public static void DoSplitBlocksIntoAssignmentsAndPredicates(ControlFlowGraph cfg)
{
cfg.Vertices.ForEach(v =>
{
if (v.BalancedCode.IsEmpty() || v.Residue.IsEmpty()) return;
var ass = (v.BalancedCode.Last() is Assign) ? v.BalancedCode.Last().AssertCast<Assign>() : null;
var lhs = ass == null ? null : ass.Lhs.AssertCast<Ref>().Sym;
var @ref = v.Residue.AssertSingle() is Ref ? v.Residue.AssertSingle().AssertCast<Ref>().Sym : null;
if (lhs != null && @ref != null && lhs.ProtoId == @ref.ProtoId)
{
// todo. this introduces a nasty bug if assigned variable is reused later
v.BalancedCode.RemoveLast();
v.Residue.SetElements(ass.Rhs);
}
if (v.BalancedCode.IsEmpty() || v.Residue.IsEmpty()) return;
var v_test = new ControlFlowBlock();
v_test.Residue.Add(v.Residue.AssertSingle());
v.Residue.RemoveElements();
var outEdges = cfg.Vedges(v, null);
cfg.RemoveEdges(outEdges);
cfg.AddVertex(v_test);
cfg.AddEdge(new ControlFlowEdge(v, v_test));
outEdges.ForEach(e => cfg.AddEdge(new ControlFlowEdge(v_test, e.Target, e.Tag)));
});
}
public static void DoEvictUnreachableCode(ControlFlowGraph cfg)
{
// todo. this still doesn't completely fix code like "while(false)"
var unreachable = cfg.Vertices.Except(cfg.Cflow());
unreachable.Contains(cfg.Finish).AssertFalse();
cfg.RemoveVertices(unreachable);
}
public static void DoRemoveReturnThunk(ControlFlowGraph cfg)
{
var preRets = cfg.Vedges(null, cfg.Finish);
if (preRets.Count() > 1) return;
var wannabe = preRets.AssertSingle().Source;
if (wannabe.BalancedCode.IsEmpty() &&
wannabe.Residue.SingleOrDefault() is Ref)
{
var retThunk = wannabe;
retThunk.BalancedCode.AssertEmpty();
var auxLocal = retThunk.Residue.AssertSingle().AssertCast<Ref>().Sym;
var rets = cfg.Vedges(null, retThunk);
rets.AssertEach(ret => ret.Tag == null);
cfg.RemoveVertex(retThunk);
rets.ForEach(ret =>
{
var src = ret.Source;
src.Residue.AssertEmpty();
var ass = src.BalancedCode.Last().AssertCast<Assign>();
var lhs = ass.Lhs.AssertCast<Ref>();
var rhs = ass.Rhs.AssertCast<Expression>();
(lhs.Sym.ProtoId == auxLocal.ProtoId).AssertTrue();
cfg.AddEdge(new ControlFlowEdge(src, cfg.Finish));
src.BalancedCode.RemoveLast();
src.Residue.Add(rhs);
});
}
}
public DfaHelper(ControlFlowGraph cfg)
{
_cfg = cfg;
// todo. use Cflow instead for being correct with exec order
_vertices = cfg.Vertices.ToList();
var allStmts = _vertices.SelectMany(cfb =>
cfb.BalancedCode.Concat(cfb.Residue.Cast<Node>())).ToReadOnly();
allStmts.ForEach((stmt, i) => PutIntoCache(stmt, i));
}
public static void DoRestoreOpAssignOperators(ControlFlowGraph cfg, Symbols symbols)
{
var dirty = true;
var dfa = new DfaHelper(cfg);
while (dirty)
{
dirty = false;
foreach (var atom in dfa.Atoms())
{
if (dirty |= TryMatchPattern1(dfa, atom)) break;
if (dirty |= TryMatchPattern2(dfa, atom)) break;
if (dirty |= TryMatchPattern3(dfa, atom)) break;
if (dirty |= TryMatchPattern4(dfa, atom)) break;
}
}
}
public static void DoNormalizeEdgeTags(ControlFlowGraph cfg)
{
cfg.Vertices.AssertEach(v => cfg.Vedges(v, null).Count() <= 2);
var binaryCondEdges = cfg.Edges().Where(e => e.Tag.Arity() == 2);
var sourceVertices = binaryCondEdges.Select(e => e.Source).Distinct();
sourceVertices.ForEach(v =>
{
var outEdges = cfg.Vedges(v, null);
outEdges.AssertEach(e => e.Tag.Arity() == 2);
var operatorType = outEdges.First().Tag.Value.ToOperatorType();
var newTags = outEdges.Select(e =>
e.Tag == operatorType.ToPredicateType() ? PredicateType.IsTrue :
e.Tag == operatorType.ToPredicateType().Negate() ? PredicateType.IsFalse :
((Func<PredicateType>)(() => { throw AssertionHelper.Fail(); }))());
var joint = Operator.Create(operatorType, v.Residue);
v.Residue.SetElements(joint);
outEdges.Zip(newTags).ForEach(e => e.Item1.Tag = e.Item2);
