feat: port session 07 — Protocol Parser, Auth extras (TPM/certidp/certstore), Internal utilities & data structures

Session 07 scope (5 features, 17 tests, ~1165 Go LOC):
- Protocol/ParserTypes.cs: ParserState enum (79 states), PublishArgument, ParseContext
- Protocol/IProtocolHandler.cs: handler interface decoupling parser from client
- Protocol/ProtocolParser.cs: Parse(), ProtoSnippet(), OverMaxControlLineLimit(),
  ProcessPub/HeaderPub/RoutedMsgArgs/RoutedHeaderMsgArgs, ClonePubArg(), GetHeader()
- tests/Protocol/ProtocolParserTests.cs: 17 tests via TestProtocolHandler stub

Auth extras from session 06 (committed separately):
- Auth/TpmKeyProvider.cs, Auth/CertificateIdentityProvider/, Auth/CertificateStore/

Internal utilities & data structures (session 06 overflow):
- Internal/AccessTimeService.cs, ElasticPointer.cs, SystemMemory.cs, ProcessStatsProvider.cs
- Internal/DataStructures/GenericSublist.cs, HashWheel.cs
- Internal/DataStructures/SubjectTree.cs, SubjectTreeNode.cs, SubjectTreeParts.cs

All 461 tests pass (460 unit + 1 integration). DB updated for features 2588-2592 and tests 2598-2614.
This commit is contained in:
Joseph Doherty
2026-02-26 13:16:56 -05:00
parent 0a54d342ba
commit 88b1391ef0
56 changed files with 9006 additions and 6 deletions

View File

@@ -0,0 +1,678 @@
// Copyright 2025 The NATS Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
namespace ZB.MOM.NatsNet.Server.Internal.DataStructures;
// Sublist is a routing mechanism to handle subject distribution and
// provides a facility to match subjects from published messages to
// interested subscribers. Subscribers can have wildcard subjects to
// match multiple published subjects.
/// <summary>
/// A value type used with <see cref="SimpleSublist"/> to track interest without
/// storing any associated data. Equivalent to Go's <c>struct{}</c>.
/// </summary>
public readonly struct EmptyStruct : IEquatable<EmptyStruct>
{
public static readonly EmptyStruct Value = default;
public bool Equals(EmptyStruct other) => true;
public override bool Equals(object? obj) => obj is EmptyStruct;
public override int GetHashCode() => 0;
public static bool operator ==(EmptyStruct left, EmptyStruct right) => true;
public static bool operator !=(EmptyStruct left, EmptyStruct right) => false;
}
/// <summary>
/// A thread-safe trie-based NATS subject routing list that efficiently stores and
/// retrieves subscriptions. Wildcards <c>*</c> (single-token) and <c>&gt;</c>
/// (full-wildcard) are supported.
/// </summary>
/// <typeparam name="T">The subscription value type. Must be non-null.</typeparam>
public class GenericSublist<T> where T : notnull
{
// Token separator and wildcard constants (mirrors Go's const block).
private const char Pwc = '*';
private const char Fwc = '>';
private const char Btsep = '.';
// -------------------------------------------------------------------------
// Public error singletons (mirrors Go's var block).
// -------------------------------------------------------------------------
/// <summary>Thrown when a subject is syntactically invalid.</summary>
public static readonly ArgumentException ErrInvalidSubject =
new("gsl: invalid subject");
/// <summary>Thrown when a subscription is not found during removal.</summary>
public static readonly KeyNotFoundException ErrNotFound =
new("gsl: no matches found");
/// <summary>Thrown when a value is already registered for the given subject.</summary>
public static readonly InvalidOperationException ErrAlreadyRegistered =
new("gsl: notification already registered");
// -------------------------------------------------------------------------
// Fields
// -------------------------------------------------------------------------
private readonly TrieLevel _root;
private uint _count;
private readonly ReaderWriterLockSlim _lock = new(LockRecursionPolicy.NoRecursion);
// -------------------------------------------------------------------------
// Construction
// -------------------------------------------------------------------------
internal GenericSublist()
{
_root = new TrieLevel();
}
/// <summary>Creates a new <see cref="GenericSublist{T}"/>.</summary>
public static GenericSublist<T> NewSublist() => new();
/// <summary>Creates a new <see cref="SimpleSublist"/>.</summary>
public static SimpleSublist NewSimpleSublist() => new();
// -------------------------------------------------------------------------
// Public API
// -------------------------------------------------------------------------
/// <summary>Returns the total number of subscriptions stored.</summary>
public uint Count
{
get
{
_lock.EnterReadLock();
try { return _count; }
finally { _lock.ExitReadLock(); }
}
}
/// <summary>
/// Inserts a subscription into the trie.
