Node* Parser::parse(UINT nHandle, INT iStartNt, UINT* pnErrorToken,
UINT nTokens, const UINT pnToklist[])
{
// If pnToklist is NULL, a sequence of integers 0..nTokens-1 will be used
// Sanity checks
if (pnErrorToken)
*pnErrorToken = 0;
if (!nTokens)
return NULL;
if (!this->m_pGrammar)
return NULL;
if (iStartNt >= 0)
// Root must be nonterminal (index < 0)
return NULL;
Nonterminal* pRootNt = (*this->m_pGrammar)[iStartNt];
if (!pRootNt)
// No or invalid root nonterminal
return NULL;
// Initialize the Earley columns
UINT i;
Column** pCol = new Column* [nTokens + 1];
for (i = 0; i < nTokens; i++)
pCol[i] = new Column(this, pnToklist ? pnToklist[i] : i);
pCol[i] = new Column(this, (UINT)-1); // Sentinel column
// Initialize parser state
State* pQ0 = NULL;
StateChunk* pChunkHead = NULL;
// Prepare the initial state
Production* p = pRootNt->getHead();
while (p) {
State* ps = new (pChunkHead) State(iStartNt, 0, p, 0, NULL);
if (!this->push(nHandle, ps, pCol[0], pQ0))
discardState(pChunkHead, ps);
p = p->getNext();
}
// Main parse loop
State* pQ = NULL;
NodeDict ndV; // Node dictionary
/*
clock_t clockStart = clock();
clock_t clockLast = clockStart;
*/
for (i = 0; i < nTokens + 1; i++) {
Column* pEi = pCol[i];
State* pState = pEi->nextState();
// printf("Column %u, token %u\n", i, pEi->getToken());
if (!pState && !pQ0) {
// No parse available at token i-1
if (pnErrorToken)
*pnErrorToken = i;
break;
}
pQ = pQ0;
pQ0 = NULL;
HNode* pH = NULL;
while (pState) {
INT iItem = pState->prodDot();
INT iNtB = pState->getNt();
UINT nStart = pState->getStart();
Node* pW = pState->getNode();
if (iItem < 0) {
// Don't push the same nonterminal more than once to the same column
if (pEi->markSeen(iItem)) {
// Earley predictor
// Push all right hand sides of this nonterminal
p = (*this->m_pGrammar)[iItem]->getHead();
while (p) {
State* psNew = new (pChunkHead) State(iItem, 0, p, i, NULL);
if (!this->push(nHandle, psNew, pEi, pQ))
discardState(pChunkHead, psNew);
p = p->getNext();
}
}
// Add elements from the H set that refer to the
// nonterminal iItem (nt_C)
HNode* ph = pH;
while (ph) {
if (ph->getNt() == iItem) {
Node* pY = this->makeNode(pState, i, ph->getV(), ndV);
State* psNew = new (pChunkHead) State(pState, pY);
if (!this->push(nHandle, psNew, pEi, pQ))
discardState(pChunkHead, psNew);
}
ph = ph->getNext();
}
}
else
if (iItem == 0) {
// Earley completer
if (!pW) {
Label label(iNtB, 0, NULL, i, i);
pW = ndV.lookupOrAdd(label);
pW->addFamily(pState->getProd(), NULL, NULL); // Epsilon production
}
if (nStart == i) {
HNode* ph = new HNode(iNtB, pW);
ph->setNext(pH);
pH = ph;
}
State* psNt = pCol[nStart]->getNtHead(iNtB);
while (psNt) {
Node* pY = this->makeNode(psNt, i, pW, ndV);
State* psNew = new (pChunkHead) State(psNt, pY);
if (!this->push(nHandle, psNew, pEi, pQ))
discardState(pChunkHead, psNew);
psNt = psNt->getNtNext();
}
}
// Move to the next item on the agenda
// (which may have been enlarged by the previous code)
pState = pEi->nextState();
}
// Clean up the H set
while (pH) {
HNode* ph = pH->getNext();
delete pH;
pH = ph;
}
// Reset the node dictionary
ndV.reset();
Node* pV = NULL;
if (pQ) {
Label label(pEi->getToken(), 0, NULL, i, i + 1);
pV = new Node(label); // Reference is deleted below
}
while (pQ) {
// Earley scanner
State* psNext = pQ->getNext();
Node* pY = this->makeNode(pQ, i + 1, pV, ndV);
// Instead of throwing away the old state and creating
// a new almost identical one, re-use the old after
// 'incrementing' it by moving the dot one step to the right
pQ->increment(pY);
ASSERT(i + 1 <= nTokens);
if (!this->push(nHandle, pQ, pCol[i + 1], pQ0))
pQ->~State();
pQ = psNext;
}
// Clean up reference to pV created above
if (pV)
pV->delRef();
/*
clock_t clockNow = clock();
clock_t clockElapsed = clockNow - clockStart;
clock_t clockThis = clockNow - clockLast;
clockLast = clockNow;
printf ("Column %u finished in %.3f sec, elapsed %.3f sec\n", i,
((float)clockThis) / CLOCKS_PER_SEC, ((float)clockElapsed) / CLOCKS_PER_SEC);
fflush(stdout);
*/
}
/*
clock_t clockNow = clock() - clockStart;
printf("Parse loop finished, elapsed %.3f sec\n",
((float)clockNow) / CLOCKS_PER_SEC);
*/
ASSERT(pQ == NULL);
ASSERT(pQ0 == NULL);
Node* pResult = NULL;
if (i > nTokens) {
// Completed the token loop
pCol[nTokens]->resetEnum();
State* ps = pCol[nTokens]->nextState();
while (ps && !pResult) {
// Look through the end states until we find one that spans the
// entire parse tree and derives the starting nonterminal
pResult = ps->getResult(iStartNt);
if (pResult)
// Save the result node from being deleted when the
// column states are deleted
pResult->addRef();
ps = pCol[nTokens]->nextState();
}
if (!pResult && pnErrorToken)
// No parse available at the last column
*pnErrorToken = nTokens;
}
/*
clockNow = clock() - clockStart;
printf("Result found, elapsed %.3f sec\n",
((float)clockNow) / CLOCKS_PER_SEC);
*/
// Cleanup
for (i = 0; i < nTokens + 1; i++)
delete pCol[i];
delete [] pCol;
freeStates(pChunkHead);
/*
clockNow = clock() - clockStart;
printf("Cleanup finished, elapsed %.3f sec\n",
((float)clockNow) / CLOCKS_PER_SEC);
*/
return pResult; // The caller should call delRef() on this after using it
}
