static bool TryAnalyze(RegexNode node, AnalysisResults results, bool isAtomicByAncestor, bool isInLoop)
{
if (!StackHelper.TryEnsureSufficientExecutionStack())
{
return(false);
}
// Track whether we've seen any nodes with various options set.
results._hasIgnoreCase |= (node.Options & RegexOptions.IgnoreCase) != 0;
results._hasRightToLeft |= (node.Options & RegexOptions.RightToLeft) != 0;
// Track whether this node is inside of a loop.
if (isInLoop)
{
(results._inLoops ??= new HashSet <RegexNode>()).Add(node);
}
if (isAtomicByAncestor)
{
// We've been told by our parent that we should be considered atomic, so add ourselves
// to the atomic collection.
results._isAtomicByAncestor.Add(node);
}
else
{
// Certain kinds of nodes incur backtracking logic themselves: add them to the backtracking collection.
// We may later find that a node contains another that has backtracking; we'll add nodes based on that
// after examining the children.
switch (node.Kind)
{
case RegexNodeKind.Alternate:
case RegexNodeKind.Loop or RegexNodeKind.Lazyloop when node.M != node.N:
case RegexNodeKind.Oneloop or RegexNodeKind.Notoneloop or RegexNodeKind.Setloop or RegexNodeKind.Onelazy or RegexNodeKind.Notonelazy or RegexNodeKind.Setlazy when node.M != node.N:
(results._mayBacktrack ??= new HashSet <RegexNode>()).Add(node);
break;
}
}
// Update state for certain node types.
bool isAtomicBySelf = false;
switch (node.Kind)
{
// Some node types add atomicity around what they wrap. Set isAtomicBySelfOrParent to true for such nodes
// even if it was false upon entering the method.
case RegexNodeKind.Atomic:
case RegexNodeKind.NegativeLookaround:
case RegexNodeKind.PositiveLookaround:
isAtomicBySelf = true;
break;
// Track any nodes that are themselves captures.
case RegexNodeKind.Capture:
results._containsCapture.Add(node);
break;
// Track whether we've recurred into a loop
case RegexNodeKind.Loop:
case RegexNodeKind.Lazyloop:
isInLoop = true;
break;
}
// Process each child.
int childCount = node.ChildCount();
for (int i = 0; i < childCount; i++)
{
RegexNode child = node.Child(i);
// Determine whether the child should be treated as atomic (whether anything
// can backtrack into it), which is influenced by whether this node (the child's
// parent) is considered atomic by itself or by its parent.
bool treatChildAsAtomic = (isAtomicByAncestor | isAtomicBySelf) && node.Kind switch
{
// If the parent is atomic, so is the child. That's the whole purpose
// of the Atomic node, and lookarounds are also implicitly atomic.
RegexNodeKind.Atomic or RegexNodeKind.NegativeLookaround or RegexNodeKind.PositiveLookaround => true,
// Each branch is considered independently, so any atomicity applied to the alternation also applies
// to each individual branch. This is true as well for conditionals.
RegexNodeKind.Alternate or RegexNodeKind.BackreferenceConditional or RegexNodeKind.ExpressionConditional => true,
// Captures don't impact atomicity: if the parent of a capture is atomic, the capture is also atomic.
RegexNodeKind.Capture => true,
// If the parent is a concatenation and this is the last node, any atomicity
// applying to the concatenation applies to this node, too.
RegexNodeKind.Concatenate => i == childCount - 1,
// For loops with a max iteration count of 1, they themselves can be considered
// atomic as can whatever they wrap, as they won't ever iterate more than once
// and thus we don't need to worry about one iteration consuming input destined
// for a subsequent iteration.
RegexNodeKind.Loop or RegexNodeKind.Lazyloop when node.N == 1 => true,
// For any other parent type, give up on trying to prove atomicity.
_ => false,
};
// Now analyze the child.
if (!TryAnalyze(child, results, treatChildAsAtomic, isInLoop))
{
return(false);
}
// If the child contains captures, so too does this parent.
if (results._containsCapture.Contains(child))
{
results._containsCapture.Add(node);
}
// If the child might require backtracking into it, so too might the parent,
// unless the parent is itself considered atomic. Here we don't consider parental
// atomicity, as we need to surface upwards to the parent whether any backtracking
// will be visible from this node to it.
if (!isAtomicBySelf && (results._mayBacktrack?.Contains(child) == true))
{
(results._mayBacktrack ??= new HashSet <RegexNode>()).Add(node);
}
}
// Successfully analyzed the node.
return(true);
}
// Processes the node, adding any prefix text to the builder.