/// Throws <see cref="ArgumentException"/> if <paramref name="subject"/> is invalid.
/// </summary>
public void Insert(string subject, T value)
{
_lock.EnterWriteLock();
try
{
InsertCore(subject, value);
}
finally
{
_lock.ExitWriteLock();
}
}
/// <summary>
/// Removes a subscription from the trie.
/// Throws <see cref="ArgumentException"/> if the subject is invalid, or
/// <see cref="KeyNotFoundException"/> if not found.
/// </summary>
public void Remove(string subject, T value)
{
_lock.EnterWriteLock();
try
{
RemoveCore(subject, value);
}
finally
{
_lock.ExitWriteLock();
}
}
/// <summary>
/// Calls <paramref name="action"/> for every value whose subscription matches
/// the literal <paramref name="subject"/>.
/// </summary>
public void Match(string subject, Action<T> action)
{
_lock.EnterReadLock();
try
{
var tokens = TokenizeForMatch(subject);
if (tokens == null) return;
MatchLevel(_root, tokens, 0, action);
}
finally
{
_lock.ExitReadLock();
}
}
/// <summary>
/// Calls <paramref name="action"/> for every value whose subscription matches
/// <paramref name="subject"/> supplied as a UTF-8 byte span.
/// </summary>
public void MatchBytes(ReadOnlySpan<byte> subject, Action<T> action)
{
Match(System.Text.Encoding.UTF8.GetString(subject), action);
}
/// <summary>
/// Returns <see langword="true"/> when at least one subscription matches
/// <paramref name="subject"/>.
/// </summary>
public bool HasInterest(string subject)
{
_lock.EnterReadLock();
try
{
var tokens = TokenizeForMatch(subject);
if (tokens == null) return false;
int dummy = 0;
return MatchLevelForAny(_root, tokens, 0, ref dummy);
}
finally
{
_lock.ExitReadLock();
}
}
/// <summary>
/// Returns the number of subscriptions that match <paramref name="subject"/>.
/// </summary>
public int NumInterest(string subject)
{
_lock.EnterReadLock();
try
{
var tokens = TokenizeForMatch(subject);
if (tokens == null) return 0;
int np = 0;
MatchLevelForAny(_root, tokens, 0, ref np);
return np;
}
finally
{
_lock.ExitReadLock();
}
}
/// <summary>
/// Returns <see langword="true"/> if the trie contains any subscription that
/// could match a subject whose tokens begin with the tokens of
/// <paramref name="subject"/>. Used for trie intersection checks.
/// </summary>
public bool HasInterestStartingIn(string subject)
{
_lock.EnterReadLock();
try
{
var tokens = TokenizeSubjectIntoSlice(subject);
return HasInterestStartingInLevel(_root, tokens, 0);
}
finally
{
_lock.ExitReadLock();
}
}
// -------------------------------------------------------------------------
// Internal helpers (accessible to tests in the same assembly).
// -------------------------------------------------------------------------
/// <summary>Returns the maximum depth of the trie. Used in tests.</summary>
internal int NumLevels() => VisitLevel(_root, 0);
// -------------------------------------------------------------------------
// Private: Insert core (lock must be held by caller)
// -------------------------------------------------------------------------
private void InsertCore(string subject, T value)
{
var sfwc = false; // seen full-wildcard token
TrieNode? n = null;
var l = _root;
// Iterate tokens split by '.' using index arithmetic to avoid allocations.
var start = 0;
while (start <= subject.Length)
{
// Find end of this token.
var end = subject.IndexOf(Btsep, start);
var isLast = end < 0;
if (isLast) end = subject.Length;
var tokenLen = end - start;
if (tokenLen == 0 || sfwc)
throw new ArgumentException(ErrInvalidSubject.Message);
if (tokenLen > 1)
{
var t = subject.Substring(start, tokenLen);
if (!l.Nodes.TryGetValue(t, out n))
{
n = new TrieNode();
l.Nodes[t] = n;
}
}
else
{
switch (subject[start])
{
case Pwc:
if (l.PwcNode == null) l.PwcNode = new TrieNode();
n = l.PwcNode;
break;
case Fwc:
if (l.FwcNode == null) l.FwcNode = new TrieNode();
n = l.FwcNode;
sfwc = true;
break;
default:
var t = subject.Substring(start, 1);
if (!l.Nodes.TryGetValue(t, out n))
{
n = new TrieNode();
l.Nodes[t] = n;
}
break;
}
}
n.Next ??= new TrieLevel();
l = n.Next;
if (isLast) break;
start = end + 1;
}
if (n == null)
throw new ArgumentException(ErrInvalidSubject.Message);
n.Subs[value] = subject;
_count++;
}
// -------------------------------------------------------------------------
// Private: Remove core (lock must be held by caller)
// -------------------------------------------------------------------------
private void RemoveCore(string subject, T value)
{
var sfwc = false;
var l = _root;
// We use a fixed-size stack-style array to track visited (level, node, token)
// triples so we can prune upward after removal. 32 is the same as Go's [32]lnt.
