2 * Copyright 2001-2006 Adrian Thurston <thurston@complang.org>
5 /* This file is part of Ragel.
7 * Ragel is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * Ragel is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Ragel; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
30 #include "parsedata.h"
31 #include "parsetree.h"
32 #include "mergesort.h"
33 #include "xmlcodegen.h"
38 char mainMachine[] = "main";
40 void Token::set( const char *str, int len )
43 data = new char[len+1];
44 memcpy( data, str, len );
48 void Token::append( const Token &other )
50 int newLength = length + other.length;
51 char *newString = new char[newLength+1];
52 memcpy( newString, data, length );
53 memcpy( newString + length, other.data, other.length );
54 newString[newLength] = 0;
59 /* Perform minimization after an operation according
60 * to the command line args. */
61 void afterOpMinimize( FsmAp *fsm, bool lastInSeq )
63 /* Switch on the prefered minimization algorithm. */
64 if ( minimizeOpt == MinimizeEveryOp || ( minimizeOpt == MinimizeMostOps && lastInSeq ) ) {
65 /* First clean up the graph. FsmAp operations may leave these
66 * lying around. There should be no dead end states. The subtract
67 * intersection operators are the only places where they may be
68 * created and those operators clean them up. */
69 fsm->removeUnreachableStates();
71 switch ( minimizeLevel ) {
73 fsm->minimizeApproximate();
75 case MinimizePartition1:
76 fsm->minimizePartition1();
78 case MinimizePartition2:
79 fsm->minimizePartition2();
82 fsm->minimizeStable();
88 /* Count the transitions in the fsm by walking the state list. */
89 int countTransitions( FsmAp *fsm )
92 StateAp *state = fsm->stateList.head;
93 while ( state != 0 ) {
94 numTrans += state->outList.length();
100 Key makeFsmKeyHex( char *str, const InputLoc &loc, ParseData *pd )
102 /* Reset errno so we can check for overflow or underflow. In the event of
103 * an error, sets the return val to the upper or lower bound being tested
106 unsigned int size = keyOps->alphType->size;
107 bool unusedBits = size < sizeof(unsigned long);
109 unsigned long ul = strtoul( str, 0, 16 );
111 if ( errno == ERANGE || ( unusedBits && ul >> (size * 8) ) ) {
112 error(loc) << "literal " << str << " overflows the alphabet type" << endl;
113 ul = 1 << (size * 8);
116 if ( unusedBits && keyOps->alphType->isSigned && ul >> (size * 8 - 1) )
117 ul |= (0xffffffff >> (size*8 ) ) << (size*8);
119 return Key( (long)ul );
122 Key makeFsmKeyDec( char *str, const InputLoc &loc, ParseData *pd )
124 /* Convert the number to a decimal. First reset errno so we can check
125 * for overflow or underflow. */
127 long long minVal = keyOps->alphType->minVal;
128 long long maxVal = keyOps->alphType->maxVal;
130 long long ll = strtoll( str, 0, 10 );
132 /* Check for underflow. */
133 if ( ( errno == ERANGE && ll < 0 ) || ll < minVal) {
134 error(loc) << "literal " << str << " underflows the alphabet type" << endl;
137 /* Check for overflow. */
138 else if ( ( errno == ERANGE && ll > 0 ) || ll > maxVal ) {
139 error(loc) << "literal " << str << " overflows the alphabet type" << endl;
143 if ( keyOps->alphType->isSigned )
144 return Key( (long)ll );
146 return Key( (unsigned long)ll );
149 /* Make an fsm key in int format (what the fsm graph uses) from an alphabet
150 * number returned by the parser. Validates that the number doesn't overflow
151 * the alphabet type. */
152 Key makeFsmKeyNum( char *str, const InputLoc &loc, ParseData *pd )
154 /* Switch on hex/decimal format. */
155 if ( str[0] == '0' && str[1] == 'x' )
156 return makeFsmKeyHex( str, loc, pd );
158 return makeFsmKeyDec( str, loc, pd );
161 /* Make an fsm int format (what the fsm graph uses) from a single character.
162 * Performs proper conversion depending on signed/unsigned property of the
164 Key makeFsmKeyChar( char c, ParseData *pd )
166 if ( keyOps->isSigned ) {
167 /* Copy from a char type. */
171 /* Copy from an unsigned byte type. */
172 return Key( (unsigned char)c );
176 /* Make an fsm key array in int format (what the fsm graph uses) from a string
177 * of characters. Performs proper conversion depending on signed/unsigned
178 * property of the alphabet. */
179 void makeFsmKeyArray( Key *result, char *data, int len, ParseData *pd )
181 if ( keyOps->isSigned ) {
182 /* Copy from a char star type. */
184 for ( int i = 0; i < len; i++ )
185 result[i] = Key(src[i]);
188 /* Copy from an unsigned byte ptr type. */
189 unsigned char *src = (unsigned char*) data;
190 for ( int i = 0; i < len; i++ )
191 result[i] = Key(src[i]);
195 /* Like makeFsmKeyArray except the result has only unique keys. They ordering
196 * will be changed. */
197 void makeFsmUniqueKeyArray( KeySet &result, char *data, int len,
198 bool caseInsensitive, ParseData *pd )
200 /* Use a transitions list for getting unique keys. */
201 if ( keyOps->isSigned ) {
202 /* Copy from a char star type. */
204 for ( int si = 0; si < len; si++ ) {
206 result.insert( key );
207 if ( caseInsensitive ) {
209 result.insert( key.toUpper() );
210 else if ( key.isUpper() )
211 result.insert( key.toLower() );
