All Downloads are FREE. Search and download functionalities are using the official Maven repository.

cvc5-cvc5-1.2.0.src.proof.lfsc.lfsc_printer.cpp Maven / Gradle / Ivy

The newest version!
/******************************************************************************
 * Top contributors (to current version):
 *   Andrew Reynolds, Hans-Jörg Schurr, Abdalrhman Mohamed
 *
 * This file is part of the cvc5 project.
 *
 * Copyright (c) 2009-2024 by the authors listed in the file AUTHORS
 * in the top-level source directory and their institutional affiliations.
 * All rights reserved.  See the file COPYING in the top-level source
 * directory for licensing information.
 * ****************************************************************************
 *
 * The printer for LFSC proofs
 */

#include "proof/lfsc/lfsc_printer.h"

#include 

#include "expr/dtype.h"
#include "expr/dtype_cons.h"
#include "expr/dtype_selector.h"
#include "expr/node_algorithm.h"
#include "expr/skolem_manager.h"
#include "options/proof_options.h"
#include "proof/lfsc/lfsc_list_sc_node_converter.h"
#include "proof/lfsc/lfsc_print_channel.h"

using namespace cvc5::internal::kind;
using namespace cvc5::internal::rewriter;

namespace cvc5::internal {
namespace proof {

LfscPrinter::LfscPrinter(Env& env,
                         LfscNodeConverter& ltp,
                         rewriter::RewriteDb* rdb)
    : EnvObj(env),
      d_tproc(ltp),
      d_assumpCounter(0),
      d_trustChildPletCounter(0),
      d_termLetPrefix("t"),
      d_assumpPrefix("a"),
      d_pletPrefix("p"),
      d_pletTrustChildPrefix("q"),
      d_rdb(rdb)
{
  NodeManager* nm = nodeManager();
  d_boolType = nm->booleanType();
  // used for the `flag` type in LFSC
  d_tt = d_tproc.mkInternalSymbol("tt", d_boolType);
  d_ff = d_tproc.mkInternalSymbol("ff", d_boolType);
}

void LfscPrinter::print(std::ostream& out, const ProofNode* pn)
{
  Trace("lfsc-print-debug") << "; ORIGINAL PROOF: " << *pn << std::endl;
  Assert (!pn->getChildren().empty());
  // closing parentheses
  std::stringstream cparen;
  const std::vector& definitions = pn->getArguments();
  std::unordered_set definedSymbols;
  for (const Node& n : definitions)
  {
    definedSymbols.insert(n[0]);
    // Note that we don't have to convert it via the term processor (for the
    // sake of inferring declared symbols), since this is already done in the
    // lfsc post processor update method for the outermost SCOPE.
  }
  const std::vector& assertions = pn->getChildren()[0]->getArguments();
  const ProofNode* pnBody = pn->getChildren()[0]->getChildren()[0].get();

  // clear the rules we have warned about
  d_trustWarned.clear();

  // [1] convert assertions to internal and set up assumption map
  Trace("lfsc-print-debug") << "; print declarations" << std::endl;
  std::vector iasserts;
  std::map passumeMap;
  for (size_t i = 0, nasserts = assertions.size(); i < nasserts; i++)
  {
    Node a = assertions[i];
    iasserts.push_back(d_tproc.convert(a));
    // remember the assumption name
    passumeMap[a] = i;
  }
  d_assumpCounter = assertions.size();

  // [2] compute the proof letification
  Trace("lfsc-print-debug") << "; compute proof letification" << std::endl;
  std::vector pletList;
  std::map pletMap;
  computeProofLetification(pnBody, pletList, pletMap);

  // [3] compute the global term letification and declared symbols and types
  Trace("lfsc-print-debug")
      << "; compute global term letification and declared symbols" << std::endl;
  LetBinding lbind(d_termLetPrefix);
  for (const Node& ia : iasserts)
  {
    lbind.process(ia);
  }
  // We do a "dry-run" of proof printing here, using the LetBinding print
  // channel. This pass traverses the proof but does not print it, but instead
  // updates the let binding data structure for all nodes that appear anywhere
  // in the proof. It is also important for the term processor for collecting
  // symbols and types that are used in the proof.
  LfscPrintChannelPre lpcp(lbind);
  LetBinding emptyLetBind(d_termLetPrefix);
  std::map::iterator itp;
  for (const ProofNode* p : pletList)
  {
    itp = pletMap.find(p);
    Assert(itp != pletMap.end());
    size_t pid = itp->second;
    pletMap.erase(p);
    printProofInternal(&lpcp, p, emptyLetBind, pletMap, passumeMap);
    pletMap[p] = pid;
    Node resType = p->getResult();
    lbind.process(d_tproc.convert(resType));
  }
  // Print the body of the outermost scope
  printProofInternal(&lpcp, pnBody, emptyLetBind, pletMap, passumeMap);