var levels = new LevelNodeToken[32];
var levelCount = 0;
TrieNode? n = null;
var start = 0;
while (start <= subject.Length)
{
var end = subject.IndexOf(Btsep, start);
var isLast = end < 0;
if (isLast) end = subject.Length;
var tokenLen = end - start;
if (tokenLen == 0 || sfwc)
throw new ArgumentException(ErrInvalidSubject.Message);
if (l == null!)
throw new KeyNotFoundException(ErrNotFound.Message);
var tokenStr = subject.Substring(start, tokenLen);
if (tokenLen > 1)
{
l.Nodes.TryGetValue(tokenStr, out n);
}
else
{
switch (tokenStr[0])
{
case Pwc:
n = l.PwcNode;
break;
case Fwc:
n = l.FwcNode;
sfwc = true;
break;
default:
l.Nodes.TryGetValue(tokenStr, out n);
break;
}
}
if (n != null)
{
if (levelCount < levels.Length)
levels[levelCount++] = new LevelNodeToken(l, n, tokenStr);
l = n.Next!;
}
else
{
l = null!;
}
if (isLast) break;
start = end + 1;
}
// Remove from the final node's subscription map.
if (!RemoveFromNode(n, value))
throw new KeyNotFoundException(ErrNotFound.Message);
_count--;
// Prune empty nodes upward.
for (var i = levelCount - 1; i >= 0; i--)
{
var (lv, nd, tk) = levels[i];
if (nd.IsEmpty())
lv.PruneNode(nd, tk);
}
}
private static bool RemoveFromNode(TrieNode? n, T value)
{
if (n == null) return false;
return n.Subs.Remove(value);
}
// -------------------------------------------------------------------------
// Private: matchLevel - recursive trie descent with callback
// Mirrors Go's matchLevel function exactly.
// -------------------------------------------------------------------------
private static void MatchLevel(TrieLevel? l, string[] tokens, int start, Action<T> action)
{
TrieNode? pwc = null;
TrieNode? n = null;
for (var i = start; i < tokens.Length; i++)
{
if (l == null) return;
// Full-wildcard at this level matches everything at/below.
if (l.FwcNode != null)
CallbacksForResults(l.FwcNode, action);
pwc = l.PwcNode;
if (pwc != null)
MatchLevel(pwc.Next, tokens, i + 1, action);
l.Nodes.TryGetValue(tokens[i], out n);
l = n?.Next;
}
// After consuming all tokens, emit subs from exact and pwc matches.
if (n != null)
CallbacksForResults(n, action);
if (pwc != null)
CallbacksForResults(pwc, action);
}
private static void CallbacksForResults(TrieNode n, Action<T> action)
{
foreach (var sub in n.Subs.Keys)
action(sub);
}
// -------------------------------------------------------------------------
// Private: matchLevelForAny - returns true on first match, counting via np
// Mirrors Go's matchLevelForAny function exactly.
// -------------------------------------------------------------------------
private static bool MatchLevelForAny(TrieLevel? l, string[] tokens, int start, ref int np)
{
TrieNode? pwc = null;
TrieNode? n = null;
for (var i = start; i < tokens.Length; i++)
{
if (l == null) return false;
if (l.FwcNode != null)
{
np += l.FwcNode.Subs.Count;
return true;
}
pwc = l.PwcNode;
if (pwc != null)
{
if (MatchLevelForAny(pwc.Next, tokens, i + 1, ref np))
return true;
}
l.Nodes.TryGetValue(tokens[i], out n);
l = n?.Next;
}
if (n != null)
{
np += n.Subs.Count;
if (n.Subs.Count > 0) return true;
}
if (pwc != null)
{
np += pwc.Subs.Count;
return pwc.Subs.Count > 0;
}
return false;
}
// -------------------------------------------------------------------------
// Private: hasInterestStartingIn - mirrors Go's hasInterestStartingIn
// -------------------------------------------------------------------------
private static bool HasInterestStartingInLevel(TrieLevel? l, string[] tokens, int start)
{
if (l == null) return false;
if (start >= tokens.Length) return true;
if (l.FwcNode != null) return true;
var found = false;
if (l.PwcNode != null)
found = HasInterestStartingInLevel(l.PwcNode.Next, tokens, start + 1);
if (!found && l.Nodes.TryGetValue(tokens[start], out var n))
found = HasInterestStartingInLevel(n.Next, tokens, start + 1);
return found;
}
// -------------------------------------------------------------------------
// Private: numLevels helper - mirrors Go's visitLevel
// -------------------------------------------------------------------------
private static int VisitLevel(TrieLevel? l, int depth)
{
if (l == null || l.NumNodes() == 0) return depth;
depth++;
var maxDepth = depth;
foreach (var n in l.Nodes.Values)
{
var d = VisitLevel(n.Next, depth);
if (d > maxDepth) maxDepth = d;
}
if (l.PwcNode != null)
{
var d = VisitLevel(l.PwcNode.Next, depth);
if (d > maxDepth) maxDepth = d;
}
if (l.FwcNode != null)
{
var d = VisitLevel(l.FwcNode.Next, depth);
if (d > maxDepth) maxDepth = d;
}
return maxDepth;
}
// -------------------------------------------------------------------------
// Private: tokenization helpers
// -------------------------------------------------------------------------
/// <summary>
/// Tokenizes a subject for match/hasInterest operations.