216 /* Copy from an unsigned byte ptr type. */
217 unsigned char *src = (unsigned char*) data;
218 for ( int si = 0; si < len; si++ ) {
220 result.insert( key );
221 if ( caseInsensitive ) {
223 result.insert( key.toUpper() );
224 else if ( key.isUpper() )
225 result.insert( key.toLower() );
231 FsmAp *dotFsm( ParseData *pd )
233 FsmAp *retFsm = new FsmAp();
234 retFsm->rangeFsm( keyOps->minKey, keyOps->maxKey );
238 FsmAp *dotStarFsm( ParseData *pd )
240 FsmAp *retFsm = new FsmAp();
241 retFsm->rangeStarFsm( keyOps->minKey, keyOps->maxKey );
245 /* Make a builtin type. Depends on the signed nature of the alphabet type. */
246 FsmAp *makeBuiltin( BuiltinMachine builtin, ParseData *pd )
248 /* FsmAp created to return. */
250 bool isSigned = keyOps->isSigned;
254 /* All characters. */
255 retFsm = dotFsm( pd );
259 /* Ascii characters 0 to 127. */
260 retFsm = new FsmAp();
261 retFsm->rangeFsm( 0, 127 );
265 /* Ascii extended characters. This is the full byte range. Dependent
266 * on signed, vs no signed. If the alphabet is one byte then just use
269 retFsm = new FsmAp();
270 retFsm->rangeFsm( -128, 127 );
273 retFsm = new FsmAp();
274 retFsm->rangeFsm( 0, 255 );
279 /* Alpha [A-Za-z]. */
280 FsmAp *upper = new FsmAp(), *lower = new FsmAp();
281 upper->rangeFsm( 'A', 'Z' );
282 lower->rangeFsm( 'a', 'z' );
283 upper->unionOp( lower );
284 upper->minimizePartition2();
290 retFsm = new FsmAp();
291 retFsm->rangeFsm( '0', '9' );
295 /* Alpha numerics [0-9A-Za-z]. */
296 FsmAp *digit = new FsmAp(), *lower = new FsmAp();
297 FsmAp *upper = new FsmAp();
298 digit->rangeFsm( '0', '9' );
299 upper->rangeFsm( 'A', 'Z' );
300 lower->rangeFsm( 'a', 'z' );
301 digit->unionOp( upper );
302 digit->unionOp( lower );
303 digit->minimizePartition2();
308 /* Lower case characters. */
309 retFsm = new FsmAp();
310 retFsm->rangeFsm( 'a', 'z' );
314 /* Upper case characters. */
315 retFsm = new FsmAp();
316 retFsm->rangeFsm( 'A', 'Z' );
320 /* Control characters. */
321 FsmAp *cntrl = new FsmAp();
322 FsmAp *highChar = new FsmAp();
323 cntrl->rangeFsm( 0, 31 );
324 highChar->concatFsm( 127 );
325 cntrl->unionOp( highChar );
326 cntrl->minimizePartition2();
331 /* Graphical ascii characters [!-~]. */
332 retFsm = new FsmAp();
333 retFsm->rangeFsm( '!', '~' );
337 /* Printable characters. Same as graph except includes space. */
338 retFsm = new FsmAp();
339 retFsm->rangeFsm( ' ', '~' );
344 FsmAp *range1 = new FsmAp();
345 FsmAp *range2 = new FsmAp();
346 FsmAp *range3 = new FsmAp();
347 FsmAp *range4 = new FsmAp();
348 range1->rangeFsm( '!', '/' );
349 range2->rangeFsm( ':', '@' );
350 range3->rangeFsm( '[', '`' );
351 range4->rangeFsm( '{', '~' );
352 range1->unionOp( range2 );
353 range1->unionOp( range3 );
354 range1->unionOp( range4 );
355 range1->minimizePartition2();
360 /* Whitespace: [\t\v\f\n\r ]. */
361 FsmAp *cntrl = new FsmAp();
362 FsmAp *space = new FsmAp();
363 cntrl->rangeFsm( '\t', '\r' );
364 space->concatFsm( ' ' );
365 cntrl->unionOp( space );
366 cntrl->minimizePartition2();
371 /* Hex digits [0-9A-Fa-f]. */
372 FsmAp *digit = new FsmAp();
373 FsmAp *upper = new FsmAp();
374 FsmAp *lower = new FsmAp();
375 digit->rangeFsm( '0', '9' );
376 upper->rangeFsm( 'A', 'F' );
377 lower->rangeFsm( 'a', 'f' );
378 digit->unionOp( upper );
379 digit->unionOp( lower );
380 digit->minimizePartition2();
385 retFsm = new FsmAp();
390 retFsm = new FsmAp();
398 /* Check if this name inst or any name inst below is referenced. */
399 bool NameInst::anyRefsRec()
404 /* Recurse on children until true. */
405 for ( NameVect::Iter ch = childVect; ch.lte(); ch++ ) {
406 if ( (*ch)->anyRefsRec() )
417 /* Initialize the structure that will collect info during the parse of a
419 ParseData::ParseData( const char *fileName, char *sectionName,
420 const InputLoc §ionLoc )
423 generatingSectionSubset(false),
425 /* 0 is reserved for global error actions. */
447 sectionName(sectionName),
448 sectionLoc(sectionLoc),
453 nextEpsilonResolvedLink(0),
454 nextLongestMatchId(1),
455 lmRequiresErrorState(false)
457 /* Initialize the dictionary of graphs. This is our symbol table. The
458 * initialization needs to be done on construction which happens at the
459 * beginning of a machine spec so any assignment operators can reference
464 /* Clean up the data collected during a parse. */
465 ParseData::~ParseData()
467 /* Delete all the nodes in the action list. Will cause all the
468 * string data that represents the actions to be deallocated. */
472 /* Make a name id in the current name instantiation scope if it is not
474 NameInst *ParseData::addNameInst( const InputLoc &loc, const char *data, bool isLabel )
476 /* Create the name instantitaion object and insert it. */
477 NameInst *newNameInst = new NameInst( loc, curNameInst, data, nextNameId++, isLabel );
478 curNameInst->childVect.append( newNameInst );
480 curNameInst->children.insertMulti( data, newNameInst );
484 void ParseData::initNameWalk()
486 curNameInst = rootName;
490 void ParseData::initExportsNameWalk()
492 curNameInst = exportsRootName;
496 /* Goes into the next child scope. The number of the child is already set up.