  // [4] print declared sorts and symbols
  // [4a] user declare function symbols
  // Note that this is buffered into an output stream preambleSymDecl and then
  // printed after types. We require printing the declared symbols here so that
  // the set of collected declared types is complete at [4b].
  Trace("lfsc-print-debug") << "; print user symbols" << std::endl;
  std::stringstream preambleSymDecl;
  const std::unordered_set& syms = d_tproc.getDeclaredSymbols();
  for (const Node& s : syms)
  {
    TypeNode st = s.getType();
    if (st.isDatatypeConstructor() || st.isDatatypeSelector()
        || st.isDatatypeTester() || st.isDatatypeUpdater()
        || definedSymbols.find(s) != definedSymbols.cend())
    {
      // Constructors, selector, testers, updaters are defined by the datatype.
      // Some definitions depend on declarations and other definitions. So, we
      // print them in order after declarations.
      continue;
    }
    Node si = d_tproc.convert(s);
    preambleSymDecl << "(define " << si << " (var "
                    << d_tproc.getOrAssignIndexForFVar(s) << " ";
    printType(preambleSymDecl, st);
    preambleSymDecl << "))" << std::endl;
  }
  // Note that definitions always use their own internal letification, since
  // their bodies are not part of the main proof. It is possible to share term
  // letification via global definitions, however, this requires further
  // analysis to ensure symbols are printed in the correct order. This is
  // not done for simplicity.
  for (const Node& def : definitions)
  {
    Node si = d_tproc.convert(def[0]);
    preambleSymDecl << "(define " << si << ' ';
    print(preambleSymDecl, def[1]);
    preambleSymDecl << ')' << std::endl;
  }
  // [4b] user declared sorts
  Trace("lfsc-print-debug") << "; print user sorts" << std::endl;
  std::stringstream preamble;
  std::unordered_set sts;
  std::unordered_set tupleArity;
  // get the types from the term processor, which has seen all terms occurring
  // in the proof at this point
  // The for loop below may add elements to the set of declared types, so we
  // copy the set to ensure that the for loop iterators do not become outdated.
  const std::unordered_set types = d_tproc.getDeclaredTypes();
  for (const TypeNode& st : types)
  {
    // note that we must get all "component types" of a type, so that
    // e.g. U is printed as a sort declaration when we have type (Array U Int).
    ensureTypeDefinitionPrinted(preamble, st, sts, tupleArity);
  }
  // print datatype definitions for the above sorts
  for (const TypeNode& stc : sts)
  {
    if (!stc.isDatatype() || stc.getKind() == Kind::PARAMETRIC_DATATYPE)
    {
      // skip the instance of a parametric datatype
      continue;
    }
    const DType& dt = stc.getDType();
    preamble << "; DATATYPE " << dt.getName() << std::endl;
    NodeManager* nm = nodeManager();
    for (size_t i = 0, ncons = dt.getNumConstructors(); i < ncons; i++)
    {
      const DTypeConstructor& cons = dt[i];
      std::string cname = d_tproc.getNameForUserNameOf(cons.getConstructor());
      Node cc = nm->mkRawSymbol(cname, stc);
      // print constructor/tester
      preamble << "(declare " << cc << " term)" << std::endl;
      for (size_t j = 0, nargs = cons.getNumArgs(); j < nargs; j++)
      {
        const DTypeSelector& arg = cons[j];
        // print selector
        std::string sname = d_tproc.getNameForUserNameOf(arg.getSelector());
        Node sc = nm->mkRawSymbol(sname, stc);
        preamble << "(declare " << sc << " term)" << std::endl;
      }
    }
    // testers and updaters are instances of parametric symbols
    // shared selectors are instance of parametric symbol "sel"
    preamble << "; END DATATYPE " << std::endl;
  }
  // [4c] user declared function symbols
  preamble << preambleSymDecl.str();

  // [5] print warnings
  for (ProofRule r : d_trustWarned)
  {
    out << "; WARNING: adding trust step for " << r << std::endl;
  }

  // [6] print the DSL rewrite rule declarations
  const std::unordered_set& dslrs = lpcp.getDslRewrites();
  for (ProofRewriteRule dslr : dslrs)
  {
    // also computes the format for the rule
    printDslRule(out, dslr, d_dslFormat[dslr]);
  }

  // [7] print the check command and term lets
  out << preamble.str();
  if (options().proof.lfscFlatten)
  {
    // print term lets as definitions
    std::stringstream cparenTmp;
    printLetList(out, cparenTmp, lbind, true);
  }
  else
  {
    // the outer check statement for the main proof
    out << "(check" << std::endl;
    cparen << ")";
    // print the term let list wrapped around the body of the final proof
    printLetList(out, cparen, lbind, false);
  }

  Trace("lfsc-print-debug") << "; print asserts" << std::endl;
  // [8] print the assertions, with letification
  // the assumption identifier mapping
  for (size_t i = 0, nasserts = iasserts.size(); i < nasserts; i++)
  {
    Node ia = iasserts[i];
    if (options().proof.lfscFlatten)
    {
      out << "(declare ";
      LfscPrintChannelOut::printId(out, i, d_assumpPrefix);
      out << " (holds ";
      printInternal(out, ia, lbind);
      out << "))" << std::endl;
    }
    else
    {
      out << "(# ";
      LfscPrintChannelOut::printId(out, i, d_assumpPrefix);
      out << " (holds ";
      printInternal(out, ia, lbind);
      out << ")" << std::endl;
      cparen << ")";
    }
  }

  Trace("lfsc-print-debug") << "; print annotation" << std::endl;
  // [9] print the annotation
  if (!options().proof.lfscFlatten)
  {
    out << "(: (holds false)" << std::endl;
    cparen << ")";
  }