/// Returns <see langword="null"/> if the subject contains an empty token,
/// because an empty token can never match any subscription in the trie.
/// Mirrors Go's inline tokenization in <c>match()</c> and <c>hasInterest()</c>.
/// </summary>
private static string[]? TokenizeForMatch(string subject)
{
if (subject.Length == 0) return null;
var tokens = new List<string>(8);
var start = 0;
for (var i = 0; i < subject.Length; i++)
{
if (subject[i] == Btsep)
{
if (i - start == 0) return null; // empty token
tokens.Add(subject.Substring(start, i - start));
start = i + 1;
}
}
// Trailing separator produces empty last token.
if (start >= subject.Length) return null;
tokens.Add(subject.Substring(start));
return tokens.ToArray();
}
/// <summary>
/// Tokenizes a subject into a string array without validation.
/// Mirrors Go's <c>tokenizeSubjectIntoSlice</c>.
/// </summary>
private static string[] TokenizeSubjectIntoSlice(string subject)
{
var tokens = new List<string>(8);
var start = 0;
for (var i = 0; i < subject.Length; i++)
{
if (subject[i] == Btsep)
{
tokens.Add(subject.Substring(start, i - start));
start = i + 1;
}
}
tokens.Add(subject.Substring(start));
return tokens.ToArray();
}
// -------------------------------------------------------------------------
// Private: Trie node and level types
// -------------------------------------------------------------------------
/// <summary>
/// A trie node holding a subscription map and an optional link to the next level.
/// Mirrors Go's <c>node[T]</c>.
/// </summary>
private sealed class TrieNode
{
/// <summary>Maps subscription value → original subject string.</summary>
public readonly Dictionary<T, string> Subs = new();
/// <summary>The next trie level below this node, or null if at a leaf.</summary>
public TrieLevel? Next;
/// <summary>
/// Returns true when the node has no subscriptions and no live children.
/// Used during removal to decide whether to prune this node.
/// Mirrors Go's <c>node.isEmpty()</c>.
/// </summary>
public bool IsEmpty() => Subs.Count == 0 && (Next == null || Next.NumNodes() == 0);
}
/// <summary>
/// A trie level containing named child nodes and special wildcard slots.
/// Mirrors Go's <c>level[T]</c>.
/// </summary>
private sealed class TrieLevel
{
public readonly Dictionary<string, TrieNode> Nodes = new();
public TrieNode? PwcNode; // '*' single-token wildcard node
public TrieNode? FwcNode; // '>' full-wildcard node
/// <summary>
/// Returns the total count of live nodes at this level.
/// Mirrors Go's <c>level.numNodes()</c>.
/// </summary>
public int NumNodes()
{
var num = Nodes.Count;
if (PwcNode != null) num++;
if (FwcNode != null) num++;
return num;
}
/// <summary>
/// Removes an empty node from this level, using reference equality to
/// distinguish wildcard slots from named slots.
/// Mirrors Go's <c>level.pruneNode()</c>.
/// </summary>
public void PruneNode(TrieNode n, string token)
{
if (ReferenceEquals(n, FwcNode))
FwcNode = null;
else if (ReferenceEquals(n, PwcNode))
PwcNode = null;
else
Nodes.Remove(token);
}
}
/// <summary>
/// Tracks a (level, node, token) triple during removal for upward pruning.
/// Mirrors Go's <c>lnt[T]</c>.
