diff options
author | Adrian Thurston <thurston@complang.org> | 2012-08-01 13:18:02 +0000 |
---|---|---|
committer | Adrian Thurston <thurston@complang.org> | 2012-08-01 13:18:02 +0000 |
commit | 6bc9727d3ac615090bf5086cae43d225d3fb552f (patch) | |
tree | 695c7894d54b8b9de40e4cf347063e99238b7b0a /colm/fsmgraph.cc | |
parent | 39c9b4a6f1014cb30ee535d4f534d0dcc4fe5905 (diff) | |
download | colm-6bc9727d3ac615090bf5086cae43d225d3fb552f.tar.gz |
revert "moved 'colm' dir to 'src'"
Colm includes a library component with headers installed to a private dir
inside include: $prefix/include/colm. We need our headers to reference each
other using this colm prefix. This needs to be true for compiling our source
and also for compiling external programs. It is conventient to have all the
source in a directory called colm and then to use -I <source-root> when
building colm. We use $prefix/include when building external programs.
This reverts commit 247904a84430b8c9151fa6afb68f01b60afb92c9.
Diffstat (limited to 'colm/fsmgraph.cc')
-rw-r--r-- | colm/fsmgraph.cc | 1408 |
1 files changed, 1408 insertions, 0 deletions
diff --git a/colm/fsmgraph.cc b/colm/fsmgraph.cc new file mode 100644 index 00000000..590d7902 --- /dev/null +++ b/colm/fsmgraph.cc @@ -0,0 +1,1408 @@ +/* + * Copyright 2006-2012 Adrian Thurston <thurston@complang.org> + */ + +/* This file is part of Colm. + * + * Colm is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * Colm is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with Colm; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + */ + +#include <assert.h> +#include <iostream> + +#include "config.h" +#include "defs.h" +#include "fsmgraph.h" +#include "mergesort.h" + +using std::cerr; +using std::endl; + +/* Make a new state. The new state will be put on the graph's + * list of state. The new state can be created final or non final. */ +FsmState *FsmGraph::addState() +{ + /* Make the new state to return. */ + FsmState *state = new FsmState(); + + if ( misfitAccounting ) { + /* Create the new state on the misfit list. All states are created + * with no foreign in transitions. */ + misfitList.append( state ); + } + else { + /* Create the new state. */ + stateList.append( state ); + } + + return state; +} + +/* Construct an FSM that is the concatenation of an array of characters. A new + * machine will be made that has len+1 states with one transition between each + * state for each integer in str. IsSigned determines if the integers are to + * be considered as signed or unsigned ints. */ +void FsmGraph::concatFsm( Key *str, int len ) +{ + /* Make the first state and set it as the start state. */ + FsmState *last = addState(); + setStartState( last ); + + /* Attach subsequent states. */ + for ( int i = 0; i < len; i++ ) { + FsmState *newState = addState(); + attachNewTrans( last, newState, str[i], str[i] ); + last = newState; + } + + /* Make the last state the final state. */ + setFinState( last ); +} + +/* Case insensitive version of concatFsm. */ +void FsmGraph::concatFsmCI( Key *str, int len ) +{ + /* Make the first state and set it as the start state. */ + FsmState *last = addState(); + setStartState( last ); + + /* Attach subsequent states. */ + for ( int i = 0; i < len; i++ ) { + FsmState *newState = addState(); + + KeySet keySet; + if ( str[i].isLower() ) + keySet.insert( str[i].toUpper() ); + if ( str[i].isUpper() ) + keySet.insert( str[i].toLower() ); + keySet.insert( str[i] ); + + for ( int i = 0; i < keySet.length(); i++ ) + attachNewTrans( last, newState, keySet[i], keySet[i] ); + + last = newState; + } + + /* Make the last state the final state. */ + setFinState( last ); +} + +/* Construct a machine that matches one character. A new machine will be made + * that has two states with a single transition between the states. IsSigned + * determines if the integers are to be considered as signed or unsigned ints. */ +void FsmGraph::concatFsm( Key chr ) +{ + /* Two states first start, second final. */ + setStartState( addState() ); + + FsmState *end = addState(); + setFinState( end ); + + /* Attach on the character. */ + attachNewTrans( startState, end, chr, chr ); +} + +/* Construct a machine that matches any character in set. A new machine will + * be made that has two states and len transitions between the them. The set + * should be ordered correctly accroding to KeyOps and should not contain + * any duplicates. */ +void FsmGraph::orFsm( Key *set, int len ) +{ + /* Two states first start, second final. */ + setStartState( addState() ); + + FsmState *end = addState(); + setFinState( end ); + + for ( int i = 1; i < len; i++ ) + assert( set[i-1] < set[i] ); + + /* Attach on all the integers in the given string of ints. */ + for ( int i = 0; i < len; i++ ) + attachNewTrans( startState, end, set[i], set[i] ); +} + +/* Construct a machine that matches a range of characters. A new machine will + * be made with two states and a range transition between them. The range will + * match any characters from low to high inclusive. Low should be less than or + * equal to high otherwise undefined behaviour results. IsSigned determines + * if the integers are to be considered as signed or unsigned ints. */ +void FsmGraph::rangeFsm( Key low, Key high ) +{ + /* Two states first start, second final. */ + setStartState( addState() ); + + FsmState *end = addState(); + setFinState( end ); + + /* Attach using the range of characters. */ + attachNewTrans( startState, end, low, high ); +} + +/* Construct a machine that a repeated range of characters. */ +void FsmGraph::rangeStarFsm( Key low, Key high) +{ + /* One state which is final and is the start