  Trace("lfsc-print-debug") << "; print proof body" << std::endl;
  // [10] print the proof body
  Assert(pn->getRule() == ProofRule::SCOPE);
  // the outermost scope can be ignored (it is the scope of the assertions,
  // which are already printed above).
  LfscPrintChannelOut lout(out);

  if (options().proof.lfscFlatten)
  {
    // print the proof letification as separate check statements, followed
    // by the main proof.
    for (size_t i = 0; i <= pletList.size(); i++)
    {
      bool isFinal = (i == pletList.size());
      const ProofNode* p = isFinal ? pnBody : pletList[i];
      Node res = p->getResult();
      std::stringstream resType;
      printInternal(resType, d_tproc.convert(res), lbind);
      out << "(check (: (holds " << resType.str() << ")" << std::endl;
      // print the letified proof
      if (isFinal)
      {
        printProofInternal(&lout, p, lbind, pletMap, passumeMap);
        out << "))" << std::endl;
      }
      else
      {
        itp = pletMap.find(p);
        Assert(itp != pletMap.end());
        size_t pid = itp->second;
        pletMap.erase(p);
        printProofInternal(&lout, p, lbind, pletMap, passumeMap);
        pletMap[p] = pid;
        out << "))" << std::endl;
        out << "(declare ";
        LfscPrintChannelOut::printId(out, pid, d_pletPrefix);
        out << " (holds " << resType.str() << "))" << std::endl;
      }
    }
  }
  else
  {
    printProofLetify(&lout, pnBody, lbind, pletList, pletMap, passumeMap);
  }
  // [11] print closing parantheses
  out << cparen.str() << std::endl;
}

void LfscPrinter::ensureTypeDefinitionPrinted(
    std::ostream& os,
    TypeNode tn,
    std::unordered_set& processed,
    std::unordered_set& tupleArityProcessed)
{
  // note that we must get all "component types" of a type, so that
  // e.g. U is printed as a sort declaration when we have type (Array U Int).
  std::unordered_set ctypes;
  expr::getComponentTypes(tn, ctypes);

  for (const TypeNode& stc : ctypes)
  {
    printTypeDefinition(os, stc, processed, tupleArityProcessed);
  }
}

void LfscPrinter::printTypeDefinition(
    std::ostream& os,
    TypeNode tn,
    std::unordered_set& processed,
    std::unordered_set& tupleArityProcessed)
{
  if (processed.find(tn) != processed.end())
  {
    return;
  }
  processed.insert(tn);
  // print uninterpreted sorts and uninterpreted sort constructors here
  if (tn.getKind() == Kind::SORT_TYPE)
  {
    os << "(declare ";
    printType(os, tn);
    uint64_t arity = 0;
    if (tn.isUninterpretedSortConstructor())
    {
      arity = tn.getUninterpretedSortConstructorArity();
    }
    std::stringstream tcparen;
    for (uint64_t i = 0; i < arity; i++)
    {
      os << " (! s" << i << " sort";
      tcparen << ")";
    }
    os << " sort" << tcparen.str() << ")" << std::endl;
  }
  else if (tn.isDatatype())
  {
    const DType& dt = tn.getDType();
    if (tn.getKind() == Kind::PARAMETRIC_DATATYPE || dt.isNullable())
    {
      // skip the instance of a parametric datatype
      // nullables don't need printing
      return;
    }
    if (dt.isTuple())
    {
      const DTypeConstructor& cons = dt[0];
      size_t arity = cons.getNumArgs();
      if (tupleArityProcessed.find(arity) == tupleArityProcessed.end())
      {
        tupleArityProcessed.insert(arity);
        if (arity>0)
        {
          os << "(declare Tuple";
          os << "_" << arity;
        }
        else
        {
          os << "(declare UnitTuple";
        }
        os << " ";
        std::stringstream tcparen;
        for (size_t j = 0, nargs = cons.getNumArgs(); j < nargs; j++)
        {
          os << "(! s" << j << " sort ";
          tcparen << ")";
        }
        os << "sort" << tcparen.str() << ")";
      }
      os << std::endl;
    }
    else
    {
      os << "(declare ";
      printType(os, tn);
      std::stringstream cdttparens;
      if (dt.isParametric())
      {
        std::vector params = dt.getParameters();
        for (const TypeNode& p : params)
        {
          os << " (! " << p << " sort";
          cdttparens << ")";
        }
      }
      os << " sort)" << cdttparens.str() << std::endl;
    }
    // must also ensure the subfield types of the datatype are printed
    std::unordered_set sftypes = dt.getSubfieldTypes();
    for (const TypeNode& sft : sftypes)
    {
      ensureTypeDefinitionPrinted(os, sft, processed, tupleArityProcessed);
    }
  }
  // all other sorts are builtin into the LFSC signature
}

void LfscPrinter::printProofLetify(
    LfscPrintChannel* out,
    const ProofNode* pn,
    const LetBinding& lbind,
    const std::vector& pletList,
    std::map& pletMap,
    std::map& passumeMap)
{
  // closing parentheses
  size_t cparen = 0;