/// </summary>
private readonly struct LevelNodeToken
{
public readonly TrieLevel Level;
public readonly TrieNode Node;
public readonly string Token;
public LevelNodeToken(TrieLevel level, TrieNode node, string token)
{
Level = level;
Node = node;
Token = token;
}
public void Deconstruct(out TrieLevel level, out TrieNode node, out string token)
{
level = Level;
node = Node;
token = Token;
}
}
}
/// <summary>
/// A lightweight sublist that tracks interest only, without storing any associated data.
/// Equivalent to Go's <c>SimpleSublist = GenericSublist[struct{}]</c>.
/// </summary>
public sealed class SimpleSublist : GenericSublist<EmptyStruct>
{
internal SimpleSublist() { }
}

View File

@@ -0,0 +1,263 @@
using System.Buffers.Binary;
namespace ZB.MOM.NatsNet.Server.Internal.DataStructures;
/// <summary>
/// A time-based hash wheel for efficiently scheduling and expiring timer tasks keyed by sequence number.
/// Each slot covers a 1-second window; the wheel has 4096 slots (covering ~68 minutes before wrapping).
/// Not thread-safe.
/// </summary>
/// <remarks>
/// Mirrors the Go <c>thw.HashWheel</c> type. Timestamps are Unix nanoseconds (<see cref="long"/>).
/// </remarks>
public sealed class HashWheel
{
/// <summary>Slot width in nanoseconds (1 second).</summary>
private const long TickDuration = 1_000_000_000L;
private const int WheelBits = 12;
private const int WheelSize = 1 << WheelBits; // 4096
private const int WheelMask = WheelSize - 1;
private const int HeaderLen = 17; // 1 magic + 8 count + 8 highSeq
public static readonly Exception ErrTaskNotFound = new InvalidOperationException("thw: task not found");
public static readonly Exception ErrInvalidVersion = new InvalidDataException("thw: encoded version not known");
private readonly Slot?[] _wheel = new Slot?[WheelSize];
private long _lowest = long.MaxValue;
private ulong _count;
// --- Slot ---
private sealed class Slot
{
public readonly Dictionary<ulong, long> Entries = new();
public long Lowest = long.MaxValue;
}
/// <summary>Creates a new empty <see cref="HashWheel"/>.</summary>
public static HashWheel NewHashWheel() => new();
private static Slot NewSlot() => new();
private long GetPosition(long expires) => (expires / TickDuration) & WheelMask;
// --- Public API ---
/// <summary>Returns the number of tasks currently scheduled.</summary>
public ulong Count => _count;
/// <summary>Schedules a new timer task.</summary>
public void Add(ulong seq, long expires)
{
var pos = (int)GetPosition(expires);
_wheel[pos] ??= NewSlot();
var slot = _wheel[pos]!;
if (!slot.Entries.ContainsKey(seq))
_count++;
slot.Entries[seq] = expires;
if (expires < slot.Lowest)
{
slot.Lowest = expires;
if (expires < _lowest)
_lowest = expires;
}
}
/// <summary>Removes a timer task.</summary>
/// <exception cref="InvalidOperationException">Thrown (as <see cref="ErrTaskNotFound"/>) when not found.</exception>
public void Remove(ulong seq, long expires)
{
var pos = (int)GetPosition(expires);
var slot = _wheel[pos];
if (slot is null || !slot.Entries.ContainsKey(seq))
throw ErrTaskNotFound;
slot.Entries.Remove(seq);
_count--;
if (slot.Entries.Count == 0)
_wheel[pos] = null;
}
/// <summary>Updates the expiration time of an existing timer task.</summary>
public void Update(ulong seq, long oldExpires, long newExpires)
{
Remove(seq, oldExpires);
Add(seq, newExpires);
}
/// <summary>
/// Expires all tasks whose timestamp is &lt;= now. The callback receives each task;
/// if it returns <see langword="true"/> the task is removed, otherwise it is kept.
/// </summary>
public void ExpireTasks(Func<ulong, long, bool> callback)
{
var now = DateTimeOffset.UtcNow.ToUnixTimeMilliseconds() * 1_000_000L;
ExpireTasksInternal(now, callback);
}
internal void ExpireTasksInternal(long ts, Func<ulong, long, bool> callback)
{
if (_lowest > ts)
return;
var globalLowest = long.MaxValue;
for (var pos = 0; pos < WheelSize; pos++)
{
var slot = _wheel[pos];
if (slot is null || slot.Lowest > ts)
{
if (slot is not null && slot.Lowest < globalLowest)
globalLowest = slot.Lowest;
continue;
}
var slotLowest = long.MaxValue;
// Snapshot keys to allow removal during iteration.
var keys = slot.Entries.Keys.ToArray();
foreach (var seq in keys)
{
var exp = slot.Entries[seq];
if (exp <= ts && callback(seq, exp))
{
slot.Entries.Remove(seq);
_count--;
continue;
}
if (exp < slotLowest)
slotLowest = exp;
}
if (slot.Entries.Count == 0)
{
_wheel[pos] = null;
}
else
{
slot.Lowest = slotLowest;
if (slotLowest < globalLowest)
globalLowest = slotLowest;
}
}
_lowest = globalLowest;
}
/// <summary>
/// Returns the earliest expiration timestamp before <paramref name="before"/>,
/// or <see cref="long.MaxValue"/> if none.