497 * We need this for the syncronous name tree and parse tree walk to work
498 * properly. It is reset on entry into a scope and advanced on poping of a
499 * scope. A call to enterNameScope should be accompanied by a corresponding
501 NameFrame ParseData::enterNameScope( bool isLocal, int numScopes )
503 /* Save off the current data. */
505 retFrame.prevNameInst = curNameInst;
506 retFrame.prevNameChild = curNameChild;
507 retFrame.prevLocalScope = localNameScope;
509 /* Enter into the new name scope. */
510 for ( int i = 0; i < numScopes; i++ ) {
511 curNameInst = curNameInst->childVect[curNameChild];
516 localNameScope = curNameInst;
521 /* Return from a child scope to a parent. The parent info must be specified as
522 * an argument and is obtained from the corresponding call to enterNameScope.
524 void ParseData::popNameScope( const NameFrame &frame )
526 /* Pop the name scope. */
527 curNameInst = frame.prevNameInst;
528 curNameChild = frame.prevNameChild+1;
529 localNameScope = frame.prevLocalScope;
532 void ParseData::resetNameScope( const NameFrame &frame )
534 /* Pop the name scope. */
535 curNameInst = frame.prevNameInst;
536 curNameChild = frame.prevNameChild;
537 localNameScope = frame.prevLocalScope;
541 void ParseData::unsetObsoleteEntries( FsmAp *graph )
543 /* Loop the reference names and increment the usage. Names that are no
544 * longer needed will be unset in graph. */
545 for ( NameVect::Iter ref = curNameInst->referencedNames; ref.lte(); ref++ ) {
547 NameInst *name = *ref;
550 /* If the name is no longer needed unset its corresponding entry. */
551 if ( name->numUses == name->numRefs ) {
552 assert( graph->entryPoints.find( name->id ) != 0 );
553 graph->unsetEntry( name->id );
554 assert( graph->entryPoints.find( name->id ) == 0 );
559 NameSet ParseData::resolvePart( NameInst *refFrom, const char *data, bool recLabelsOnly )
561 /* Queue needed for breadth-first search, load it with the start node. */
562 NameInstList nameQueue;
563 nameQueue.append( refFrom );
566 while ( nameQueue.length() > 0 ) {
567 /* Pull the next from location off the queue. */
568 NameInst *from = nameQueue.detachFirst();
570 /* Look for the name. */
571 NameMapEl *low, *high;
572 if ( from->children.findMulti( data, low, high ) ) {
573 /* Record all instances of the name. */
574 for ( ; low <= high; low++ )
575 result.insert( low->value );
578 /* Name not there, do breadth-first operation of appending all
579 * childrent to the processing queue. */
580 for ( NameVect::Iter name = from->childVect; name.lte(); name++ ) {
581 if ( !recLabelsOnly || (*name)->isLabel )
582 nameQueue.append( *name );
586 /* Queue exhausted and name never found. */
590 void ParseData::resolveFrom( NameSet &result, NameInst *refFrom,
591 const NameRef &nameRef, int namePos )
593 /* Look for the name in the owning scope of the factor with aug. */
594 NameSet partResult = resolvePart( refFrom, nameRef[namePos], false );
596 /* If there are more parts to the name then continue on. */
597 if ( ++namePos < nameRef.length() ) {
598 /* There are more components to the name, search using all the part
599 * results as the base. */
600 for ( NameSet::Iter name = partResult; name.lte(); name++ )
601 resolveFrom( result, *name, nameRef, namePos );
604 /* This is the last component, append the part results to the final
606 result.insert( partResult );
610 /* Write out a name reference. */
611 ostream &operator<<( ostream &out, const NameRef &nameRef )
614 if ( nameRef[pos] == 0 ) {
618 out << nameRef[pos++];
619 for ( ; pos < nameRef.length(); pos++ )
620 out << "::" << nameRef[pos];
624 ostream &operator<<( ostream &out, const NameInst &nameInst )
626 /* Count the number fully qualified name parts. */
628 NameInst *curParent = nameInst.parent;
629 while ( curParent != 0 ) {
631 curParent = curParent->parent;
634 /* Make an array and fill it in. */
635 curParent = nameInst.parent;
636 NameInst **parents = new NameInst*[numParents];
637 for ( int p = numParents-1; p >= 0; p-- ) {
638 parents[p] = curParent;
639 curParent = curParent->parent;
642 /* Write the parents out, skip the root. */
643 for ( int p = 1; p < numParents; p++ )
644 out << "::" << ( parents[p]->name != 0 ? parents[p]->name : "<ANON>" );
646 /* Write the name and cleanup. */
647 out << "::" << ( nameInst.name != 0 ? nameInst.name : "<ANON>" );
652 struct CmpNameInstLoc
654 static int compare( const NameInst *ni1, const NameInst *ni2 )
656 if ( ni1->loc.line < ni2->loc.line )
658 else if ( ni1->loc.line > ni2->loc.line )
660 else if ( ni1->loc.col < ni2->loc.col )
662 else if ( ni1->loc.col > ni2->loc.col )
668 void errorStateLabels( const NameSet &resolved )
670 MergeSort<NameInst*, CmpNameInstLoc> mergeSort;
671 mergeSort.sort( resolved.data, resolved.length() );
672 for ( NameSet::Iter res = resolved; res.lte(); res++ )
673 error((*res)->loc) << " -> " << **res << endl;
677 NameInst *ParseData::resolveStateRef( const NameRef &nameRef, InputLoc &loc, Action *action )
679 NameInst *nameInst = 0;
681 /* Do the local search if the name is not strictly a root level name
683 if ( nameRef[0] != 0 ) {
684 /* If the action is referenced, resolve all of them. */
685 if ( action != 0 && action->actionRefs.length() > 0 ) {
686 /* Look for the name in all referencing scopes. */
688 for ( ActionRefs::Iter actRef = action->actionRefs; actRef.lte(); actRef++ )