  // define the let proofs
  if (!pletList.empty())
  {
    std::map::iterator itp;
    for (const ProofNode* p : pletList)
    {
      itp = pletMap.find(p);
      Assert(itp != pletMap.end());
      size_t pid = itp->second;
      pletMap.erase(p);
      printPLet(out, p, pid, d_pletPrefix, lbind, pletMap, passumeMap);
      pletMap[p] = pid;
      // printPLet opens two parentheses
      cparen = cparen + 2;
    }
    out->printEndLine();
  }

  // [2] print the proof body
  printProofInternal(out, pn, lbind, pletMap, passumeMap);

  // print the closing parenthesis
  out->printCloseRule(cparen);
}

void LfscPrinter::printPLet(LfscPrintChannel* out,
                            const ProofNode* p,
                            size_t pid,
                            const std::string& prefix,
                            const LetBinding& lbind,
                            const std::map& pletMap,
                            std::map& passumeMap)
{
  // print (plet _ _
  out->printOpenLfscRule(LfscRule::PLET);
  out->printHole();
  out->printHole();
  out->printEndLine();
  // print the letified proof
  printProofInternal(out, p, lbind, pletMap, passumeMap);
  // print the lambda (\ __pX
  out->printOpenLfscRule(LfscRule::LAMBDA);
  out->printId(pid, prefix);
  out->printEndLine();
}

void LfscPrinter::printProofInternal(
    LfscPrintChannel* out,
    const ProofNode* pn,
    const LetBinding& lbind,
    const std::map& pletMap,
    std::map& passumeMap)
{
  // the stack
  std::vector visit;
  // whether we have to process children
  std::unordered_set processingChildren;
  // helper iterators
  std::unordered_set::iterator pit;
  std::map::const_iterator pletIt;
  std::map::iterator passumeIt;
  Node curn;
  TypeNode curtn;
  const ProofNode* cur;
  visit.push_back(PExpr(pn));
  do
  {
    curn = visit.back().d_node;
    curtn = visit.back().d_typeNode;
    cur = visit.back().d_pnode;
    visit.pop_back();
    // case 1: printing a proof
    if (cur != nullptr)
    {
      ProofRule r = cur->getRule();
      // maybe it is letified
      pletIt = pletMap.find(cur);
      if (pletIt != pletMap.end())
      {
        // a letified proof
        out->printId(pletIt->second, d_pletPrefix);
        continue;
      }
      pit = processingChildren.find(cur);
      if (pit == processingChildren.end())
      {
        bool isLambda = false;
        if (r == ProofRule::LFSC_RULE)
        {
          Assert(!cur->getArguments().empty());
          LfscRule lr = getLfscRule(cur->getArguments()[0]);
          isLambda = (lr == LfscRule::LAMBDA);
        }
        if (r == ProofRule::ASSUME)
        {
          // an assumption, must have a name
          passumeIt = passumeMap.find(cur->getResult());
          Assert(passumeIt != passumeMap.end());
          out->printId(passumeIt->second, d_assumpPrefix);
        }
        else if (r == ProofRule::ENCODE_EQ_INTRO)
        {
          // just add child
          visit.push_back(PExpr(cur->getChildren()[0].get()));
        }
        else if (isLambda)
        {
          Assert(cur->getArguments().size() == 3);
          // lambdas are handled specially. We print in a self contained way
          // here.
          // allocate an assumption, if necessary
          size_t pid;
          Node assumption = cur->getArguments()[2];
          passumeIt = passumeMap.find(assumption);
          if (passumeIt == passumeMap.end())
          {
            pid = d_assumpCounter;
            d_assumpCounter++;
            passumeMap[assumption] = pid;
          }
          else
          {
            pid = passumeIt->second;
          }
          // make the node whose name is the assumption id, where notice that
          // the type of this node does not matter
          std::stringstream pidNodeName;
          LfscPrintChannelOut::printId(pidNodeName, pid, d_assumpPrefix);
          // must be an internal symbol so that it is not turned into (bvar ...)
          Node pidNode =
              d_tproc.mkInternalSymbol(pidNodeName.str(), d_boolType);
          // print "(\ "
          out->printOpenRule(cur);
          // print the identifier
          out->printNode(pidNode);
          // Print the body of the proof with a fresh proof letification. We can
          // keep the assumption map and the let binding (for terms).
          std::vector pletListNested;
          std::map pletMapNested;
          const ProofNode* curBody = cur->getChildren()[0].get();
          computeProofLetification(curBody, pletListNested, pletMapNested);
          printProofLetify(
              out, curBody, lbind, pletListNested, pletMapNested, passumeMap);
          // print ")"
          out->printCloseRule();
        }
        else
        {
          // assert that we should traverse cur when letifying
          Assert(d_lpltc.shouldTraverse(cur));
          // a normal rule application, compute the proof arguments, which
          // notice in the case of PI also may modify our passumeMap.
          std::vector args;
          if (computeProofArgs(cur, args))
          {
            processingChildren.insert(cur);
            // will revisit this proof node to close parentheses
            visit.push_back(PExpr(cur));
            std::reverse(args.begin(), args.end());
            visit.insert(visit.end(), args.begin(), args.end());
            // print the rule name
            out->printOpenRule(cur);
          }
          else
          {
            // Could not print the rule, trust for now.
            // If we are expanding trusted steps, its children are printed as
            // plet applications that wrap this term, so that all subproofs are
            // recorded in the proof.
            size_t cparenTrustChild = 0;
            if (options().proof.lfscExpandTrust)
            {
              const std::vector>& children =
                  cur->getChildren();
              for (const std::shared_ptr& c : children)
              {
                size_t pid = d_trustChildPletCounter;
                d_trustChildPletCounter++;
                printPLet(out,
                          c.get(),
                          pid,
                          d_pletTrustChildPrefix,
                          lbind,
                          pletMap,
                          passumeMap);
                cparenTrustChild = cparenTrustChild + 2;
              }
            }
            Node res = d_tproc.convert(cur->getResult());
            res = lbind.convert(res, true);
            out->printTrust(res, r);
            d_trustWarned.insert(r);
            out->printCloseRule(cparenTrustChild);
          }
        }
      }
      else
      {
        processingChildren.erase(cur);
        out->printCloseRule();
      }
    }
    // case 2: printing a node
    else if (!curn.isNull())
    {
      // it has already been converted to internal form, we letify it here
      Node curni = lbind.convert(curn, true);
      out->printNode(curni);
    }
    // case 3: printing a type node
    else if (!curtn.isNull())
    {
      out->printTypeNode(curtn);
    }
    // case 4: a hole
    else
    {
      out->printHole();
    }
  } while (!visit.empty());
}