});
cfg.Edges().AssertEach(e => e.Tag.Arity() <= 1);
cfg.Vertices.Where(v => v.Residue.IsNotEmpty()).AssertEach(v => v.Residue.Count() == 1);
}
public static void DoStripOffRedundancies(ControlFlowGraph cfg)
{
var allNodes = cfg.Vertices.SelectMany(cfb => Seq.Concat(cfb.BalancedCode, cfb.Residue.Cast<Node>()));
allNodes.ForEach(StripOffRedundanciesInPlace);
}
public static void DoRestoreCollectionInitializers(ControlFlowGraph cfg)
{
// here we transform "new T[0]" expressions into collection initializers
var allStmts = cfg.Vertices.SelectMany(cfb => cfb.BalancedCode).ToReadOnly();
var allNodes = allStmts.SelectMany(s => s.Family()).ToReadOnly();
var emptyArrayCtors = allNodes.OfType<Eval>().Where(eval =>
{
var ctor = eval.InvokedCtor();
if (ctor == null || !ctor.DeclaringType.IsArray) return false;
var app = eval == null ? null : eval.Callee;
var arrlen = app == null ? null : app.Args.SingleOrDefault2() as Const;
if (arrlen == null || (!(arrlen.Value is int) && !(arrlen.Value is long))) return false;
var i_arrlen = arrlen.Value.AssertCoerce<long>();
return i_arrlen == 0;
}).ToReadOnly();
emptyArrayCtors.ForEach(eac => eac.Parent.ReplaceRecursive(eac, new CollectionInit(eac)));
// now we proceed to decompile stuff like "var a = new T[n]; a[0] = foo0; ... a[n-1] = foo;"
var arrayCtors = allStmts.OfType<Assign>().Where(ass =>
{
if (!(ass.Rhs is Eval)) return false;
var ctor = ass.Rhs.InvokedCtor();
return ctor == null ? false : ctor.DeclaringType.IsArray;
}).ToReadOnly();
// todo #1. this needs to be like RestoreOpAssignOperators
// to be capable of restoring recursive collection initializers
// todo #2. what about multidimensional initializers for jagges and/or rects?
// for the latter t0do n0te that for dimensions higher than 3 arrays of objects define
// GetValue(params) and SetValue(params) methods => this won't work correctly
foreach (var ctor in arrayCtors)
{
var cfb = cfg.Vertices.Single(cfb1 => cfb1.BalancedCode.Contains(ctor));
var map = new Dictionary<long, Expression>();
var inits = cfb.BalancedCode.SkipWhile(n => n != ctor).Skip(1).TakeWhile(n =>
{
var m = n.InvokedMethod();
if (m != null && ((m.DeclaringType == typeof(Array) && m.Name == "SetValue") ||
(m.DeclaringType.IsArray && m.Name == "Set")))
{
var args = n.InvocationArgs().AssertNotNull();
if (args.Count() != 3) return false;
var arrayRef = args.First();
if (!arrayRef.Equiv(ctor.Lhs)) return false;
var value = args.Third() as Expression;
if (value == null) return false;
var index = args.Second() as Const;
if (index == null || (!(index.Value is int) && !(index.Value is long))) return false;
var i_index = index.Value.AssertCoerce<long>();
map.Add(i_index, value);
return true;
}
else
{
return false;
}
}).ToReadOnly();
var arrlen = ctor.Rhs.InvocationArgs().SingleOrDefault2() as Const;
if (arrlen == null || (!(arrlen.Value is int) && !(arrlen.Value is long))) continue;
var i_arrlen = arrlen.Value.AssertCoerce<long>();
if (map.Count() != i_arrlen) continue;
if (!Set.Equal(0L.Unfold(i => i + 1, i => i < i_arrlen), map.Keys)) continue;
var elements = map.OrderBy(kvp => kvp.Key).Select(kvp => kvp.Value).ToReadOnly();
var arrayInit = new CollectionInit(ctor.Rhs.AssertCast<Eval>(), elements);
inits.ForEach(init => cfg.Remove(init));
ctor.Rhs = arrayInit;
}
}
public static Block DoDecompileScopes(ControlFlowGraph cfg)
{
// todo. do not crash on irregular control flow but rather emit labels/gotos
return cfg.DecompileScopes();
}
public static void DoRestoreObjectInitializers(ControlFlowGraph cfg)
{
// todo. to be implemented
}
public static void DoRestoreConditionalOperators(ControlFlowGraph cfg)
{
// todo. to be implemented
}
public static void DoDecompileComplexConditions(ControlFlowGraph cfg)