689 resolveFrom( resolved, *actRef, nameRef, 0 );
691 if ( resolved.length() > 0 ) {
692 /* Take the first one. */
693 nameInst = resolved[0];
694 if ( resolved.length() > 1 ) {
695 /* Complain about the multiple references. */
696 error(loc) << "state reference " << nameRef <<
697 " resolves to multiple entry points" << endl;
698 errorStateLabels( resolved );
704 /* If not found in the local scope, look in global. */
705 if ( nameInst == 0 ) {
707 int fromPos = nameRef[0] != 0 ? 0 : 1;
708 resolveFrom( resolved, rootName, nameRef, fromPos );
710 if ( resolved.length() > 0 ) {
711 /* Take the first. */
712 nameInst = resolved[0];
713 if ( resolved.length() > 1 ) {
714 /* Complain about the multiple references. */
715 error(loc) << "state reference " << nameRef <<
716 " resolves to multiple entry points" << endl;
717 errorStateLabels( resolved );
722 if ( nameInst == 0 ) {
723 /* If not found then complain. */
724 error(loc) << "could not resolve state reference " << nameRef << endl;
729 void ParseData::resolveNameRefs( InlineList *inlineList, Action *action )
731 for ( InlineList::Iter item = *inlineList; item.lte(); item++ ) {
732 switch ( item->type ) {
733 case InlineItem::Entry: case InlineItem::Goto:
734 case InlineItem::Call: case InlineItem::Next: {
735 /* Resolve, pass action for local search. */
736 NameInst *target = resolveStateRef( *item->nameRef, item->loc, action );
738 /* Name lookup error reporting is handled by resolveStateRef. */
740 /* Check if the target goes into a longest match. */
741 NameInst *search = target->parent;
742 while ( search != 0 ) {
743 if ( search->isLongestMatch ) {
744 error(item->loc) << "cannot enter inside a longest "
745 "match construction as an entry point" << endl;
748 search = search->parent;
751 /* Record the reference in the name. This will cause the
752 * entry point to survive to the end of the graph
753 * generating walk. */
754 target->numRefs += 1;
757 item->nameTarg = target;
764 /* Some of the item types may have children. */
765 if ( item->children != 0 )
766 resolveNameRefs( item->children, action );
770 /* Resolve references to labels in actions. */
771 void ParseData::resolveActionNameRefs()
773 for ( ActionList::Iter act = actionList; act.lte(); act++ ) {
774 /* Only care about the actions that are referenced. */
775 if ( act->actionRefs.length() > 0 )
776 resolveNameRefs( act->inlineList, act );
780 /* Walk a name tree starting at from and fill the name index. */
781 void ParseData::fillNameIndex( NameInst *from )
783 /* Fill the value for from in the name index. */
784 nameIndex[from->id] = from;
786 /* Recurse on the implicit final state and then all children. */
787 if ( from->final != 0 )
788 fillNameIndex( from->final );
789 for ( NameVect::Iter name = from->childVect; name.lte(); name++ )
790 fillNameIndex( *name );
793 void ParseData::makeRootNames()
795 /* Create the root name. */
796 rootName = new NameInst( InputLoc(), 0, 0, nextNameId++, false );
797 exportsRootName = new NameInst( InputLoc(), 0, 0, nextNameId++, false );
800 /* Build the name tree and supporting data structures. */
801 void ParseData::makeNameTree( GraphDictEl *dictEl )
803 /* Set up curNameInst for the walk. */
807 /* A start location has been specified. */
808 dictEl->value->makeNameTree( dictEl->loc, this );
811 /* First make the name tree. */
812 for ( GraphList::Iter glel = instanceList; glel.lte(); glel++ ) {
813 /* Recurse on the instance. */
814 glel->value->makeNameTree( glel->loc, this );
818 /* The number of nodes in the tree can now be given by nextNameId */
819 nameIndex = new NameInst*[nextNameId];
820 memset( nameIndex, 0, sizeof(NameInst*)*nextNameId );
821 fillNameIndex( rootName );
822 fillNameIndex( exportsRootName );
826 void ParseData::createBuiltin( const char *name, BuiltinMachine builtin )
828 Expression *expression = new Expression( builtin );
829 Join *join = new Join( expression );
830 JoinOrLm *joinOrLm = new JoinOrLm( join );
831 VarDef *varDef = new VarDef( name, joinOrLm );
832 GraphDictEl *graphDictEl = new GraphDictEl( name, varDef );
833 graphDict.insert( graphDictEl );
836 /* Initialize the graph dict with builtin types. */
837 void ParseData::initGraphDict( )
839 createBuiltin( "any", BT_Any );
840 createBuiltin( "ascii", BT_Ascii );
841 createBuiltin( "extend", BT_Extend );
842 createBuiltin( "alpha", BT_Alpha );
843 createBuiltin( "digit", BT_Digit );
844 createBuiltin( "alnum", BT_Alnum );
845 createBuiltin( "lower", BT_Lower );
846 createBuiltin( "upper", BT_Upper );
847 createBuiltin( "cntrl", BT_Cntrl );
848 createBuiltin( "graph", BT_Graph );
849 createBuiltin( "print", BT_Print );
850 createBuiltin( "punct", BT_Punct );
851 createBuiltin( "space", BT_Space );
852 createBuiltin( "xdigit", BT_Xdigit );
853 createBuiltin( "null", BT_Lambda );
854 createBuiltin( "zlen", BT_Lambda );
855 createBuiltin( "empty", BT_Empty );
858 /* Set the alphabet type. If the types are not valid returns false. */
859 bool ParseData::setAlphType( const InputLoc &loc, char *s1, char *s2 )
862 userAlphType = findAlphType( s1, s2 );
864 return userAlphType != 0;
867 /* Set the alphabet type. If the types are not valid returns false. */
868 bool ParseData::setAlphType( const InputLoc &loc, char *s1 )
871 userAlphType = findAlphType( s1 );