bool LfscPrinter::computeProofArgs(const ProofNode* pn,
                                   std::vector& pargs)
{
  const std::vector>& children = pn->getChildren();
  std::vector cs;
  for (const std::shared_ptr& c : children)
  {
    cs.push_back(c.get());
  }
  ProofRule r = pn->getRule();
  const std::vector& args = pn->getArguments();
  std::vector as;
  for (const Node& a : args)
  {
    Node ac = d_tproc.convert(a);
    Assert(!ac.isNull());
    as.push_back(ac);
  }
  // The proof expression stream, which packs the next expressions (proofs,
  // terms, sorts, LFSC datatypes) into a print-expression vector pargs. This
  // stream can be used via "pf << e" which appends an expression to the
  // vector maintained by this stream.
  PExprStream pf(pargs, d_tt, d_ff);
  // hole
  PExpr h;
  Trace("lfsc-print-debug2")
      << "Compute proof args " << r << " #children= " << cs.size()
      << " #args=" << as.size() << std::endl;
  switch (r)
  {
    // SAT
    case ProofRule::RESOLUTION:
      pf << h << h << h << cs[0] << cs[1] << args[0].getConst() << as[1];
      break;
    case ProofRule::REORDERING: pf << h << as[0] << cs[0]; break;
    case ProofRule::FACTORING: pf << h << h << cs[0]; break;
    // Boolean
    case ProofRule::SPLIT: pf << as[0]; break;
    case ProofRule::NOT_NOT_ELIM: pf << h << cs[0]; break;
    case ProofRule::CONTRA: pf << h << cs[0] << cs[1]; break;
    case ProofRule::MODUS_PONENS:
    case ProofRule::EQ_RESOLVE: pf << h << h << cs[0] << cs[1]; break;
    case ProofRule::NOT_AND: pf << h << h << cs[0]; break;
    case ProofRule::NOT_OR_ELIM:
    case ProofRule::AND_ELIM: pf << h << h << args[0] << cs[0]; break;
    case ProofRule::IMPLIES_ELIM:
    case ProofRule::NOT_IMPLIES_ELIM1:
    case ProofRule::NOT_IMPLIES_ELIM2:
    case ProofRule::EQUIV_ELIM1:
    case ProofRule::EQUIV_ELIM2:
    case ProofRule::NOT_EQUIV_ELIM1:
    case ProofRule::NOT_EQUIV_ELIM2:
    case ProofRule::XOR_ELIM1:
    case ProofRule::XOR_ELIM2:
    case ProofRule::NOT_XOR_ELIM1:
    case ProofRule::NOT_XOR_ELIM2: pf << h << h << cs[0]; break;
    case ProofRule::ITE_ELIM1:
    case ProofRule::ITE_ELIM2:
    case ProofRule::NOT_ITE_ELIM1:
    case ProofRule::NOT_ITE_ELIM2: pf << h << h << h << cs[0]; break;
    // CNF
    case ProofRule::CNF_AND_POS:
    case ProofRule::CNF_OR_NEG:
      // print second argument as a raw integer (mpz)
      pf << h << as[0] << args[1];
      break;
    case ProofRule::CNF_AND_NEG: pf << h << as[0]; break;
    case ProofRule::CNF_OR_POS:
      pf << as[0];
      break;
      break;
    case ProofRule::CNF_IMPLIES_POS:
    case ProofRule::CNF_IMPLIES_NEG1:
    case ProofRule::CNF_IMPLIES_NEG2:
    case ProofRule::CNF_EQUIV_POS1:
    case ProofRule::CNF_EQUIV_POS2:
    case ProofRule::CNF_EQUIV_NEG1:
    case ProofRule::CNF_EQUIV_NEG2:
    case ProofRule::CNF_XOR_POS1:
    case ProofRule::CNF_XOR_POS2:
    case ProofRule::CNF_XOR_NEG1:
    case ProofRule::CNF_XOR_NEG2: pf << as[0][0] << as[0][1]; break;
    case ProofRule::CNF_ITE_POS1:
    case ProofRule::CNF_ITE_POS2:
    case ProofRule::CNF_ITE_POS3:
    case ProofRule::CNF_ITE_NEG1:
    case ProofRule::CNF_ITE_NEG2:
    case ProofRule::CNF_ITE_NEG3: pf << as[0][0] << as[0][1] << as[0][2]; break;
    // equality
    case ProofRule::REFL: pf << as[0]; break;
    case ProofRule::SYMM: pf << h << h << cs[0]; break;
    case ProofRule::TRANS: pf << h << h << h << cs[0] << cs[1]; break;
    case ProofRule::TRUE_INTRO:
    case ProofRule::FALSE_INTRO:
    case ProofRule::TRUE_ELIM:
    case ProofRule::FALSE_ELIM: pf << h << cs[0]; break;
    // arithmetic
    case ProofRule::ARITH_MULT_POS:
    case ProofRule::ARITH_MULT_NEG:
    {
      pf << h << as[0] << as[1] << d_tproc.convertType(as[0].getType());
    }
    break;
    case ProofRule::ARITH_TRICHOTOMY:
    {
