Allow the preprocessor to cache the lexed tokens, so that we can do efficient lookahead and backtracking.

1) New public methods added:
  -EnableBacktrackAtThisPos
  -DisableBacktrack
  -Backtrack
  -isBacktrackEnabled

2) LookAhead() implementation is replaced with a more efficient one.
3) LookNext() is removed.

llvm-svn: 54611
This commit is contained in:
Argyrios Kyrtzidis 2008-08-10 13:15:22 +00:00
parent 07bb087ac1
commit b3dd1e0889
5 changed files with 146 additions and 129 deletions

View File

@ -71,6 +71,9 @@ class Preprocessor {
bool DisableMacroExpansion : 1; // True if macro expansion is disabled.
bool InMacroArgs : 1; // True if parsing fn macro invocation args.
/// CacheTokens - True when the lexed tokens are cached for backtracking.
bool CacheTokens : 1;
/// Identifiers - This is mapping/lookup information for all identifiers in
/// the program, including program keywords.
IdentifierTable Identifiers;
@ -139,10 +142,24 @@ class Preprocessor {
unsigned NumCachedTokenLexers;
TokenLexer *TokenLexerCache[TokenLexerCacheSize];
/// PeekedToken - Cache the token that was retrieved through LookNext().
/// This is a valid token (its Location is valid) when LookNext() is
/// called and gets invalid again when it is "consumed" by Lex().
Token PeekedToken;
// Cached tokens state.
typedef std::vector<Token> CachedTokensTy;
/// CachedTokens - Cached tokens are stored here when we do backtracking or
/// lookahead. They are "lexed" by the CachingLex() method.
CachedTokensTy CachedTokens;
/// CachedLexPos - The position of the cached token that CachingLex() should
/// "lex" next. If it points beyond the CachedTokens vector, it means that
/// a normal Lex() should be invoked.
CachedTokensTy::size_type CachedLexPos;
/// CachedBacktrackPos - Gets set by the EnableBacktrackAtThisPos() method,
/// to indicate the position where CachedLexPos should be set when the
/// BackTrack() method is invoked.
CachedTokensTy::size_type CachedBacktrackPos;
public:
Preprocessor(Diagnostic &diags, const LangOptions &opts, TargetInfo &target,
SourceManager &SM, HeaderSearch &Headers);
@ -258,7 +275,45 @@ public:
/// lexer stack. This should only be used in situations where the current
/// state of the top-of-stack lexer is known.
void RemoveTopOfLexerStack();
/// EnableBacktrackAtThisPos - From the point that this method is called, and
/// until DisableBacktrack() or Backtrack() is called, the Preprocessor keeps
/// track of the lexed tokens so that a subsequent Backtrack() call will make
/// the Preprocessor re-lex the same tokens.
///
/// EnableBacktrackAtThisPos should not be called again until DisableBacktrack
/// or Backtrack is called.
///
/// NOTE: *DO NOT* forget to call either DisableBacktrack() or Backtrack() at
/// some point after EnableBacktrackAtThisPos. If you don't, caching of tokens
/// will continue indefinitely.
///
void EnableBacktrackAtThisPos() {
assert(!CacheTokens && "Backtrack is already enabled!");
CacheTokens = true;
CachedBacktrackPos = CachedLexPos;
EnterCachingLexMode();
}
/// DisableBacktrack - Stop the caching of tokens that was enabled by
/// EnableBacktrackAtThisPos().
void DisableBacktrack() {
assert(CacheTokens && "Backtrack is not enabled!");
CacheTokens = false;
}
/// Backtrack - Make Preprocessor re-lex the tokens that were lexed since
/// EnableBacktrackAtThisPos() was previously called.
void Backtrack() {
assert(CacheTokens && "Backtrack is not enabled!");
CacheTokens = false;
CachedLexPos = CachedBacktrackPos;
}
/// isBacktrackEnabled - True if EnableBacktrackAtThisPos() was called and
/// caching of tokens is on.
bool isBacktrackEnabled() const { return CacheTokens; }
/// Lex - To lex a token from the preprocessor, just pull a token from the
/// current lexer or macro object.
void Lex(Token &Result) {
@ -266,11 +321,8 @@ public:
CurLexer->Lex(Result);
else if (CurTokenLexer)
CurTokenLexer->Lex(Result);
else {
// We have a peeked token that hasn't been consumed yet.
Result = PeekedToken;
ConsumedPeekedToken();
}
else
CachingLex(Result);
}
/// LexNonComment - Lex a token. If it's a comment, keep lexing until we get
@ -300,32 +352,12 @@ public:
/// returned by Lex(), LookAhead(1) returns the token after it, etc. This
/// returns normal tokens after phase 5. As such, it is equivalent to using
/// 'Lex', not 'LexUnexpandedToken'.
///
/// NOTE: is a relatively expensive method, so it should not be used in common
/// code paths if possible!
///
Token LookAhead(unsigned N);
/// LookNext - Returns the next token that would be returned by Lex() without
/// consuming it.
const Token &LookNext() {
if (PeekedToken.getLocation().isInvalid()) {
// We don't have a peeked token that hasn't been consumed yet.
// Peek it now.
PeekToken();
}
return PeekedToken;
const Token &LookAhead(unsigned N) {
if (CachedLexPos + N < CachedTokens.size())
return CachedTokens[CachedLexPos+N];
else
return PeekAhead(N+1);
}
private:
/// PeekToken - Lexes one token into PeekedToken and pushes CurLexer,
/// CurLexerToken into the IncludeMacroStack before setting them to null.
void PeekToken();
/// ConsumedPeekedToken - Called when Lex() is about to return the PeekedToken
/// and have it "consumed".
void ConsumedPeekedToken();
public:
/// Diag - Forwarding function for diagnostics. This emits a diagnostic at
/// the specified Token's location, translating the token's start
@ -523,6 +555,17 @@ private:
bool isAngled, const DirectoryLookup *FromDir,
const DirectoryLookup *&CurDir);
//===--------------------------------------------------------------------===//
// Caching stuff.
void CachingLex(Token &Result);
bool InCachingLexMode() const { return CurLexer == 0 && CurTokenLexer == 0; }
void EnterCachingLexMode();
void ExitCachingLexMode() {
if (InCachingLexMode())
RemoveTopOfLexerStack();
}
const Token &PeekAhead(unsigned N);
//===--------------------------------------------------------------------===//
/// Handle*Directive - implement the various preprocessor directives. These
/// should side-effect the current preprocessor object so that the next call