state. */ + setStartState( addState() ); + setFinState( startState ); + + /* Attach start to start using range of characters. */ + attachNewTrans( startState, startState, low, high ); +} + +/* Construct a machine that matches the empty string. A new machine will be + * made with only one state. The new state will be both a start and final + * state. IsSigned determines if the machine has a signed or unsigned + * alphabet. Fsm operations must be done on machines with the same alphabet + * signedness. */ +void FsmGraph::lambdaFsm( ) +{ + /* Give it one state with no transitions making it + * the start state and final state. */ + setStartState( addState() ); + setFinState( startState ); +} + +/* Construct a machine that matches nothing at all. A new machine will be + * made with only one state. It will not be final. */ +void FsmGraph::emptyFsm( ) +{ + /* Give it one state with no transitions making it + * the start state and final state. */ + setStartState( addState() ); +} + +void FsmGraph::transferOutData( FsmState *destState, FsmState *srcState ) +{ + for ( TransList::Iter trans = destState->outList; trans.lte(); trans++ ) { + if ( trans->toState != 0 ) { + /* Get the actions data from the outActionTable. */ + trans->actionTable.setActions( srcState->outActionTable ); + + /* Get the priorities from the outPriorTable. */ + trans->priorTable.setPriors( srcState->outPriorTable ); + } + } +} + +/* Kleene star operator. Makes this machine the kleene star of itself. Any + * transitions made going out of the machine and back into itself will be + * notified that they are leaving transitions by having the leavingFromState + * callback invoked. */ +void FsmGraph::starOp( ) +{ + /* For the merging process. */ + MergeData md; + + /* Turn on misfit accounting to possibly catch the old start state. */ + setMisfitAccounting( true ); + + /* Create the new new start state. It will be set final after the merging + * of the final states with the start state is complete. */ + FsmState *prevStartState = startState; + unsetStartState(); + setStartState( addState() ); + + /* Merge the new start state with the old one to isolate it. */ + mergeStates( md, startState, prevStartState ); + + /* Merge the start state into all final states. Except the start state on + * the first pass. If the start state is set final we will be doubling up + * its transitions, which will get transfered to any final states that + * follow it in the final state set. This will be determined by the order + * of items in the final state set. To prevent this we just merge with the + * start on a second pass. */ + for ( StateSet::Iter st = finStateSet; st.lte(); st++ ) { + if ( *st != startState ) + mergeStatesLeaving( md, *st, startState ); + } + + /* Now it is safe to merge the start state with itself (provided it + * is set final). */ + if ( startState->isFinState() ) + mergeStatesLeaving( md, startState, startState ); + + /* Now ensure the new start state is a final state. */ + setFinState( startState ); + + /* Fill in any states that were newed up as combinations of others. */ + fillInStates( md ); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +void FsmGraph::repeatOp( int times ) +{ + /* Must be 1 and up. 0 produces null machine and requires deleting this. */ + assert( times > 0 ); + + /* A repeat of one does absolutely nothing. */ + if ( times == 1 ) + return; + + /* Make a machine to make copies from. */ + FsmGraph *copyFrom = new FsmGraph( *this ); + + /* Concatentate duplicates onto the end up until before the last. */ + for ( int i = 1; i < times-1; i++ ) { + FsmGraph *dup = new FsmGraph( *copyFrom ); + doConcat( dup, 0, false ); + } + + /* Now use the copyFrom on the end. */ + doConcat( copyFrom, 0, false ); +} + +void FsmGraph::optionalRepeatOp( int times ) +{ + /* Must be 1 and up. 0 produces null machine and requires deleting this. */ + assert( times > 0 ); + + /* A repeat of one optional merely allows zero string. */ + if ( times == 1 ) { + setFinState( startState ); + return; + } + + /* Make a machine to make copies from. */ + FsmGraph *copyFrom = new FsmGraph( *this ); + + /* The state set used in the from end of the concatentation. Starts with + * the initial final state set, then after each concatenation, gets set to + * the the final states that come from the the duplicate. */ + StateSet lastFinSet( finStateSet ); + + /* Set the initial state to zero to allow zero copies. */ + setFinState( startState ); + + /* Concatentate duplicates onto the end up until before the last. */ + for ( int i = 1; i < times-1; i++ ) { + /* Make a duplicate for concating and set the fin bits to graph 2 so we + * can pick out it's final states after the optional style concat. */ + FsmGraph *dup = new FsmGraph( *copyFrom ); + dup->setFinBits( SB_GRAPH2 ); + doConcat( dup, &lastFinSet, true ); + + /* Clear the last final state set and make the new one by taking only + * the final states that come from graph 2.