      // should be robust to different orderings
      pf << h << h << h << cs[0] << cs[1];
    }
    break;
    case ProofRule::INT_TIGHT_UB:
    case ProofRule::INT_TIGHT_LB:
    {
      Node res = pn->getResult();
      Assert(res.getNumChildren() == 2);
      Assert(res[1].isConst());
      pf << h << h << d_tproc.convert(res[1]) << cs[0];
    }
    break;
    // strings
    case ProofRule::STRING_LENGTH_POS:
      pf << as[0] << d_tproc.convertType(as[0].getType()) << h;
      break;
    case ProofRule::STRING_LENGTH_NON_EMPTY: pf << h << h << cs[0]; break;
    case ProofRule::RE_INTER: pf << h << h << h << cs[0] << cs[1]; break;
    case ProofRule::CONCAT_EQ:
      pf << h << h << h << args[0].getConst()
         << d_tproc.convertType(children[0]->getResult()[0].getType()) << cs[0];
      break;
    case ProofRule::CONCAT_UNIFY:
      pf << h << h << h << h << args[0].getConst()
         << d_tproc.convertType(children[0]->getResult()[0].getType()) << cs[0]
         << cs[1];
      break;
    case ProofRule::CONCAT_CSPLIT:
      pf << h << h << h << h << args[0].getConst()
         << d_tproc.convertType(children[0]->getResult()[0].getType()) << cs[0]
         << cs[1];
      break;
    case ProofRule::CONCAT_CONFLICT:
      pf << h << h << args[0].getConst()
         << d_tproc.convertType(children[0]->getResult()[0].getType()) << cs[0];
      break;
    case ProofRule::RE_UNFOLD_POS:
      if (children[0]->getResult()[1].getKind() != Kind::REGEXP_CONCAT)
      {
        return false;
      }
      pf << h << h << h << cs[0];
      break;
    case ProofRule::STRING_EAGER_REDUCTION:
    {
      Kind k = as[0].getKind();
      if (k == Kind::STRING_TO_CODE || k == Kind::STRING_CONTAINS
          || k == Kind::STRING_INDEXOF)
      {
        pf << h << as[0] << as[0][0].getType();
      }
      else
      {
        // not yet defined for other kinds
        return false;
      }
    }
    break;
    case ProofRule::STRING_REDUCTION:
    {
      Kind k = as[0].getKind();
      if (k == Kind::STRING_SUBSTR || k == Kind::STRING_INDEXOF)
      {
        pf << h << as[0] << d_tproc.convertType(as[0][0].getType());
      }
      else
      {
        // not yet defined for other kinds
        return false;
      }
    }
    break;
    // quantifiers
    case ProofRule::SKOLEM_INTRO:
    {
      pf << d_tproc.convert(SkolemManager::getUnpurifiedForm(args[0]));
    }
    break;
    // ---------- arguments of non-translated rules go here
    case ProofRule::LFSC_RULE:
    {
      LfscRule lr = getLfscRule(args[0]);
      // lambda should be processed elsewhere
      Assert(lr != LfscRule::LAMBDA);
      // Note that `args` has 2 builtin arguments, thus the first real argument
      // begins at index 2
      switch (lr)
      {
        case LfscRule::DEFINITION: pf << as[1][0]; break;
        case LfscRule::SCOPE: pf << h << as[2] << cs[0]; break;
        case LfscRule::NEG_SYMM: pf << h << h << cs[0]; break;
        case LfscRule::CONG: pf << h << h << h << h << cs[0] << cs[1]; break;
        case LfscRule::AND_INTRO1: pf << h << cs[0]; break;
        case LfscRule::NOT_AND_REV: pf << h << h << cs[0]; break;
        case LfscRule::PROCESS_SCOPE: pf << h << h << as[2] << cs[0]; break;
        case LfscRule::AND_INTRO2: pf << h << h << cs[0] << cs[1]; break;
        case LfscRule::ARITH_SUM_UB: pf << h << h << h << cs[0] << cs[1]; break;
        case LfscRule::CONCAT_CONFLICT_DEQ:
          pf << h << h << h << h << as[2].getConst()
             << d_tproc.convertType(children[0]->getResult()[0].getType())
             << cs[0] << cs[1];
          break;
        case LfscRule::INSTANTIATE:
          pf << h << h << h << h << as[2] << cs[0];
          break;
        case LfscRule::BETA_REDUCE: pf << h << as[2]; break;
        default: return false; break;
      }
    }