View File

@ -198,10 +198,7 @@ private:
/// Note that this differs from the Preprocessor's LookAhead method, because
/// the Parser always has one token lexed that the preprocessor doesn't.
///
/// NOTE: is a relatively expensive method, so it should not be used in common
/// code paths if possible!
///
Token GetLookAheadToken(unsigned N) {
const Token &GetLookAheadToken(unsigned N) {
if (N == 0 || Tok.is(tok::eof)) return Tok;
return PP.LookAhead(N-1);
}
@ -209,7 +206,7 @@ private:
/// NextToken - This peeks ahead one token and returns it without
/// consuming it.
const Token &NextToken() {
return PP.LookNext();
return PP.LookAhead(0);
}

View File

@ -0,0 +1,63 @@
//===--- PPCaching.cpp - Handle caching lexed tokens ----------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements pieces of the Preprocessor interface that manage the
// caching of lexed tokens.
//
//===----------------------------------------------------------------------===//
#include "clang/Lex/Preprocessor.h"
using namespace clang;
void Preprocessor::CachingLex(Token &Result) {
if (CachedLexPos < CachedTokens.size()) {
Result = CachedTokens[CachedLexPos++];
return;
}
ExitCachingLexMode();
Lex(Result);
if (!CacheTokens) {
// All cached tokens were consumed.
CachedTokens.clear();
CachedLexPos = 0;
return;
}
// We should cache the lexed token.
EnterCachingLexMode();
if (Result.isNot(tok::eof)) {
CachedTokens.push_back(Result);
++CachedLexPos;
}
}
void Preprocessor::EnterCachingLexMode() {
if (InCachingLexMode())
return;
IncludeMacroStack.push_back(IncludeStackInfo(CurLexer, CurDirLookup,
CurTokenLexer));
CurLexer = 0;
CurTokenLexer = 0;
}
const Token &Preprocessor::PeekAhead(unsigned N) {
assert(CachedLexPos + N > CachedTokens.size() && "Confused caching.");
ExitCachingLexMode();
for (unsigned C = CachedLexPos + N - CachedTokens.size(); C > 0; --C) {
CachedTokens.push_back(Token());
Lex(CachedTokens.back());
}
EnterCachingLexMode();
return CachedTokens.back();
}