873 return userAlphType != 0;
876 bool ParseData::setVariable( char *var, InlineList *inlineList )
880 if ( strcmp( var, "p" ) == 0 )
882 else if ( strcmp( var, "pe" ) == 0 )
884 else if ( strcmp( var, "eof" ) == 0 )
885 eofExpr = inlineList;
886 else if ( strcmp( var, "cs" ) == 0 )
888 else if ( strcmp( var, "data" ) == 0 )
889 dataExpr = inlineList;
890 else if ( strcmp( var, "top" ) == 0 )
891 topExpr = inlineList;
892 else if ( strcmp( var, "stack" ) == 0 )
893 stackExpr = inlineList;
894 else if ( strcmp( var, "act" ) == 0 )
895 actExpr = inlineList;
896 else if ( strcmp( var, "ts" ) == 0 )
897 tokstartExpr = inlineList;
898 else if ( strcmp( var, "te" ) == 0 )
899 tokendExpr = inlineList;
906 /* Initialize the key operators object that will be referenced by all fsms
908 void ParseData::initKeyOps( )
910 /* Signedness and bounds. */
911 HostType *alphType = alphTypeSet ? userAlphType : hostLang->defaultAlphType;
912 thisKeyOps.setAlphType( alphType );
914 if ( lowerNum != 0 ) {
915 /* If ranges are given then interpret the alphabet type. */
916 thisKeyOps.minKey = makeFsmKeyNum( lowerNum, rangeLowLoc, this );
917 thisKeyOps.maxKey = makeFsmKeyNum( upperNum, rangeHighLoc, this );
920 thisCondData.lastCondKey = thisKeyOps.maxKey;
923 void ParseData::printNameInst( NameInst *nameInst, int level )
925 for ( int i = 0; i < level; i++ )
927 cerr << (nameInst->name != 0 ? nameInst->name : "<ANON>") <<
928 " id: " << nameInst->id <<
929 " refs: " << nameInst->numRefs <<
930 " uses: " << nameInst->numUses << endl;
931 for ( NameVect::Iter name = nameInst->childVect; name.lte(); name++ )
932 printNameInst( *name, level+1 );
935 /* Remove duplicates of unique actions from an action table. */
936 void ParseData::removeDups( ActionTable &table )
938 /* Scan through the table looking for unique actions to
939 * remove duplicates of. */
940 for ( int i = 0; i < table.length(); i++ ) {
941 /* Remove any duplicates ahead of i. */
942 for ( int r = i+1; r < table.length(); ) {
943 if ( table[r].value == table[i].value )
951 /* Remove duplicates from action lists. This operates only on transition and
952 * eof action lists and so should be called once all actions have been
953 * transfered to their final resting place. */
954 void ParseData::removeActionDups( FsmAp *graph )
956 /* Loop all states. */
957 for ( StateList::Iter state = graph->stateList; state.lte(); state++ ) {
958 /* Loop all transitions. */
959 for ( TransList::Iter trans = state->outList; trans.lte(); trans++ )
960 removeDups( trans->actionTable );
961 removeDups( state->toStateActionTable );
962 removeDups( state->fromStateActionTable );
963 removeDups( state->eofActionTable );
967 Action *ParseData::newAction( const char *name, InlineList *inlineList )
972 loc.fileName = "<NONE>";
974 Action *action = new Action( loc, name, inlineList, nextCondId++ );
975 action->actionRefs.append( rootName );
976 actionList.append( action );
980 void ParseData::initLongestMatchData()
982 if ( lmList.length() > 0 ) {
983 /* The initTokStart action resets the token start. */
984 InlineList *il1 = new InlineList;
985 il1->append( new InlineItem( InputLoc(), InlineItem::LmInitTokStart ) );
986 initTokStart = newAction( "initts", il1 );
987 initTokStart->isLmAction = true;
989 /* The initActId action gives act a default value. */
990 InlineList *il4 = new InlineList;
991 il4->append( new InlineItem( InputLoc(), InlineItem::LmInitAct ) );
992 initActId = newAction( "initact", il4 );
993 initActId->isLmAction = true;
995 /* The setTokStart action sets tokstart. */
996 InlineList *il5 = new InlineList;
997 il5->append( new InlineItem( InputLoc(), InlineItem::LmSetTokStart ) );
998 setTokStart = newAction( "ts", il5 );
999 setTokStart->isLmAction = true;
1001 /* The setTokEnd action sets tokend. */
1002 InlineList *il3 = new InlineList;
1003 il3->append( new InlineItem( InputLoc(), InlineItem::LmSetTokEnd ) );
1004 setTokEnd = newAction( "te", il3 );
1005 setTokEnd->isLmAction = true;
1007 /* The action will also need an ordering: ahead of all user action
1009 initTokStartOrd = curActionOrd++;
1010 initActIdOrd = curActionOrd++;
1011 setTokStartOrd = curActionOrd++;
1012 setTokEndOrd = curActionOrd++;
1016 /* After building the graph, do some extra processing to ensure the runtime
1017 * data of the longest mactch operators is consistent. */
1018 void ParseData::setLongestMatchData( FsmAp *graph )
1020 if ( lmList.length() > 0 ) {
1021 /* Make sure all entry points (targets of fgoto, fcall, fnext, fentry)
1022 * init the tokstart. */
1023 for ( EntryMap::Iter en = graph->entryPoints; en.lte(); en++ ) {
1024 /* This is run after duplicates are removed, we must guard against
1025 * inserting a duplicate. */
1026 ActionTable &actionTable = en->value->toStateActionTable;
1027 if ( ! actionTable.hasAction( initTokStart ) )
1028 actionTable.setAction( initTokStartOrd, initTokStart );
1031 /* Find the set of states that are the target of transitions with
1032 * actions that have calls. These states will be targeted by fret
1035 for ( StateList::Iter state = graph->stateList; state.lte(); state++ ) {
1036 for ( TransList::Iter trans = state->outList; trans.lte(); trans++ ) {
1037 for ( ActionTable::Iter ati = trans->actionTable; ati.lte(); ati++ ) {
1038 if ( ati->value->anyCall && trans->toState != 0 )
1039 states.insert( trans->toState );