/// </summary>
public long GetNextExpiration(long before) =>
_lowest < before ? _lowest : long.MaxValue;
// --- Encode / Decode ---
/// <summary>
/// Serializes the wheel to a byte array. <paramref name="highSeq"/> is stored
/// in the header and returned by <see cref="Decode"/>.
/// </summary>
public byte[] Encode(ulong highSeq)
{
// Preallocate conservatively: header + up to 2 varints per entry.
var buf = new List<byte>(HeaderLen + (int)(_count * 16));
buf.Add(1); // magic version
AppendUint64LE(buf, _count);
AppendUint64LE(buf, highSeq);
foreach (var slot in _wheel)
{
if (slot is null)
continue;
foreach (var (seq, ts) in slot.Entries)
{
AppendVarint(buf, ts);
AppendUvarint(buf, seq);
}
}
return buf.ToArray();
}
/// <summary>
/// Replaces this wheel's contents with those from a binary snapshot.
/// Returns the <c>highSeq</c> stored in the header.
/// </summary>
public ulong Decode(ReadOnlySpan<byte> b)
{
if (b.Length < HeaderLen)
throw (InvalidDataException)ErrInvalidVersion;
if (b[0] != 1)
throw (InvalidDataException)ErrInvalidVersion;
// Reset wheel.
Array.Clear(_wheel);
_lowest = long.MaxValue;
_count = 0;
var count = BinaryPrimitives.ReadUInt64LittleEndian(b[1..]);
var stamp = BinaryPrimitives.ReadUInt64LittleEndian(b[9..]);
var pos = HeaderLen;
for (ulong i = 0; i < count; i++)
{
var ts = ReadVarint(b, ref pos);
var seq = ReadUvarint(b, ref pos);
Add(seq, ts);
}
return stamp;
}
// --- Encoding helpers ---
private static void AppendUint64LE(List<byte> buf, ulong v)
{
buf.Add((byte)v);
buf.Add((byte)(v >> 8));
buf.Add((byte)(v >> 16));
buf.Add((byte)(v >> 24));
buf.Add((byte)(v >> 32));
buf.Add((byte)(v >> 40));
buf.Add((byte)(v >> 48));
buf.Add((byte)(v >> 56));
}
private static void AppendVarint(List<byte> buf, long v)
{
// ZigZag encode like Go's binary.AppendVarint.
var uv = (ulong)((v << 1) ^ (v >> 63));
AppendUvarint(buf, uv);
}
private static void AppendUvarint(List<byte> buf, ulong v)
{
while (v >= 0x80)
{
buf.Add((byte)(v | 0x80));
v >>= 7;
}
buf.Add((byte)v);
}
private static long ReadVarint(ReadOnlySpan<byte> b, ref int pos)
{
var uv = ReadUvarint(b, ref pos);
var v = (long)(uv >> 1);
if ((uv & 1) != 0)
v = ~v;
return v;
}
private static ulong ReadUvarint(ReadOnlySpan<byte> b, ref int pos)
{
ulong x = 0;
int s = 0;
while (pos < b.Length)
{
var by = b[pos++];
x |= (ulong)(by & 0x7F) << s;
if ((by & 0x80) == 0)
return x;
s += 7;
}
throw new InvalidDataException("thw: unexpected EOF in varint");
}
}

View File

@@ -0,0 +1,488 @@
// Copyright 2023-2025 The NATS Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
namespace ZB.MOM.NatsNet.Server.Internal.DataStructures;
/// <summary>
/// An adaptive radix trie (ART) for storing subject information on literal NATS subjects.
/// Uses dynamic nodes (4/10/16/48/256 children), path compression, and lazy expansion.
/// Supports exact lookup, wildcard matching ('*' and '>'), and ordered/fast iteration.
/// Not thread-safe.