*/ + lastFinSet.empty(); + for ( int i = 0; i < finStateSet.length(); i++ ) { + /* If the state came from graph 2, add it to the last set and clear + * the bits. */ + FsmState *fs = finStateSet[i]; + if ( fs->stateBits & SB_GRAPH2 ) { + lastFinSet.insert( fs ); + fs->stateBits &= ~SB_GRAPH2; + } + } + } + + /* Now use the copyFrom on the end, no bits set, no bits to clear. */ + doConcat( copyFrom, &lastFinSet, true ); +} + + +/* Fsm concatentation worker. Supports treating the concatentation as optional, + * which essentially leaves the final states of machine one as final. */ +void FsmGraph::doConcat( FsmGraph *other, StateSet *fromStates, bool optional ) +{ + /* For the merging process. */ + StateSet finStateSetCopy, startStateSet; + MergeData md; + + /* Turn on misfit accounting for both graphs. */ + setMisfitAccounting( true ); + other->setMisfitAccounting( true ); + + /* Get the other's start state. */ + FsmState *otherStartState = other->startState; + + /* Unset other's start state before bringing in the entry points. */ + other->unsetStartState(); + + /* Bring in the rest of other's entry points. */ + copyInEntryPoints( other ); + other->entryPoints.empty(); + + /* Bring in other's states into our state lists. */ + stateList.append( other->stateList ); + misfitList.append( other->misfitList ); + + /* If from states is not set, then get a copy of our final state set before + * we clobber it and use it instead. */ + if ( fromStates == 0 ) { + finStateSetCopy = finStateSet; + fromStates = &finStateSetCopy; + } + + /* Unset all of our final states and get the final states from other. */ + if ( !optional ) + unsetAllFinStates(); + finStateSet.insert( other->finStateSet ); + + /* Since other's lists are empty, we can delete the fsm without + * affecting any states. */ + delete other; + + /* Merge our former final states with the start state of other. */ + for ( int i = 0; i < fromStates->length(); i++ ) { + FsmState *state = fromStates->data[i]; + + /* Merge the former final state with other's start state. */ + mergeStatesLeaving( md, state, otherStartState ); + + /* If the former final state was not reset final then we must clear + * the state's out trans data. If it got reset final then it gets to + * keep its out trans data. This must be done before fillInStates gets + * called to prevent the data from being sourced. */ + if ( ! state->isFinState() ) + clearOutData( state ); + } + + /* Fill in any new states made from merging. */ + fillInStates( md ); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +/* Concatenates other to the end of this machine. Other is deleted. Any + * transitions made leaving this machine and entering into other are notified + * that they are leaving transitions by having the leavingFromState callback + * invoked. */ +void FsmGraph::concatOp( FsmGraph *other ) +{ + /* Assert same signedness and return graph concatenation op. */ + doConcat( other, 0, false ); +} + + +void FsmGraph::doOr( FsmGraph *other ) +{ + /* For the merging process. */ + MergeData md; + + /* Build a state set consisting of both start states */ + StateSet startStateSet; + startStateSet.insert( startState ); + startStateSet.insert( other->startState ); + + /* Both of the original start states loose their start state status. */ + unsetStartState(); + other->unsetStartState(); + + /* Bring in the rest of other's entry points. */ + copyInEntryPoints( other ); + other->entryPoints.empty(); + + /* Merge the lists. This will move all the states from other + * into this. No states will be deleted. */ + stateList.append( other->stateList ); + misfitList.append( other->misfitList ); + + /* Move the final set data from other into this. */ + finStateSet.insert(other->finStateSet); + other->finStateSet.empty(); + + /* Since other's list is empty, we can delete the fsm without + * affecting any states. */ + delete other; + + /* Create a new start state. */ + setStartState( addState() ); + + /* Merge the start states. */ + mergeStates( md, startState, startStateSet.data, startStateSet.length() ); + + /* Fill in any new states made from merging. */ + fillInStates( md ); +} + +/* Unions other with this machine. Other is deleted. */ +void FsmGraph::unionOp( FsmGraph *other ) +{ + /* Turn on misfit accounting for both graphs. */ + setMisfitAccounting( true ); + other->setMisfitAccounting( true ); + + /* Call Worker routine. */ + doOr( other ); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +/* Intersects other with this machine. Other is deleted. */ +void FsmGraph::intersectOp( FsmGraph *other ) +{ + /* Turn on misfit accounting for both graphs. */ + setMisfitAccounting( true ); + other->setMisfitAccounting( true ); + + /* Set the fin bits on this and other to want each other. */ + setFinBits( SB_GRAPH1 ); + other->setFinBits( SB_GRAPH2 ); + + /* Call worker Or routine. */ + doOr( other ); + + /* Unset any final states that are no longer to + * be final due to final bits. */ + unsetIncompleteFinals(); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); + + /* Remove states that have no path to a final state. */ + removeDeadEndStates(); +} + +/* Set subtracts other machine from this machine. Other is deleted. */ +void FsmGraph::subtractOp( FsmGraph *other ) +{ + /* Turn on misfit accounting for both graphs. */ + setMisfitAccounting( true ); + other->setMisfitAccounting( true ); + + /* Set the fin bits of other to be killers. */ + other->setFinBits( SB_GRAPH1 ); + + /* Call worker Or routine. */ + doOr( other ); + + /* Unset any final states that are no longer to + * be final due to final bits. */ + unsetKilledFinals(); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); + + /* Remove states that have no path to a final state. */ + removeDeadEndStates(); +} + +bool FsmGraph::inEptVect( EptVect *eptVect, FsmState *state ) +{ + if ( eptVect != 0 ) { + /* Vect is there, walk it looking for state. */ + for ( int i = 0; i < eptVect->length(); i++ ) { + if ( eptVect->data[i].targ == state ) + return true; + } + } + return false; +} + +/* Fill epsilon vectors in a root state from a given starting point. Epmploys + * a depth first search through the graph of epsilon transitions. */ +void FsmGraph::epsilonFillEptVectFrom( FsmState *root, FsmState *from, bool parentLeaving ) +{ + /* Walk the epsilon transitions out of the state. */ + for ( EpsilonTrans::Iter