1045 /* Init tokstart upon entering the above collected states. */
1046 for ( StateSet::Iter ps = states; ps.lte(); ps++ ) {
1047 /* This is run after duplicates are removed, we must guard against
1048 * inserting a duplicate. */
1049 ActionTable &actionTable = (*ps)->toStateActionTable;
1050 if ( ! actionTable.hasAction( initTokStart ) )
1051 actionTable.setAction( initTokStartOrd, initTokStart );
1056 /* Make the graph from a graph dict node. Does minimization and state sorting. */
1057 FsmAp *ParseData::makeInstance( GraphDictEl *gdNode )
1059 /* Build the graph from a walk of the parse tree. */
1060 FsmAp *graph = gdNode->value->walk( this );
1062 /* Resolve any labels that point to multiple states. Any labels that are
1063 * still around are referenced only by gotos and calls and they need to be
1064 * made into deterministic entry points. */
1065 graph->deterministicEntry();
1068 * All state construction is now complete.
1071 /* Transfer actions from the out action tables to eof action tables. */
1072 for ( StateSet::Iter state = graph->finStateSet; state.lte(); state++ )
1073 graph->transferOutActions( *state );
1075 /* Transfer global error actions. */
1076 for ( StateList::Iter state = graph->stateList; state.lte(); state++ )
1077 graph->transferErrorActions( state, 0 );
1079 if ( ::wantDupsRemoved )
1080 removeActionDups( graph );
1082 /* Remove unreachable states. There should be no dead end states. The
1083 * subtract and intersection operators are the only places where they may
1084 * be created and those operators clean them up. */
1085 graph->removeUnreachableStates();
1087 /* No more fsm operations are to be done. Action ordering numbers are
1088 * no longer of use and will just hinder minimization. Clear them. */
1089 graph->nullActionKeys();
1091 /* Transition priorities are no longer of use. We can clear them
1092 * because they will just hinder minimization as well. Clear them. */
1093 graph->clearAllPriorities();
1095 if ( minimizeOpt != MinimizeNone ) {
1096 /* Minimize here even if we minimized at every op. Now that function
1097 * keys have been cleared we may get a more minimal fsm. */
1098 switch ( minimizeLevel ) {
1099 case MinimizeApprox:
1100 graph->minimizeApproximate();
1102 case MinimizeStable:
1103 graph->minimizeStable();
1105 case MinimizePartition1:
1106 graph->minimizePartition1();
1108 case MinimizePartition2:
1109 graph->minimizePartition2();
1114 graph->compressTransitions();
1119 void ParseData::printNameTree()
1121 /* Print the name instance map. */
1122 for ( NameVect::Iter name = rootName->childVect; name.lte(); name++ )
1123 printNameInst( *name, 0 );
1125 cerr << "name index:" << endl;
1126 /* Show that the name index is correct. */
1127 for ( int ni = 0; ni < nextNameId; ni++ ) {
1129 const char *name = nameIndex[ni]->name;
1130 cerr << ( name != 0 ? name : "<ANON>" ) << endl;
1134 FsmAp *ParseData::makeSpecific( GraphDictEl *gdNode )
1136 /* Build the name tree and supporting data structures. */
1137 makeNameTree( gdNode );
1139 /* Resove name references from gdNode. */
1141 gdNode->value->resolveNameRefs( this );
1143 /* Do not resolve action references. Since we are not building the entire
1144 * graph there's a good chance that many name references will fail. This
1145 * is okay since generating part of the graph is usually only done when
1146 * inspecting the compiled machine. */
1148 /* Same story for extern entry point references. */
1150 /* Flag this case so that the XML code generator is aware that we haven't
1151 * looked up name references in actions. It can then avoid segfaulting. */
1152 generatingSectionSubset = true;
1154 /* Just building the specified graph. */
1156 FsmAp *mainGraph = makeInstance( gdNode );
1161 FsmAp *ParseData::makeAll()
1163 /* Build the name tree and supporting data structures. */
1166 /* Resove name references in the tree. */
1168 for ( GraphList::Iter glel = instanceList; glel.lte(); glel++ )
1169 glel->value->resolveNameRefs( this );
1171 /* Resolve action code name references. */
1172 resolveActionNameRefs();
1174 /* Force name references to the top level instantiations. */
1175 for ( NameVect::Iter inst = rootName->childVect; inst.lte(); inst++ )
1176 (*inst)->numRefs += 1;
1178 FsmAp *mainGraph = 0;
1179 FsmAp **graphs = new FsmAp*[instanceList.length()];
1182 /* Make all the instantiations, we know that main exists in this list. */
1184 for ( GraphList::Iter glel = instanceList; glel.lte(); glel++ ) {
1185 if ( strcmp( glel->key, mainMachine ) == 0 ) {
1186 /* Main graph is always instantiated. */
1187 mainGraph = makeInstance( glel );
1190 /* Instantiate and store in others array. */
1191 graphs[numOthers++] = makeInstance( glel );
1195 if ( mainGraph == 0 )
1196 mainGraph = graphs[--numOthers];
1198 if ( numOthers > 0 ) {
1199 /* Add all the other graphs into main. */
1200 mainGraph->globOp( graphs, numOthers );
1207 void ParseData::analyzeAction( Action *action, InlineList *inlineList )
1209 /* FIXME: Actions used as conditions should be very constrained. */
1210 for ( InlineList::Iter item = *inlineList; item.lte(); item++ ) {
1211 if ( item->type == InlineItem::Call || item->type == InlineItem::CallExpr )
1212 action->anyCall = true;
1214 /* Need to recurse into longest match items. */
1215 if ( item->type == InlineItem::LmSwitch ) {