/// </summary>
public sealed class SubjectTree<T>
{
internal ISubjectTreeNode<T>? _root;
private int _size;
/// <summary>Returns the number of entries stored in the tree.</summary>
public int Size() => _size;
/// <summary>Returns true if the tree has no entries.</summary>
public bool Empty() => _size == 0;
/// <summary>Clears all entries from the tree.</summary>
public SubjectTree<T> Reset()
{
_root = null;
_size = 0;
return this;
}
/// <summary>
/// Inserts a value into the tree under the given subject.
/// If the subject already exists, returns the old value with updated=true.
/// Subjects containing byte 127 (the noPivot sentinel) are rejected silently.
/// </summary>
public (T? oldVal, bool updated) Insert(ReadOnlySpan<byte> subject, T value)
{
if (subject.IndexOf(SubjectTreeParts.NoPivot) >= 0)
return (default, false);
var subjectBytes = subject.ToArray();
var (old, updated) = DoInsert(ref _root, subjectBytes, value, 0);
if (!updated)
_size++;
return (old, updated);
}
/// <summary>
/// Finds the value stored at the given literal subject.
/// Returns (value, true) if found, (default, false) otherwise.
/// </summary>
public (T? val, bool found) Find(ReadOnlySpan<byte> subject)
{
var si = 0;
var n = _root;
var subjectBytes = subject.ToArray();
while (n != null)
{
if (n.IsLeaf)
{
var ln = (SubjectTreeLeaf<T>)n;
return ln.Match(subjectBytes.AsSpan(si))
? (ln.Value, true)
: (default, false);
}
var prefix = n.Prefix;
if (prefix.Length > 0)
{
var end = Math.Min(si + prefix.Length, subjectBytes.Length);
if (!subjectBytes.AsSpan(si, end - si).SequenceEqual(prefix.AsSpan(0, end - si)))
return (default, false);
si += prefix.Length;
}
var next = n.FindChild(SubjectTreeParts.Pivot(subjectBytes, si));
if (next == null) return (default, false);
n = next;
}
return (default, false);
}
/// <summary>
/// Deletes the entry at the given literal subject.
/// Returns (value, true) if deleted, (default, false) if not found.
/// </summary>
public (T? val, bool found) Delete(ReadOnlySpan<byte> subject)
{
if (_root == null || subject.IsEmpty) return (default, false);
var subjectBytes = subject.ToArray();
var (val, deleted) = DoDelete(ref _root, subjectBytes, 0);
if (deleted) _size--;
return (val, deleted);
}
/// <summary>
/// Matches all stored subjects against a filter that may contain wildcards ('*' and '>').
/// Invokes fn for each match. Return false from the callback to stop early.
/// </summary>
public void Match(ReadOnlySpan<byte> filter, Func<byte[], T, bool> fn)
{
if (_root == null || filter.IsEmpty || fn == null) return;
var parts = SubjectTreeParts.GenParts(filter.ToArray());
MatchNode(_root, parts, Array.Empty<byte>(), fn);
}
/// <summary>
/// Like Match but returns false if the callback stopped iteration early.
/// Returns true if matching ran to completion.
/// </summary>
public bool MatchUntil(ReadOnlySpan<byte> filter, Func<byte[], T, bool> fn)
{
if (_root == null || filter.IsEmpty || fn == null) return true;
var parts = SubjectTreeParts.GenParts(filter.ToArray());
return MatchNode(_root, parts, Array.Empty<byte>(), fn);
}
/// <summary>
/// Walks all entries in lexicographical order.
/// Return false from the callback to stop early.
/// </summary>
public void IterOrdered(Func<byte[], T, bool> fn)
{
if (_root == null || fn == null) return;
IterNode(_root, Array.Empty<byte>(), ordered: true, fn);
}
/// <summary>
/// Walks all entries in storage order (no ordering guarantee).
/// Return false from the callback to stop early.
/// </summary>
public void IterFast(Func<byte[], T, bool> fn)
{
if (_root == null || fn == null) return;
IterNode(_root, Array.Empty<byte>(), ordered: false, fn);
}
// -------------------------------------------------------------------------
// Internal recursive insert
// -------------------------------------------------------------------------
private static (T? old, bool updated) DoInsert(ref ISubjectTreeNode<T>? np, byte[] subject, T value, int si)
{
if (np == null)
{
np = new SubjectTreeLeaf<T>(subject, value);
return (default, false);
}
if (np.IsLeaf)
{
var ln = (SubjectTreeLeaf<T>)np;
if (ln.Match(subject.AsSpan(si)))
{
var oldVal = ln.Value;
ln.Value = value;
return (oldVal, true);
}
// Split the leaf: compute common prefix between existing suffix and new subject tail.
var cpi = SubjectTreeParts.CommonPrefixLen(ln.Suffix, subject.AsSpan(si));
var nn = new SubjectTreeNode4<T>(subject[si..(si + cpi)]);
ln.SetSuffix(ln.Suffix[cpi..]);
si += cpi;
var p = SubjectTreeParts.Pivot(ln.Suffix, 0);
if (cpi > 0 && si < subject.Length && p == subject[si])
{
// Same pivot after the split — recurse to separate further.