{
var flow = cfg.Cflow(cfg.Start);
while (true)
{
var vs = flow.FirstOrDefault(v =>
v.Residue.Count() == 1 &&
cfg.Vedges(v, null).Count() == 2 &&
cfg.Vedges(v, null).All(e => e.Target.BalancedCode.IsEmpty() && e.Target.Residue.Count() == 1));
if (vs == null) break;
var conv = cfg.ConvStrict(vs);
var ass = conv.BalancedCode.AssertFirst().AssertCast<Assign>();
(ass.Lhs is Ref && ass.Rhs is Loophole).AssertTrue();
var parts = cfg.Cflow(vs, conv);
var innerEdges = cfg.Edges(parts, parts).ToList();
cfg.Edges(null, parts).Except(innerEdges).AssertEach(e => e.Target == vs);
cfg.Edges(parts, null).Except(innerEdges).AssertEach(e => e.Source == vs || e.Source == conv);
while (true)
{
var somethingWasChanged = false;
foreach (var pivot in parts)
{
var pivot_inEdges = cfg.Vedges(null, pivot);
if (pivot_inEdges.Count() != 1) continue;
var e_pred2pivot = pivot_inEdges.AssertSingle();
var pred = e_pred2pivot.Source;
var pred_outEdges = cfg.Vedges(pred, null);
if (pred_outEdges.Count() != 2) continue;
var e_pred2target1 = pred_outEdges.AssertSingle(e => e.Target != pivot);
var target1 = e_pred2target1.Target;
var e_pivot2target1 = cfg.Vedge(pivot, target1);
if (e_pivot2target1 == null) continue;
var pivot_outEdges = cfg.Vedges(pivot, null);
if (pivot_outEdges.Count() != 2) continue;
var e_pivot2target2 = pivot_outEdges.AssertSingle(e => e.Target != target1);
var target2 = e_pivot2target2.Target;
var @operator = e_pred2target1.Condition == PredicateType.IsTrue ? OperatorType.OrElse :
e_pred2target1.Condition == PredicateType.IsFalse ? OperatorType.AndAlso :
((Func<OperatorType>)(() => { throw AssertionHelper.Fail(); }))();
var clause_left = pred.Residue.AssertSingle();
var clause_right = pivot.Residue.AssertSingle();
var negate_rhs = e_pred2target1.Condition != e_pivot2target1.Condition;
if (negate_rhs) clause_right = Operator.Not(clause_right);
var junction = Operator.Create(@operator, clause_left, clause_right);
cfg.RemoveVertex(pivot);
cfg.AddEdge(new ControlFlowEdge(pred, target2, e_pred2target1.Condition.Negate()));
pred.Residue.SetElements(junction);
somethingWasChanged |= true;
}
if (!somethingWasChanged) break;
}
parts = cfg.Cflow(vs, conv);
(parts.Count() == 4).AssertTrue();
var @const = parts.Except(vs, conv).AssertSingle(v => v.Residue.AssertSingle() is Const);
var vnext = parts.Except(vs, conv, @const).AssertSingle();
(cfg.Vedge(@const, vnext) == null && cfg.Vedge(vnext, @const) == null).AssertTrue();
cfg.Vedge(vs, vnext).IsConditional.AssertTrue();
cfg.Vedge(vs, @const).IsConditional.AssertTrue();
cfg.Vedge(vnext, conv).IsUnconditional.AssertTrue();
cfg.Vedge(@const, conv).IsUnconditional.AssertTrue();
var estart = vs.Residue.AssertSingle();
var enext = vnext.Residue.AssertSingle();
var cond_const = @const.Residue.AssertSingle().AssertCast<Const>().Value.AssertCast<int>();
var cond_edge = cfg.Vedge(vs, @const).Condition;
var val_const = cond_const == 1 ? true :
cond_const == 0 ? false :
((Func<bool>)(() => { throw AssertionHelper.Fail(); }))();
var val_edge = cond_edge == PredicateType.IsTrue ? true :
cond_edge == PredicateType.IsFalse ? false :
((Func<bool>)(() => { throw AssertionHelper.Fail(); }))();
var operator1 = val_const ? OperatorType.OrElse : OperatorType.AndAlso;
var clause_left1 = val_edge && val_const ? estart : Operator.Not(estart);
var clause_right1 = !val_edge && !val_const ? Operator.Not(enext) : enext;
var junction1 = Operator.Create(operator1, clause_left1, clause_right1);
var conv_outEdges = cfg.Vedges(conv, null);
var conv_inEdges = cfg.Vedges(null, conv).Except(cfg.Vedges(parts, conv));
cfg.RemoveVertices(@const, vnext, conv);
conv_outEdges.ForEach(e => cfg.AddEdge(new ControlFlowEdge(vs, e.Target, e.Tag)));
