Advance() == '>') {
+      token_ = kToken_SHR;
+    } else {
+      token_ = kToken_SAR;
+      stream_->Back();
+    }
+  } else {
+    stream_->Back();
+    token_ = ch;
+  }
+}
+
+bool AsmJsLexer::IsIdentifierStart(uc32 ch) {
+  return (ch >= 'A' && ch <= 'Z') || (ch >= 'a' && ch <= 'z') || ch == '_' ||
+         ch == '$';
+}
+
+bool AsmJsLexer::IsIdentifierPart(uc32 ch) {
+  return IsIdentifierStart(ch) || (ch >= '0' && ch <= '9');
+}
+
+bool AsmJsLexer::IsNumberStart(uc32 ch) {
+  return ch == '.' || (ch >= '0' && ch <= '9');
+}
+
+}  // namespace internal
+}  // namespace v8