1216 LongestMatch *lm = item->longestMatch;
1217 for ( LmPartList::Iter lmi = *lm->longestMatchList; lmi.lte(); lmi++ ) {
1218 if ( lmi->action != 0 )
1219 analyzeAction( action, lmi->action->inlineList );
1223 if ( item->type == InlineItem::LmOnLast ||
1224 item->type == InlineItem::LmOnNext ||
1225 item->type == InlineItem::LmOnLagBehind )
1227 LongestMatchPart *lmi = item->longestMatchPart;
1228 if ( lmi->action != 0 )
1229 analyzeAction( action, lmi->action->inlineList );
1232 if ( item->children != 0 )
1233 analyzeAction( action, item->children );
1238 /* Check actions for bad uses of fsm directives. We don't go inside longest
1239 * match items in actions created by ragel, since we just want the user
1241 void ParseData::checkInlineList( Action *act, InlineList *inlineList )
1243 for ( InlineList::Iter item = *inlineList; item.lte(); item++ ) {
1245 if ( act->numEofRefs > 0 ) {
1246 switch ( item->type ) {
1247 /* Currently no checks. */
1254 if ( item->children != 0 )
1255 checkInlineList( act, item->children );
1259 void ParseData::checkAction( Action *action )
1261 /* Check for actions with calls that are embedded within a longest match
1263 if ( !action->isLmAction && action->numRefs() > 0 && action->anyCall ) {
1264 for ( ActionRefs::Iter ar = action->actionRefs; ar.lte(); ar++ ) {
1265 NameInst *check = *ar;
1266 while ( check != 0 ) {
1267 if ( check->isLongestMatch ) {
1268 error(action->loc) << "within a scanner, fcall is permitted"
1269 " only in pattern actions" << endl;
1272 check = check->parent;
1277 checkInlineList( action, action->inlineList );
1281 void ParseData::analyzeGraph( FsmAp *graph )
1283 for ( ActionList::Iter act = actionList; act.lte(); act++ )
1284 analyzeAction( act, act->inlineList );
1286 for ( StateList::Iter st = graph->stateList; st.lte(); st++ ) {
1287 /* The transition list. */
1288 for ( TransList::Iter trans = st->outList; trans.lte(); trans++ ) {
1289 for ( ActionTable::Iter at = trans->actionTable; at.lte(); at++ )
1290 at->value->numTransRefs += 1;
1293 for ( ActionTable::Iter at = st->toStateActionTable; at.lte(); at++ )
1294 at->value->numToStateRefs += 1;
1296 for ( ActionTable::Iter at = st->fromStateActionTable; at.lte(); at++ )
1297 at->value->numFromStateRefs += 1;
1299 for ( ActionTable::Iter at = st->eofActionTable; at.lte(); at++ )
1300 at->value->numEofRefs += 1;
1302 for ( StateCondList::Iter sc = st->stateCondList; sc.lte(); sc++ ) {
1303 for ( CondSet::Iter sci = sc->condSpace->condSet; sci.lte(); sci++ )
1304 (*sci)->numCondRefs += 1;
1308 /* Checks for bad usage of directives in action code. */
1309 for ( ActionList::Iter act = actionList; act.lte(); act++ )
1313 void ParseData::makeExportsNameTree()
1315 /* Make a name tree for the exports. */
1316 initExportsNameWalk();
1318 /* First make the name tree. */
1319 for ( GraphDict::Iter gdel = graphDict; gdel.lte(); gdel++ ) {
1320 if ( gdel->value->isExport ) {
1321 /* Recurse on the instance. */
1322 gdel->value->makeNameTree( gdel->loc, this );
1327 void ParseData::makeExports()
1329 makeExportsNameTree();
1331 /* Resove name references in the tree. */
1332 initExportsNameWalk();
1333 for ( GraphDict::Iter gdel = graphDict; gdel.lte(); gdel++ ) {
1334 if ( gdel->value->isExport )
1335 gdel->value->resolveNameRefs( this );
1338 /* Make all the instantiations, we know that main exists in this list. */
1339 initExportsNameWalk();
1340 for ( GraphDict::Iter gdel = graphDict; gdel.lte(); gdel++ ) {
1341 /* Check if this var def is an export. */
1342 if ( gdel->value->isExport ) {
1343 /* Build the graph from a walk of the parse tree. */
1344 FsmAp *graph = gdel->value->walk( this );
1346 /* Build the graph from a walk of the parse tree. */
1347 if ( !graph->checkSingleCharMachine() ) {
1348 error(gdel->loc) << "bad export machine, must define "
1349 "a single character" << endl;
1352 /* Safe to extract the key and declare the export. */
1353 Key exportKey = graph->startState->outList.head->lowKey;
1354 exportList.append( new Export( gdel->value->name, exportKey ) );
1361 /* Construct the machine and catch failures which can occur during
1363 void ParseData::prepareMachineGen( GraphDictEl *graphDictEl )
1366 /* This machine construction can fail. */
1367 prepareMachineGenTBWrapped( graphDictEl );
1369 catch ( FsmConstructFail fail ) {
1370 switch ( fail.reason ) {
1371 case FsmConstructFail::CondNoKeySpace: {
1372 InputLoc &loc = alphTypeSet ? alphTypeLoc : sectionLoc;
1373 error(loc) << "sorry, no more characters are "
1374 "available in the alphabet space" << endl;
1375 error(loc) << " for conditions, please use a "
1376 "smaller alphtype or reduce" << endl;
1377 error(loc) << " the span of characters on which "
1378 "conditions are embedded" << endl;
1385 void ParseData::prepareMachineGenTBWrapped( GraphDictEl *graphDictEl )
1390 initLongestMatchData();
1392 /* Make the graph, do minimization. */
1393 if ( graphDictEl == 0 )
1394 sectionGraph = makeAll();
1396 sectionGraph = makeSpecific( graphDictEl );
1398 /* Compute exports from the export definitions. */
1401 /* If any errors have occured in the input file then don't write anything. */
1402 if ( gblErrorCount > 0 )
1405 analyzeGraph( sectionGraph );
1407 /* Depends on the graph analysis. */
1408 setLongestMatchData( sectionGraph );
1410 /* Decide if an error state is necessary.