ep = from->epsilonTrans; ep.lte(); ep++ ) { + /* Find the entry point, if the it does not resove, ignore it. */ + EntryMapEl *enLow, *enHigh; + if ( entryPoints.findMulti( *ep, enLow, enHigh ) ) { + /* Loop the targets. */ + for ( EntryMapEl *en = enLow; en <= enHigh; en++ ) { + /* Do not add the root or states already in eptVect. */ + FsmState *targ = en->value; + if ( targ != from && !inEptVect(root->eptVect, targ) ) { + /* Maybe need to create the eptVect. */ + if ( root->eptVect == 0 ) + root->eptVect = new EptVect(); + + /* If moving to a different graph or if any parent is + * leaving then we are leaving. */ + bool leaving = parentLeaving || + root->owningGraph != targ->owningGraph; + + /* All ok, add the target epsilon and recurse. */ + root->eptVect->append( EptVectEl(targ, leaving) ); + epsilonFillEptVectFrom( root, targ, leaving ); + } + } + } + } +} + +void FsmGraph::shadowReadWriteStates( MergeData &md ) +{ + /* Init isolatedShadow algorithm data. */ + for ( StateList::Iter st = stateList; st.lte(); st++ ) + st->isolatedShadow = 0; + + /* Any states that may be both read from and written to must + * be shadowed. */ + for ( StateList::Iter st = stateList; st.lte(); st++ ) { + /* Find such states by looping through stateVect lists, which give us + * the states that will be read from. May cause us to visit the states + * that we are interested in more than once. */ + if ( st->eptVect != 0 ) { + /* For all states that will be read from. */ + for ( EptVect::Iter ept = *st->eptVect; ept.lte(); ept++ ) { + /* Check for read and write to the same state. */ + FsmState *targ = ept->targ; + if ( targ->eptVect != 0 ) { + /* State is to be written to, if the shadow is not already + * there, create it. */ + if ( targ->isolatedShadow == 0 ) { + FsmState *shadow = addState(); + mergeStates( md, shadow, targ ); + targ->isolatedShadow = shadow; + } + + /* Write shadow into the state vector so that it is the + * state that the epsilon transition will read from. */ + ept->targ = targ->isolatedShadow; + } + } + } + } +} + +void FsmGraph::resolveEpsilonTrans( MergeData &md ) +{ + /* Walk the state list and invoke recursive worker on each state. */ + for ( StateList::Iter st = stateList; st.lte(); st++ ) + epsilonFillEptVectFrom( st, st, false ); + + /* Prevent reading from and writing to of the same state. */ + shadowReadWriteStates( md ); + + /* For all states that have epsilon transitions out, draw the transitions, + * clear the epsilon transitions. */ + for ( StateList::Iter st = stateList; st.lte(); st++ ) { + /* If there is a state vector, then create the pre-merge state. */ + if ( st->eptVect != 0 ) { + /* Merge all the epsilon targets into the state. */ + for ( EptVect::Iter ept = *st->eptVect; ept.lte(); ept++ ) { + if ( ept->leaving ) + mergeStatesLeaving( md, st, ept->targ ); + else + mergeStates( md, st, ept->targ ); + } + + /* Clean up the target list. */ + delete st->eptVect; + st->eptVect = 0; + } + + /* Clear the epsilon transitions vector. */ + st->epsilonTrans.empty(); + } +} + +void FsmGraph::epsilonOp() +{ + /* For merging process. */ + MergeData md; + + setMisfitAccounting( true ); + + for ( StateList::Iter st = stateList; st.lte(); st++ ) + st->owningGraph = 0; + + /* Perform merges. */ + resolveEpsilonTrans( md ); + + /* Epsilons can caused merges which leave behind unreachable states. */ + fillInStates( md ); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +/* Make a new maching by joining together a bunch of machines without making + * any transitions between them. A negative finalId results in there being no + * final id. */ +void FsmGraph::joinOp( int startId, int finalId, FsmGraph **others, int numOthers ) +{ + /* For the merging process. */ + MergeData md; + + /* Set the owning machines. Start at one. Zero is reserved for the start + * and final states. */ + for ( StateList::Iter st = stateList; st.lte(); st++ ) + st->owningGraph = 1; + for ( int m = 0; m < numOthers; m++ ) { + for ( StateList::Iter st = others[m]->stateList; st.lte(); st++ ) + st->owningGraph = 2+m; + } + + /* All machines loose start state status. */ + unsetStartState(); + for ( int m = 0; m < numOthers; m++ ) + others[m]->unsetStartState(); + + /* Bring the other machines into this. */ + for ( int m = 0; m < numOthers; m++ ) { + /* Bring in the rest of other's entry points. */ + copyInEntryPoints( others[m] ); + others[m]->entryPoints.empty(); + + /* Merge the lists. This will move all the states from other into + * this. No states will be deleted. */ + stateList.append( others[m]->stateList ); + assert( others[m]->misfitList.length() == 0 ); + + /* Move the final set data from other into this. */ + finStateSet.insert( others[m]->finStateSet ); + others[m]->finStateSet.empty(); + + /* Since other's list is empty, we can delete the fsm without + * affecting any states. */ + delete others[m]; + } + + /* Look up the start entry point. */ + EntryMapEl *enLow = 0, *enHigh = 0; + bool findRes = entryPoints.findMulti( startId, enLow, enHigh ); + if ( ! findRes ) { + /* No start state. Set a default one and proceed with the join. Note + * that the result of the join will be a very uninteresting machine. */ + setStartState( addState() ); + } + else { + /* There is at least one start state, create a state that will become + * the new start state. */ + FsmState *newStart = addState(); + setStartState( newStart ); + + /* The start state is in an owning machine class all it's own. */ + newStart->owningGraph = 0; + + /* Create the set of states to merge from. */ + StateSet stateSet; + for ( EntryMapEl *en = enLow; en <= enHigh; en++ ) + stateSet.insert( en->value ); + + /* Merge in the set of start states into the new start state. */ + mergeStates( md, newStart, stateSet.data, stateSet.length() ); + } + + /* Take a copy of the final state set, before unsetting them all. This + * will allow us to call clearOutData on the states that don't get + * final state status back back. */ + StateSet finStateSetCopy = finStateSet; + + /* Now all final states are unset. */ + unsetAllFinStates(); + + if ( finalId >= 0 ) { + /* Create the implicit final state. */ + FsmState *finState = addState(); + setFinState( finState ); + + /* Assign an entry into the final state on the final state entry id. Note + * that there may already be an entry on this id. That's ok. Also set the + * final state owning machine id. It's in a class all it's own. */ + setEntry( finalId, finState ); + finState->owningGraph = 0; + } + + /* Hand over to workers for resolving epsilon trans. This will merge states + * with the targets of their epsilon transitions. */ + resolveEpsilonTrans( md ); + + /* Invoke the relinquish final callback on any states that did not get + * final state status back. */ + for ( StateSet::Iter st = finStateSetCopy; st.lte(); st++ ) { + if ( !((*st)->stateBits & SB_ISFINAL) ) + clearOutData( *st ); + } + + /* Fill in any new states made from merging. */ + fillInStates( md ); + + /* Joining can be messy. Instead of having misfit accounting on (which is + * tricky here) do a full cleaning. */ + removeUnreachableStates(); +} + +void FsmGraph::globOp( FsmGraph **others, int numOthers ) +{ + /* All other machines loose start states status. */ + for ( int m = 0; m < numOthers; m++ ) + others[m]->unsetStartState(); + + /* Bring the other machines into this. */ + for ( int m = 0; m < numOthers; m++ ) { + /* Bring in the rest of other's entry points. */ + copyInEntryPoints( others[m] ); + others[m]->entryPoints.empty(); + + /* Merge the lists. This will move all the states from other into + * this. No states will be deleted. */ + stateList.append( others[m]->stateList ); + assert( others[m]->misfitList.length() == 0 ); + + /* Move the final set data from other into this. */ + finStateSet.insert( others[m]->finStateSet ); + others[m]->finStateSet.empty(); + + /* Since other's list is empty, we can delete the fsm without + * affecting any states. */ + delete others[m]; + } +} + +void FsmGraph::deterministicEntry() +{ + /* For the merging process. */ + MergeData md; + + /* States may loose their entry points, turn on misfit accounting. */ + setMisfitAccounting( true ); + + /* Get a copy of the entry map then clear all the entry points. As we + * iterate the old entry map finding duplicates we will add the entry + * points for the new states that we create. */ + EntryMap prevEntry = entryPoints; + unsetAllEntryPoints(); + + for ( int enId = 0; enId < prevEntry.length(); ) { + /* Count the number of states on this entry key. */ + int highId = enId; + while ( highId < prevEntry.length() && prevEntry[enId].key == prevEntry[highId].key ) + highId += 1; + + int numIds = highId - enId; + if ( numIds == 1 ) { + /* Only a single entry point, just set the entry. */ + setEntry( prevEntry[enId].key, prevEntry[enId].value ); + } + else { + /* Multiple entry points, need to create a new state and merge in + * all the targets of entry points. */ + FsmState *newEntry = addState(); + for ( int en = enId; en < highId; en++ ) + mergeStates( md, newEntry, prevEntry[en].value ); + + /* Add the new state as the single entry point. */ + setEntry( prevEntry[enId].key, newEntry ); + } + + enId += numIds; + } + + /* The old start state may be unreachable. Remove the misfits and turn off + * misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +/* Unset any final states that are no longer to be final due to final bits. */ +void FsmGraph::unsetKilledFinals() +{ + /* Duplicate the final state set before we begin modifying it. */ + StateSet fin( finStateSet ); + + for ( int s = 0; s < fin.length(); s++ ) { + /* Check for killing bit. */ + FsmState *state = fin.data[s]; + if ( state->stateBits & SB_GRAPH1 ) { + /* One final state is a killer, set to non-final. */ + unsetFinState( state ); + } + + /* Clear all killing bits. Non final states should never have had those + * state bits set in the first place. */ + state->stateBits &= ~SB_GRAPH1; + } +} + +/* Unset any final states that are no longer to be final due to final bits. */ +void FsmGraph::unsetIncompleteFinals() +{ + /* Duplicate the final state set before we begin modifying it. */ + StateSet fin( finStateSet ); + + for ( int s = 0; s < fin.length(); s++ ) { + /* Check for one set but not the other. */ + FsmState *state = fin.data[s]; + if ( state->stateBits & SB_BOTH && + (state->stateBits & SB_BOTH) != SB_BOTH ) + { + /* One state wants the other but it is not there. */ + unsetFinState( state ); + } + + /* Clear wanting bits. Non final states should never have had those + * state bits set in the first place. */ + state->stateBits &= ~SB_BOTH; + } +} + +/* Ensure that the start state is free of entry points (aside from the fact + * that it is the start state). If the start state has entry points then Make a + * new start state by merging with the old one. Useful before modifying start + * transitions. If the existing start state has any entry points other than the + * start state entry then modifying its transitions changes more than the start + * transitions. So isolate the start state by separating it out such that it + * only has start stateness as it's entry point. */ +void FsmGraph::isolateStartState( ) +{ + /* For the merging process. */ + MergeData md; + + /* Bail out if the start state is