    break;
    case ProofRule::DSL_REWRITE:
    {
      ProofRewriteRule di = ProofRewriteRule::NONE;
      if (!rewriter::getRewriteRule(args[0], di))
      {
        Assert(false);
      }
      Trace("lfsc-print-debug2") << "Printing dsl rule " << di << std::endl;
      const rewriter::RewriteProofRule& rpr = d_rdb->getRule(di);
      const std::vector& varList = rpr.getVarList();
      Assert(as.size() == varList.size() + 1);
      // print holes/terms based on whether variables are explicit
      for (size_t i = 1, nargs = as.size(); i < nargs; i++)
      {
        Node v = varList[i - 1];
        if (rpr.isExplicitVar(v))
        {
          // If the variable is a list variable, we must convert its value to
          // the proper term. This is based on its context.
          if (as[i].getKind() == Kind::SEXPR)
          {
            Assert(args[i].getKind() == Kind::SEXPR);
            NodeManager* nm = nodeManager();
            Kind k = rpr.getListContext(v);
            // notice we use d_tproc.getNullTerminator and not
            // expr::getNullTerminator here, which has subtle differences
            // e.g. re.empty vs (str.to_re "").
            Node null = d_tproc.getNullTerminator(k, v.getType());
            Node t;
            if (as[i].getNumChildren() == 1)
            {
              // Singleton list uses null terminator. We first construct an
              // original term and convert it.
              Node tt = nm->mkNode(k, args[i][0], null);
              tt = d_tproc.convert(tt);
              // Since conversion adds a null terminator, we have that
              // tt is of the form (f t (f null null)). We reconstruct
              // the proper term (f t null) below.
              Assert(tt.getNumChildren() == 2);
              Assert(tt[1].getNumChildren() == 2);
              std::vector tchildren;
              if (tt.getMetaKind() == metakind::PARAMETERIZED)
              {
                tchildren.push_back(tt.getOperator());
              }
              tchildren.push_back(tt[0]);
              tchildren.push_back(tt[1][0]);
              t = nm->mkNode(tt.getKind(), tchildren);
            }
            else
            {
              if (k == Kind::UNDEFINED_KIND)
              {
                Unhandled() << "Unknown context for list variable " << v
                            << " in rule " << di;
              }
              if (as[i].getNumChildren() == 0)
              {
                t = null;
              }
              else
              {
                // re-convert it
                std::vector vec(args[i].begin(), args[i].end());
                t = nm->mkNode(k, vec);
              }
              t = d_tproc.convert(t);
            }
            pf << t;
          }
          else
          {
            pf << as[i];
          }
        }
        else
        {
          pf << h;
        }
      }
      // print child proofs, which is based on the format computed for the rule
      size_t ccounter = 0;
      std::map>::iterator itf =
          d_dslFormat.find(di);
      if (itf == d_dslFormat.end())
      {
        // We may not have computed the format yet, e.g. if we are printing
        // via the pre print channel. In this case, just print all the children.
        for (const ProofNode* c : cs)
        {
          pf << c;
        }
      }
      else
      {
        for (const Node& f : itf->second)
        {
          if (f.isNull())
          {
            // this position is a hole
            pf << h;
            continue;
          }
          Assert(ccounter < cs.size());
          pf << cs[ccounter];
          ccounter++;
        }
        Assert(ccounter == cs.size());
      }
    }
    break;
    default:
    {
      return false;
      break;
    }
  }
  return true;
}