1411 * 1. There is an error transition
1412 * 2. There is a gap in the transitions
1413 * 3. The longest match operator requires it. */
1414 if ( lmRequiresErrorState || sectionGraph->hasErrorTrans() )
1415 sectionGraph->errState = sectionGraph->addState();
1417 /* State numbers need to be assigned such that all final states have a
1418 * larger state id number than all non-final states. This enables the
1419 * first_final mechanism to function correctly. We also want states to be
1420 * ordered in a predictable fashion. So we first apply a depth-first
1421 * search, then do a stable sort by final state status, then assign
1424 sectionGraph->depthFirstOrdering();
1425 sectionGraph->sortStatesByFinal();
1426 sectionGraph->setStateNumbers( 0 );
1429 void ParseData::generateXML( ostream &out )
1433 /* Make the generator. */
1434 XMLCodeGen codeGen( sectionName, this, sectionGraph, out );
1436 /* Write out with it. */
1439 if ( printStatistics ) {
1440 cerr << "fsm name : " << sectionName << endl;
1441 cerr << "num states: " << sectionGraph->stateList.length() << endl;
1446 /* Send eof to all parsers. */
1447 void terminateAllParsers( )
1449 /* FIXME: a proper token is needed here. Suppose we should use the
1450 * location of EOF in the last file that the parser was referenced in. */
1452 loc.fileName = "<EOF>";
1455 for ( ParserDict::Iter pdel = parserDict; pdel.lte(); pdel++ )
1456 pdel->value->token( loc, Parser_tk_eof, 0, 0 );
1459 void writeLanguage( std::ostream &out )
1462 switch ( hostLang->lang ) {
1463 case HostLang::C: out << "C"; break;
1464 case HostLang::D: out << "D"; break;
1465 case HostLang::Java: out << "Java"; break;
1466 case HostLang::Ruby: out << "Ruby"; break;
1467 case HostLang::CSharp: out << "C#"; break;
1473 void writeMachines( std::ostream &out, std::string hostData,
1474 const char *inputFileName, XmlParser &xmlParser )
1476 if ( machineSpec == 0 && machineName == 0 ) {
1477 /* No machine spec or machine name given. Generate everything. */
1478 for ( ParserDict::Iter parser = parserDict; parser.lte(); parser++ ) {
1479 ParseData *pd = parser->value->pd;
1480 if ( pd->instanceList.length() > 0 )
1481 pd->prepareMachineGen( 0 );
1484 if ( gblErrorCount == 0 ) {
1485 out << "<ragel version=\"" VERSION "\" filename=\"" << inputFileName << "\"";
1486 writeLanguage( out );
1488 for ( ParserDict::Iter parser = parserDict; parser.lte(); parser++ ) {
1489 ParseData *pd = parser->value->pd;
1490 if ( pd->instanceList.length() > 0 )
1491 pd->generateXML( out );
1494 out << "</ragel>\n";
1497 else if ( parserDict.length() > 0 ) {
1498 /* There is either a machine spec or machine name given. */
1499 ParseData *parseData = 0;
1500 GraphDictEl *graphDictEl = 0;
1502 /* Traverse the sections, break out when we find a section/machine
1503 * that matches the one specified. */
1504 for ( ParserDict::Iter parser = parserDict; parser.lte(); parser++ ) {
1505 ParseData *checkPd = parser->value->pd;
1506 if ( machineSpec == 0 || strcmp( checkPd->sectionName, machineSpec ) == 0 ) {
1507 GraphDictEl *checkGdEl = 0;
1508 if ( machineName == 0 || (checkGdEl =
1509 checkPd->graphDict.find( machineName )) != 0 )
1511 /* Have a machine spec and/or machine name that matches
1512 * the -M/-S options. */
1513 parseData = checkPd;
1514 graphDictEl = checkGdEl;
1520 if ( parseData == 0 )
1521 error() << "could not locate machine specified with -S and/or -M" << endl;
1523 /* Section/Machine to emit was found. Prepare and emit it. */
1524 parseData->prepareMachineGen( graphDictEl );
1525 if ( gblErrorCount == 0 ) {
1526 out << "<ragel version=\"" VERSION "\" filename=\"" << inputFileName << "\"";
1527 writeLanguage( out );
1529 parseData->generateXML( out );
1531 out << "</ragel>\n";