already isolated. */ + if ( isStartStateIsolated() ) + return; + + /* Turn on misfit accounting to possibly catch the old start state. */ + setMisfitAccounting( true ); + + /* This will be the new start state. The existing start + * state is merged with it. */ + FsmState *prevStartState = startState; + unsetStartState(); + setStartState( addState() ); + + /* Merge the new start state with the old one to isolate it. */ + mergeStates( md, startState, prevStartState ); + + /* Stfil and stateDict will be empty because the merging of the old start + * state into the new one will not have any conflicting transitions. */ + assert( md.stateDict.treeSize == 0 ); + assert( md.stfillHead == 0 ); + + /* The old start state may be unreachable. Remove the misfits and turn off + * misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +#if COLM_LOG_CONDS +void logCondSpace( CondSpace *condSpace ) +{ + if ( condSpace == 0 ) + cerr << "<empty>"; + else { + for ( CondSet::Iter csi = condSpace->condSet.last(); csi.gtb(); csi-- ) { + if ( ! csi.last() ) + cerr << ','; + (*csi)->actionName( cerr ); + } + } +} + +void logNewExpansion( Expansion *exp ) +{ + cerr << "created expansion:" << endl; + cerr << " range: " << exp->lowKey.getVal() << " .. " << + exp->highKey.getVal() << endl; + + cerr << " fromCondSpace: "; + logCondSpace( exp->fromCondSpace ); + cerr << endl; + cerr << " fromVals: " << exp->fromVals << endl; + + cerr << " toCondSpace: "; + logCondSpace( exp->toCondSpace ); + cerr << endl; + cerr << " toValsList: "; + for ( LongVect::Iter to = exp->toValsList; to.lte(); to++ ) + cerr << " " << *to; + cerr << endl; +} +#endif + + +void FsmGraph::findTransExpansions( ExpansionList &expansionList, + FsmState *destState, FsmState *srcState ) +{ + PairIter<FsmTrans, StateCond> transCond( destState->outList.head, + srcState->stateCondList.head ); + for ( ; !transCond.end(); transCond++ ) { + if ( transCond.userState == RangeOverlap ) { + Expansion *expansion = new Expansion( transCond.s1Tel.lowKey, + transCond.s1Tel.highKey ); + expansion->fromTrans = new FsmTrans(*transCond.s1Tel.trans); + expansion->fromTrans->fromState = 0; + expansion->fromTrans->toState = transCond.s1Tel.trans->toState; + expansion->fromCondSpace = 0; + expansion->fromVals = 0; + CondSpace *srcCS = transCond.s2Tel.trans->condSpace; + expansion->toCondSpace = srcCS; + + long numTargVals = (1 << srcCS->condSet.length()); + for ( long targVals = 0; targVals < numTargVals; targVals++ ) + expansion->toValsList.append( targVals ); + + #ifdef COLM_LOG_CONDS + if ( colm_log_conds ) { + logNewExpansion( expansion ); + } + #endif + expansionList.append( expansion ); + } + } +} + +void FsmGraph::findCondExpInTrans( ExpansionList &expansionList, FsmState *state, + Key lowKey, Key highKey, CondSpace *fromCondSpace, CondSpace *toCondSpace, + long fromVals, LongVect &toValsList ) +{ + FsmTrans searchTrans; + searchTrans.lowKey = fromCondSpace->baseKey + fromVals * keyOps->alphSize() + + (lowKey - keyOps->minKey); + searchTrans.highKey = fromCondSpace->baseKey + fromVals * keyOps->alphSize() + + (highKey - keyOps->minKey); + searchTrans.prev = searchTrans.next = 0; + + PairIter<FsmTrans> pairIter( state->outList.head, &searchTrans ); + for ( ; !pairIter.end(); pairIter++ ) { + if ( pairIter.userState == RangeOverlap ) { + Expansion *expansion = new Expansion( lowKey, highKey ); + expansion->fromTrans = new FsmTrans(*pairIter.s1Tel.trans); + expansion->fromTrans->fromState = 0; + expansion->fromTrans->toState = pairIter.s1Tel.trans->toState; + expansion->fromCondSpace = fromCondSpace; + expansion->fromVals = fromVals; + expansion->toCondSpace = toCondSpace; + expansion->toValsList = toValsList; + + expansionList.append( expansion ); + #ifdef COLM_LOG_CONDS + if ( colm_log_conds ) { + logNewExpansion( expansion ); + } + #endif + } + } +} + +void FsmGraph::findCondExpansions( ExpansionList &expansionList, + FsmState *destState, FsmState *srcState ) +{ + PairIter<StateCond, StateCond> condCond( destState->stateCondList.head, + srcState->stateCondList.head ); + for ( ; !condCond.end(); condCond++ ) { + if ( condCond.userState == RangeOverlap ) { + /* Loop over all existing condVals . */ + CondSet &destCS = condCond.s1Tel.trans->condSpace->condSet; + long destLen = destCS.length(); + + /* Find the items in src cond set that are not in dest + * cond set. These are the items that we must expand. */ + CondSet srcOnlyCS = condCond.s2Tel.trans->condSpace->condSet; + for ( CondSet::Iter dcsi = destCS; dcsi.lte(); dcsi++ ) + srcOnlyCS.remove( *dcsi ); + long srcOnlyLen = srcOnlyCS.length(); + + if ( srcOnlyCS.length() > 0 ) { + #ifdef COLM_LOG_CONDS + if ( colm_log_conds ) { + cerr << "there are " << srcOnlyCS.length() << " item(s) that are " + "only in the srcCS" << endl; + } + #endif + + CondSet mergedCS = destCS; + mergedCS.insert( condCond.s2Tel.trans->condSpace->condSet ); + + CondSpace *fromCondSpace = addCondSpace( destCS ); + CondSpace *toCondSpace = addCondSpace( mergedCS ); + + /* Loop all values in the dest space. */ + for ( long destVals = 0; destVals < (1 << destLen); destVals++ ) { + long basicVals = 0; + for ( CondSet::Iter csi = destCS; csi.lte(); csi++ ) { + if ( destVals & (1 << csi.pos()) ) { + Action **cim = mergedCS.find( *csi ); + long bitPos = (cim - mergedCS.data); + basicVals |= 1 << bitPos; + } + } + + /* Loop all new values. */ + LongVect expandToVals; + for ( long soVals = 0; soVals < (1 << srcOnlyLen); soVals++ ) { + long targVals = basicVals; + for ( CondSet::Iter csi = srcOnlyCS; csi.lte(); csi++ ) { + if ( soVals & (1 << csi.pos()) ) { + Action **cim = mergedCS.find( *csi ); + long bitPos = (cim - mergedCS.data); + targVals |= 1 << bitPos; + } + } + expandToVals.append( targVals ); + } + + findCondExpInTrans( expansionList, destState, + condCond.s1Tel.lowKey, condCond.s1Tel.highKey, + fromCondSpace, toCondSpace, destVals, expandToVals ); + } + } + } + } +} + +void FsmGraph::doExpand( MergeData &md, FsmState *destState, ExpansionList &expList1 ) +{ + for ( ExpansionList::Iter exp = expList1; exp.lte(); exp++ ) { + for ( LongVect::Iter to = exp->toValsList; to.lte(); to++ ) { + long targVals = *to; + + /* We will use the copy of the transition that was made when the + * expansion was created. It will get used multiple times. Each + * time we must set up the keys, everything else is constant and + * and already prepared. */ + FsmTrans *srcTrans = exp->fromTrans; + + srcTrans->lowKey = exp->toCondSpace->baseKey + + targVals * keyOps->alphSize() + (exp->lowKey - keyOps->minKey); + srcTrans->highKey = exp->toCondSpace->baseKey + + targVals * keyOps->alphSize() + (exp->highKey - keyOps->minKey); + + TransList srcList; + srcList.append( srcTrans ); + outTransCopy( md, destState, srcList.head ); + srcList.abandon(); + } + } +} + + +void FsmGraph::doRemove( MergeData &md, FsmState *destState, ExpansionList &expList1 ) +{ + for ( ExpansionList::Iter exp = expList1; exp.lte(); exp++ ) { + Removal removal; + if ( exp->fromCondSpace == 0 ) { + removal.lowKey = exp->lowKey; + removal.highKey = exp->highKey; + } + else { + removal.lowKey = exp->fromCondSpace->baseKey + + exp->fromVals * keyOps->alphSize() + (exp->lowKey - keyOps->minKey); + removal.highKey = exp->fromCondSpace->baseKey + + exp->fromVals * keyOps->alphSize() + (exp->highKey - keyOps->minKey); + } + removal.next = 0; + + TransList destList; + PairIter<FsmTrans, Removal> pairIter( destState->outList.head, &removal ); + for ( ; !pairIter.end(); pairIter++ ) { + switch ( pairIter.userState ) { + case RangeInS1: { + FsmTrans *destTrans = pairIter.s1Tel.trans; + destTrans->lowKey = pairIter.s1Tel.lowKey; + destTrans->highKey = pairIter.s1Tel.highKey; + destList.append( destTrans ); + break; + } + case RangeInS2: + break; + case RangeOverlap: { + FsmTrans *trans = pairIter.s1Tel.trans; + detachTrans( trans->fromState, trans->toState, trans ); + delete trans; + break; + } + case BreakS1: { + pairIter.s1Tel.trans = dupTrans( destState, + pairIter.s1Tel.trans ); + break; + } + case BreakS2: + break; + } + } + destState->outList.transfer( destList ); + } +} + +void FsmGraph::mergeStateConds( FsmState *destState, FsmState *srcState ) +{ + StateCondList destList; + PairIter<StateCond> pairIter( destState->stateCondList.head, + srcState->stateCondList.head ); + for ( ; !pairIter.end(); pairIter++ ) { + switch ( pairIter.userState ) { + case RangeInS1: { + StateCond *destCond = pairIter.s1Tel.trans; + destCond->lowKey = pairIter.s1Tel.lowKey; + destCond->highKey = pairIter.s1Tel.highKey; + destList.append( destCond ); + break; + } + case RangeInS2: { + StateCond *newCond = new StateCond( *pairIter.s2Tel.trans ); + newCond->lowKey = pairIter.s2Tel.lowKey; + newCond->highKey = pairIter.s2Tel.highKey; + destList.append( newCond ); + break; + } + case RangeOverlap: { + StateCond *destCond = pairIter.s1Tel.trans; + StateCond *srcCond = pairIter.s2Tel.trans; + CondSet mergedCondSet; + mergedCondSet.insert( destCond->condSpace->condSet ); + mergedCondSet.insert( srcCond->condSpace->condSet ); + destCond->condSpace = addCondSpace( mergedCondSet ); + + destCond->lowKey = pairIter.s1Tel.lowKey; + destCond->highKey = pairIter.s1Tel.highKey; + destList.append( destCond ); + break; + } + case BreakS1: + pairIter.s1Tel.trans = new StateCond( *pairIter.s1Tel.trans ); + break; + + case BreakS2: + break; + } + } + destState->stateCondList.transfer( destList ); +} + +/* A state merge which represents the drawing in of leaving transitions. If + * there is any out data then we duplicate the souce state, transfer the out + * data, then merge in the state. The new state will be reaped because it will + * not be given any in transitions. */ +void FsmGraph::mergeStatesLeaving( MergeData &md, FsmState *destState, FsmState *srcState ) +{ + if ( !hasOutData( destState ) ) + mergeStates( md, destState, srcState ); + else { + FsmState *ssMutable = addState(); + mergeStates( md, ssMutable, srcState ); + transferOutData( ssMutable, destState ); + + for ( ActionSet::Iter cond = destState->outCondSet; cond.lte(); cond++ ) + embedCondition( md, ssMutable, *cond ); + + mergeStates( md, destState, ssMutable ); + } +} + +void FsmGraph::mergeStates( MergeData &md, FsmState *destState, + FsmState **srcStates, int numSrc ) +{ + for ( int s = 0; s < numSrc; s++ ) + mergeStates( md, destState, srcStates[s] ); +} + +void FsmGraph::mergeStates( MergeData &md, FsmState *destState, FsmState *srcState ) +{ + ExpansionList expList1; + ExpansionList expList2; + + findTransExpansions( expList1, destState, srcState ); + findCondExpansions( expList1, destState, srcState ); + findTransExpansions( expList2, srcState, destState ); + findCondExpansions( expList2, srcState, destState ); + + mergeStateConds( destState, srcState ); + + outTransCopy( md, destState, srcState->outList.head ); + + doExpand( md, destState, expList1 ); + doExpand( md, destState, expList2 ); + + doRemove( md, destState, expList1 ); + doRemove( md, destState, expList2 ); + + expList1.empty(); + expList2.empty(); + + /* Get its bits and final state status. */ + destState->stateBits |= ( srcState->stateBits & ~SB_ISFINAL ); + if ( srcState->isFinState() ) + setFinState( destState ); + + /* Draw in any properties of srcState into destState. */ + if ( srcState == destState ) { + /* Duplicate the list to protect against write to source. The + * priorities sets are not copied in because that would have no + * effect. */ + destState->epsilonTrans.append( EpsilonTrans( srcState->epsilonTrans ) ); + + /* Get all actions, duplicating to protect against write to source. */ + destState->toStateActionTable.setActions( + ActionTable( srcState->toStateActionTable ) ); + destState->fromStateActionTable.setActions( + ActionTable( srcState->fromStateActionTable ) ); + destState->outActionTable.setActions( ActionTable( srcState->outActionTable ) ); + destState->outCondSet.insert( ActionSet( srcState->outCondSet ) ); + destState->errActionTable.setActions( ErrActionTable( srcState->errActionTable ) ); + destState->eofActionTable.setActions( ActionTable( srcState->eofActionTable ) ); + } + else { + /* Get the epsilons, out priorities. */ + destState->epsilonTrans.append( srcState->epsilonTrans ); + destState->outPriorTable.setPriors( srcState->outPriorTable ); + + /* Get all actions. */ + destState->toStateActionTable.setActions( srcState->toStateActionTable ); + destState->fromStateActionTable.setActions( srcState->fromStateActionTable ); + destState->outActionTable.setActions( srcState->outActionTable ); + destState->outCondSet.insert( srcState->outCondSet ); + destState->errActionTable.setActions( srcState->errActionTable ); + destState->eofActionTable.setActions( srcState->eofActionTable ); + } +} + +void FsmGraph::fillInStates( MergeData &md ) +{ + /* Merge any states that are awaiting merging. This will likey cause + * other states to be added to the stfil list. */ + FsmState *state = md.stfillHead; + while ( state != 0 ) { + StateSet *stateSet = &state->stateDictEl->stateSet; + mergeStates( md, state, stateSet->data, stateSet->length() ); + state = state->alg.next; + } + + /* Delete the state sets of all states that are on the fill list. */ + state = md.stfillHead; + while ( state != 0 ) { + /* Delete and reset the state set. */ + delete state->stateDictEl; + state->stateDictEl = 0; + + /* Next state in the stfill list. */ + state = state->alg.next; + } + + /* StateDict will still have its ptrs/size set but all of it's element + * will be deleted so we don't need to clean it up. */ +} + +void FsmGraph::findEmbedExpansions( ExpansionList &expansionList, + FsmState *destState, Action *condAction ) +{ + StateCondList destList; + PairIter<FsmTrans, StateCond> transCond( destState->outList.head, + destState->stateCondList.head ); + for ( ; !transCond.end(); transCond++ ) { + switch ( transCond.userState ) { + case RangeInS1: { + if ( transCond.s1Tel.lowKey <= keyOps->maxKey ) { + assert( transCond.s1Tel.highKey <= keyOps->maxKey ); + + /* Make a new state cond. */ + StateCond *newStateCond = new StateCond( transCond.s1Tel.lowKey, + transCond.s1Tel.highKey ); + newStateCond->condSpace = addCondSpace( CondSet( condAction ) ); + destList.append( newStateCond ); + + /* Create the expansion. */ + Expansion *expansion = new Expansion( transCond.s1Tel.lowKey, + transCond.s1Tel.highKey ); + expansion->fromTrans = new FsmTrans(*transCond.s1Tel.trans); + expansion->fromTrans->fromState = 0; + expansion->fromTrans->toState = transCond.s1Tel.trans->toState; + expansion->fromCondSpace = 0; + expansion->fromVals = 0; + expansion->toCondSpace = newStateCond->condSpace; + expansion->toValsList.append( 1 ); + #ifdef COLM_LOG_CONDS + if ( colm_log_conds ) { + logNewExpansion( expansion ); + } + #endif + expansionList.append( expansion ); + } + break; + } + case RangeInS2: { + /* Enhance state cond and find the expansion. */ + StateCond *stateCond = transCond.s2Tel.trans; + stateCond->lowKey = transCond.s2Tel.lowKey; + stateCond->highKey = transCond.s2Tel.highKey; + + CondSet &destCS = stateCond->condSpace->condSet; + long destLen = destCS.length(); + CondSpace *fromCondSpace = stateCond->condSpace; + + CondSet mergedCS = destCS; + mergedCS.insert( condAction ); + CondSpace *toCondSpace = addCondSpace( mergedCS ); + stateCond->condSpace = toCondSpace; + destList.append( stateCond ); + + /* Loop all values in the dest space. */ + for ( long destVals = 0; destVals < (1 << destLen); destVals++ ) { + long basicVals = 0; + for ( CondSet::Iter csi = destCS; csi.lte(); csi++ ) { + if ( destVals & (1 << csi.pos()) ) { + Action **cim = mergedCS.find( *csi ); + long bitPos = (cim - mergedCS.data); + basicVals |= 1 << bitPos; + } + } + + long targVals = basicVals; + Action **cim = mergedCS.find( condAction ); + long bitPos = (cim - mergedCS.data); + targVals |= 1 << bitPos; + + LongVect expandToVals( targVals ); + findCondExpInTrans( expansionList, destState, + transCond.s2Tel.lowKey, transCond.s2Tel.highKey, + fromCondSpace, toCondSpace, destVals, expandToVals ); + } + break; + } + + + case RangeOverlap: + case BreakS1: + case BreakS2: + assert( false ); + break; + } + } + + destState->stateCondList.transfer( destList ); +} + +void FsmGraph::embedCondition( FsmState *state, Action *condAction ) +{ + MergeData md; + ExpansionList expList; + + /* Turn on misfit accounting to possibly catch the old start state. */ + setMisfitAccounting( true ); + + /* Worker. */ + embedCondition( md, state, condAction ); + + /* Fill in any states that were newed up as combinations of others. */ + fillInStates( md ); + + /* Remove the misfits and turn off misfit accounting. */ + removeMisfits(); + setMisfitAccounting( false ); +} + +void FsmGraph::embedCondition( MergeData &md, FsmState *state, Action *condAction ) +{ + ExpansionList expList; + + findEmbedExpansions( expList, state, condAction ); + doExpand( md, state, expList ); + doRemove( md, state, expList ); + expList.empty(); +} |