View File

@ -60,94 +60,6 @@ Lexer *Preprocessor::getCurrentFileLexer() const {
return 0;
}
/// LookAhead - This peeks ahead N tokens and returns that token without
/// consuming any tokens. LookAhead(0) returns 'Tok', LookAhead(1) returns
/// the token after Tok, etc.
///
/// NOTE: is a relatively expensive method, so it should not be used in common
/// code paths if possible!
///
Token Preprocessor::LookAhead(unsigned N) {
// FIXME: Optimize the case where multiple lookahead calls are used back to
// back. Consider if the the parser contained (dynamically):
// Lookahead(1); Lookahead(1); Lookahead(1)
// This would return the same token 3 times, but would end up making lots of
// token stream lexers to do it. To handle this common case, see if the top
// of the lexer stack is a TokenStreamLexer with macro expansion disabled. If
// so, see if it has 'N' tokens available in it. If so, just return the
// token.
// FIXME: Optimize the case when the parser does multiple nearby lookahead
// calls. For example, consider:
// Lookahead(0); Lookahead(1); Lookahead(2);
// The previous optimization won't apply, and there won't be any space left in
// the array that was previously new'd. To handle this, always round up the
// size we new to a multiple of 16 tokens. If the previous buffer has space
// left, we can just grow it. This means we only have to do the new 1/16th as
// often.
// Optimized LookAhead(0) case.
if (N == 0)
return LookNext();
Token *LookaheadTokens = new Token[N+1];
// Read N+1 tokens into LookaheadTokens. After this loop, Tok is the token
// to return.
Token Tok;
unsigned NumTokens = 0;
for (; N != ~0U; --N, ++NumTokens) {
Lex(Tok);
LookaheadTokens[NumTokens] = Tok;
// If we got to EOF, don't lex past it. This will cause LookAhead to return
// the EOF token.
if (Tok.is(tok::eof))
break;
}
// Okay, at this point, we have the token we want to return in Tok. However,
// we read it and a bunch of other stuff (in LookaheadTokens) that we must
// allow subsequent calls to 'Lex' to return. To do this, we push a new token
// lexer onto the lexer stack with the tokens we read here. This passes
// ownership of LookaheadTokens to EnterTokenStream.
//
// Note that we disable macro expansion of the tokens from this buffer, since
// any macros have already been expanded, and the internal preprocessor state
// may already read past new macros. Consider something like LookAhead(1) on
// X
// #define X 14
// Y
// The lookahead call should return 'Y', and the next Lex call should return
// 'X' even though X -> 14 has already been entered as a macro.
//
EnterTokenStream(LookaheadTokens, NumTokens, true /*DisableExpansion*/,
true /*OwnsTokens*/);
return Tok;
}
/// PeekToken - Lexes one token into PeekedToken and pushes CurLexer,
/// CurLexerToken into the IncludeMacroStack before setting them to null.
void Preprocessor::PeekToken() {
Lex(PeekedToken);
// Cache the current Lexer, TokenLexer and set them both to null.
// When Lex() is called, PeekedToken will be "consumed".
IncludeMacroStack.push_back(IncludeStackInfo(CurLexer, CurDirLookup,
CurTokenLexer));
CurLexer = 0;
CurTokenLexer = 0;
}
/// ConsumedPeekedToken - Called when Lex() is about to return the PeekedToken
/// and have it "consumed".
void Preprocessor::ConsumedPeekedToken() {
assert(PeekedToken.getLocation().isValid() && "Confused Peeking?");
// Restore CurLexer, TokenLexer.
RemoveTopOfLexerStack();
// Make PeekedToken invalid.
PeekedToken.startToken();
}
//===----------------------------------------------------------------------===//
// Methods for Entering and Callbacks for leaving various contexts

View File

@ -68,6 +68,9 @@ Preprocessor::Preprocessor(Diagnostic &diags, const LangOptions &opts,
InMacroArgs = false;
NumCachedTokenLexers = 0;
CacheTokens = false;
CachedLexPos = 0;
// "Poison" __VA_ARGS__, which can only appear in the expansion of a macro.
// This gets unpoisoned where it is allowed.
(Ident__VA_ARGS__ = getIdentifierInfo("__VA_ARGS__"))->setIsPoisoned();
@ -579,4 +582,3 @@ void Preprocessor::HandleIdentifier(Token &Identifier) {
if (II.isExtensionToken() && Features.C99)
Diag(Identifier, diag::ext_token_used);
}