conv_inEdges.ForEach(e => cfg.AddEdge(new ControlFlowEdge(e.Source, vs, e.Tag)));
vs.BalancedCode.Add(new Assign(ass.Lhs, junction1));
vs.BalancedCode.AddElements(conv.BalancedCode.Skip(1));
vs.Residue.SetElements(conv.Residue);
}
cfg.Edges().AssertEach(e => e.Tag.Arity() <= 1);
cfg.Vertices.Where(v => v.Residue.IsNotEmpty()).AssertEach(v => v.Residue.Count() == 1);
cfg.Vertices.AssertNone(v => v.Residue.IsNotEmpty() &&
cfg.Vedges(v, null).Any(e => e.IsUnconditional && e.Target != cfg.Finish));
}
public static ControlFlowGraph DoCreateCarcass(IMethodBody cil, out ReadOnlyDictionary<ControlFlowBlock, ReadOnlyCollection<IILOp>> blocks2parts)
{
// create the control flow graph
var cfg = new ControlFlowGraph();
// partition the code into blocks with continuous control flow
// todo. support switches and protected regions
var targets = new HashSet<IILOp>(cil.OfType<Branch>().Select(br => br.Target));
var l_partitions = new List<ReadOnlyCollection<IILOp>>();
var l_partition = new List<IILOp>();
Action qualifyPartition = () => { if (l_partition.IsNotEmpty()) { l_partitions.Add(l_partition.ToReadOnly()); l_partition = new List<IILOp>(); } };
foreach (var op in cil)
{
if (op is Branch || op is Ret) qualifyPartition();
else
{
if (targets.Contains(op)) qualifyPartition();
l_partition.Add(op);
if (op is Throw) qualifyPartition();
}
}
qualifyPartition();
var partitions = l_partitions.ToReadOnly();
// create blocks and map those to ops and partitions
blocks2parts = partitions.ToDictionary(p => new ControlFlowBlock(), p => p).ToReadOnly();
blocks2parts.ForEach(kvp => cfg.AddVertex(kvp.Key));
var op2blocks = new Dictionary<IILOp, ControlFlowBlock>();
blocks2parts.ForEach(kvp => kvp.Value.ForEach(op => op2blocks.Add(op, kvp.Key)));
cil.ForEach(op => { if (!op2blocks.ContainsKey(op)) op2blocks.Add(op, null); });
// prepare to link the blocks
Action<IILOp, IILOp, CilPredicateType?> link = (op1, op2, cil_pred) =>
{
var source = op1 == null ? cfg.Start : op2blocks[op1];
var target = op2 == null ? cfg.Finish : op2blocks[op2];
var hir_pred = cil_pred == null ? (HirPredicateType?)null :
(HirPredicateType)Enum.Parse(typeof(HirPredicateType), cil_pred.Value.ToString());
cfg.AddEdge(new ControlFlowEdge(source, target, hir_pred));
};
// link the blocks (down from 300+ LOC to this simple loop =))
if (cil.IsEmpty()) link(null, null, null);
foreach (var op in cil)
{
// todo. support switches here
if (op is Switch) throw AssertionHelper.Fail();
// todo. support general case of control flow
// n0te. throw needs something on stack, so br > throw is impossible
Func<IILOp, bool> isJmp = op1 => op1 is Ret || op1 is Branch;
if (isJmp(op) && isJmp(op.Prev)) continue;
if (isJmp(op))
{
Func<IILOp, CilPredicateType?> pred = op1 => op1 is Ret ? null : op1 is Branch ? ((Branch)op1).PredicateType : ((Func<CilPredicateType?>)(() => { throw AssertionHelper.Fail(); }))();
Func<IILOp, bool> uncond = op1 => isJmp(op1) && pred(op1) == null;
Func<IILOp, bool> cond = op1 => isJmp(op1) && pred(op1) != null;
Func<IILOp, IILOp> target = null; target = op1 =>
op1 is Ret ? null : op1 is Branch ? target(((Branch)op1).Target) : op1;
(target(op) is Branch).AssertFalse();
if (target(op) is Ret) link(op.Prev, null, pred(op));
else link(op.Prev, target(op), pred(op));
isJmp(op.Next).AssertImplies(uncond(op.Next));
if (cond(op)) link(op.Prev, target(op.Next), pred(op).Negate());
}
else if (op is Throw)
{
// do nothing - throw doesn't create links
}
else
{
if (op.Prev == null) link(null, op, null);
if (isJmp(op.Next)) continue;
var blk = op2blocks.GetOrDefault(op);
var blk_next = op2blocks.GetOrDefault(op.Next);
if (blk != blk_next) link(op, op.Next, null);
}
}
// yield control to the next step of the pipeline
return cfg;
}