void LfscPrinter::computeProofLetification(
    const ProofNode* pn,
    std::vector& pletList,
    std::map& pletMap)
{
  // use callback to specify to stop at LAMBDA
  ProofLetify::computeProofLet(pn, pletList, pletMap, 2, &d_lpltc);
}

void LfscPrinter::print(std::ostream& out, Node n)
{
  Node ni = d_tproc.convert(n);
  printLetify(out, ni);
}

void LfscPrinter::printLetify(std::ostream& out, Node n)
{
  // closing parentheses
  std::stringstream cparen;

  LetBinding lbind(d_termLetPrefix);
  lbind.process(n);

  // [1] print the letification
  printLetList(out, cparen, lbind);

  // [2] print the body
  printInternal(out, n, lbind);

  out << cparen.str();
}

void LfscPrinter::printLetList(std::ostream& out,
                               std::ostream& cparen,
                               LetBinding& lbind,
                               bool asDefs)
{
  std::vector letList;
  lbind.letify(letList);
  std::map::const_iterator it;
  for (size_t i = 0, nlets = letList.size(); i < nlets; i++)
  {
    Node nl = letList[i];
    size_t id = lbind.getId(nl);
    Assert(id != 0);
    if (asDefs)
    {
      out << "(define ";
      LfscPrintChannelOut::printId(out, id, d_termLetPrefix);
      out << " ";
      // do not letify the top term
      printInternal(out, nl, lbind, false);
      out << ")" << std::endl;
    }
    else
    {
      out << "(@ ";
      LfscPrintChannelOut::printId(out, id, d_termLetPrefix);
      out << " ";
      // do not letify the top term
      printInternal(out, nl, lbind, false);
      out << std::endl;
      cparen << ")";
    }
  }
}

void LfscPrinter::printInternal(std::ostream& out, Node n)
{
  LetBinding lbind(d_termLetPrefix);
  printInternal(out, n, lbind);
}
void LfscPrinter::printInternal(std::ostream& out,
                                Node n,
                                LetBinding& lbind,
                                bool letTop)
{
  Node nc = lbind.convert(n, letTop);
  LfscPrintChannelOut::printNodeInternal(out, nc);
}

void LfscPrinter::printType(std::ostream& out, TypeNode tn)
{
  TypeNode tni = d_tproc.convertType(tn);
  LfscPrintChannelOut::printTypeNodeInternal(out, tni);
}

void LfscPrinter::printDslRule(std::ostream& out,
                               ProofRewriteRule id,
                               std::vector& format)
{
  const rewriter::RewriteProofRule& rpr = d_rdb->getRule(id);
  const std::vector& varList = rpr.getVarList();
  const std::vector& uvarList = rpr.getUserVarList();
  const std::vector& conds = rpr.getConditions();
  Node conc = rpr.getConclusion();

  std::stringstream oscs;
  std::stringstream odecl;

  std::stringstream rparen;
  odecl << "(declare ";
  LfscPrintChannelOut::printProofRewriteRule(odecl, id);
  std::vector vlsubs;
  // streams for printing the computation of term in side conditions or
  // list semantics substitutions
  std::stringstream argList;
  std::stringstream argListTerms;
  // the list variables
  std::unordered_set listVars;
  argList << "(";
  // use the names from the user variable list (uvarList)
  for (const Node& v : uvarList)
  {
    std::stringstream sss;
    sss << v;
    Node s = d_tproc.mkInternalSymbol(sss.str(), v.getType());
    odecl << " (! " << sss.str() << " term";
    argList << "(" << sss.str() << " term)";
    argListTerms << " " << sss.str();
    rparen << ")";
    vlsubs.push_back(s);
    // remember if v was a list variable, we must convert these in side
    // condition printing below
    if (expr::isListVar(v))
    {
      listVars.insert(s);
    }
  }
  argList << ")";
  // print conditions
  size_t termCount = 0;
  size_t scCount = 0;
  // print conditions, then conclusion
  for (size_t i = 0, nconds = conds.size(); i <= nconds; i++)
  {
    bool isConclusion = i == nconds;
    Node term = isConclusion ? conc : conds[i];
    Node sterm = term.substitute(
        varList.begin(), varList.end(), vlsubs.begin(), vlsubs.end());
    if (expr::hasListVar(term))
    {
      Assert(!listVars.empty());
      scCount++;
      std::stringstream scName;
      scName << "dsl.sc." << scCount << "." << id;
      // generate the side condition
      oscs << "(function " << scName.str() << " " << argList.str() << " term"
           << std::endl;
      // body must be converted to incorporate list semantics for substitutions
      // first traversal applies nary_elim to required n-ary applications
      LfscListScNodeConverter llsncp(nodeManager(), d_tproc, listVars, true);
      Node tscp;
      if (isConclusion)
      {
        Assert(sterm.getKind() == Kind::EQUAL);
        // optimization: don't need nary_elim for heads
        tscp = llsncp.convert(sterm[1]);
        tscp = sterm[0].eqNode(tscp);
      }
      else
      {
        tscp = llsncp.convert(sterm);
      }
      // second traversal converts to LFSC form
      Node t = d_tproc.convert(tscp);
      // third traversal applies nary_concat where list variables are used
      LfscListScNodeConverter llsnc(nodeManager(), d_tproc, listVars, false);
      Node tsc = llsnc.convert(t);
      oscs << "  ";
      print(oscs, tsc);
      oscs << ")" << std::endl;
      termCount++;
      // introduce a term computed by side condition
      odecl << " (! _t" << termCount << " term";
      rparen << ")";
      format.push_back(Node::null());
      // side condition, which is an implicit argument
      odecl << " (! _s" << scCount << " (^ (" << scName.str();
      rparen << ")";
      // arguments to side condition
      odecl << argListTerms.str() << ") ";
      // matches condition
      odecl << "_t" << termCount << ")";
      if (!isConclusion)
      {
        // the child proof
        odecl << " (! _u" << i;
        rparen << ")";
        format.push_back(term);
      }
      odecl << " (holds _t" << termCount << ")";
      continue;
    }
    // ordinary condition/conclusion, print the term directly
    if (!isConclusion)
    {
      odecl << " (! _u" << i;
      rparen << ")";
      format.push_back(term);
    }
    odecl << " (holds ";
    Node t = d_tproc.convert(sterm);
    print(odecl, t);
    odecl << ")";
  }
  odecl << rparen.str() << ")" << std::endl;
  // print the side conditions
  out << oscs.str();
  // print the rule declaration
  out << odecl.str();
}

}  // namespace proof
}  // namespace cvc5::internal




© 2015 - 2024 Weber Informatics LLC | Privacy Policy