// Returns whether processing should continue with subsequent nodes.
static bool Process(RegexNode node, ref ValueStringBuilder vsb)
{
if (!StackHelper.TryEnsureSufficientExecutionStack())
{
// If we're too deep on the stack, just give up finding any more prefix.
return(false);
}
// We don't bother to handle reversed input, so process at most one node
// when handling RightToLeft.
bool rtl = (node.Options & RegexOptions.RightToLeft) != 0;
switch (node.Type)
{
// Concatenation
case RegexNode.Concatenate:
{
int childCount = node.ChildCount();
for (int i = 0; i < childCount; i++)
{
if (!Process(node.Child(i), ref vsb))
{
return(false);
}
}
return(!rtl);
}
// Alternation: find a string that's a shared prefix of all branches
case RegexNode.Alternate:
{
int childCount = node.ChildCount();
// Store the initial branch into the target builder
int initialLength = vsb.Length;
bool keepExploring = Process(node.Child(0), ref vsb);
int addedLength = vsb.Length - initialLength;
// Then explore the rest of the branches, finding the length
// a prefix they all share in common with the initial branch.
if (addedLength != 0)
{
var alternateSb = new ValueStringBuilder(64);
// Process each branch. If we reach a point where we've proven there's
// no overlap, we can bail early.
for (int i = 1; i < childCount && addedLength != 0; i++)
{
alternateSb.Length = 0;
// Process the branch. We want to keep exploring after this alternation,
// but we can't if either this branch doesn't allow for it or if the prefix
// supplied by this branch doesn't entirely match all the previous ones.
keepExploring &= Process(node.Child(i), ref alternateSb);
keepExploring &= alternateSb.Length == addedLength;
addedLength = Math.Min(addedLength, alternateSb.Length);
for (int j = 0; j < addedLength; j++)
{
if (vsb[initialLength + j] != alternateSb[j])
{
addedLength = j;
keepExploring = false;
break;
}
}
}
alternateSb.Dispose();
// Then cull back on what was added based on the other branches.
vsb.Length = initialLength + addedLength;
}
return(!rtl && keepExploring);
}
// One character
case RegexNode.One when(node.Options& RegexOptions.IgnoreCase) == 0:
vsb.Append(node.Ch);
return(!rtl);
// Multiple characters
case RegexNode.Multi when(node.Options& RegexOptions.IgnoreCase) == 0:
vsb.Append(node.Str);
return(!rtl);
// Loop of one character
case RegexNode.Oneloop or RegexNode.Oneloopatomic or RegexNode.Onelazy when node.M > 0 && (node.Options & RegexOptions.IgnoreCase) == 0:
const int SingleCharIterationLimit = 32; // arbitrary cut-off to avoid creating super long strings unnecessarily
int count = Math.Min(node.M, SingleCharIterationLimit);
vsb.Append(node.Ch, count);
return(count == node.N && !rtl);
// Loop of a node
case RegexNode.Loop or RegexNode.Lazyloop when node.M > 0:
{
const int NodeIterationLimit = 4; // arbitrary cut-off to avoid creating super long strings unnecessarily
int limit = Math.Min(node.M, NodeIterationLimit);
for (int i = 0; i < limit; i++)
{
if (!Process(node.Child(0), ref vsb))
{
return(false);
}
}
return(limit == node.N && !rtl);
}
// Grouping nodes for which we only care about their single child
case RegexNode.Atomic:
case RegexNode.Capture:
return(Process(node.Child(0), ref vsb));
// Zero-width anchors and assertions
case RegexNode.Bol:
case RegexNode.Eol:
case RegexNode.Boundary:
case RegexNode.ECMABoundary:
case RegexNode.NonBoundary:
case RegexNode.NonECMABoundary:
case RegexNode.Beginning:
case RegexNode.Start:
case RegexNode.EndZ:
case RegexNode.End:
case RegexNode.Empty:
case RegexNode.UpdateBumpalong:
case RegexNode.Require:
case RegexNode.Prevent:
return(true);
// Give up for anything else
default:
return(false);
}
}