DoInsert(ref np, subject, value, si);
nn.AddChild(p, np!);
}
else
{
var nl = new SubjectTreeLeaf<T>(subject[si..], value);
nn.AddChild(SubjectTreeParts.Pivot(nl.Suffix, 0), nl);
nn.AddChild(SubjectTreeParts.Pivot(ln.Suffix, 0), ln);
}
np = nn;
return (default, false);
}
// Non-leaf node.
var prefix = np.Prefix;
if (prefix.Length > 0)
{
var cpi = SubjectTreeParts.CommonPrefixLen(prefix, subject.AsSpan(si));
if (cpi >= prefix.Length)
{
// Full prefix match: move past this node.
si += prefix.Length;
var pivotByte = SubjectTreeParts.Pivot(subject, si);
var existingChild = np.FindChild(pivotByte);
if (existingChild != null)
{
var before = existingChild;
var (old, upd) = DoInsert(ref existingChild, subject, value, si);
// Only re-register if the child reference changed identity (grew or split).
if (!ReferenceEquals(before, existingChild))
{
np.DeleteChild(pivotByte);
np.AddChild(pivotByte, existingChild!);
}
return (old, upd);
}
if (np.IsFull)
np = np.Grow();
np.AddChild(SubjectTreeParts.Pivot(subject, si), new SubjectTreeLeaf<T>(subject[si..], value));
return (default, false);
}
else
{
// Partial prefix match — insert a new node4 above the current node.
var newPrefix = subject[si..(si + cpi)];
si += cpi;
var splitNode = new SubjectTreeNode4<T>(newPrefix);
((SubjectTreeMeta<T>)np).SetPrefix(prefix[cpi..]);
// Use np.Prefix (updated) to get the correct pivot for the demoted node.
splitNode.AddChild(SubjectTreeParts.Pivot(np.Prefix, 0), np);
splitNode.AddChild(
SubjectTreeParts.Pivot(subject.AsSpan(si), 0),
new SubjectTreeLeaf<T>(subject[si..], value));
np = splitNode;
}
}
else
{
// No prefix on this node.
var pivotByte = SubjectTreeParts.Pivot(subject, si);
var existingChild = np.FindChild(pivotByte);
if (existingChild != null)
{
var before = existingChild;
var (old, upd) = DoInsert(ref existingChild, subject, value, si);
if (!ReferenceEquals(before, existingChild))
{
np.DeleteChild(pivotByte);
np.AddChild(pivotByte, existingChild!);
}
return (old, upd);
}
if (np.IsFull)
np = np.Grow();
np.AddChild(SubjectTreeParts.Pivot(subject, si), new SubjectTreeLeaf<T>(subject[si..], value));
}
return (default, false);
}
// -------------------------------------------------------------------------
// Internal recursive delete
// -------------------------------------------------------------------------
private static (T? val, bool deleted) DoDelete(ref ISubjectTreeNode<T>? np, byte[] subject, int si)
{
if (np == null || subject.Length == 0) return (default, false);
var n = np;
if (n.IsLeaf)
{
var ln = (SubjectTreeLeaf<T>)n;
if (ln.Match(subject.AsSpan(si)))
{
np = null;
return (ln.Value, true);
}
return (default, false);
}
// Check prefix.
var prefix = n.Prefix;
if (prefix.Length > 0)
{
if (subject.Length < si + prefix.Length)
return (default, false);
if (!subject.AsSpan(si, prefix.Length).SequenceEqual(prefix))
return (default, false);
si += prefix.Length;
}
var p = SubjectTreeParts.Pivot(subject, si);
var childNode = n.FindChild(p);
if (childNode == null) return (default, false);
if (childNode.IsLeaf)
{
var childLeaf = (SubjectTreeLeaf<T>)childNode;
if (childLeaf.Match(subject.AsSpan(si)))
{
n.DeleteChild(p);
TryShrink(ref np!, prefix);
return (childLeaf.Value, true);
}
return (default, false);
}
// Recurse into non-leaf child.
var (val, deleted) = DoDelete(ref childNode, subject, si);
if (deleted)
{
if (childNode == null)
{
// Child was nulled out — remove slot and try to shrink.
n.DeleteChild(p);
TryShrink(ref np!, prefix);
}
else
{
// Child changed identity — re-register.
n.DeleteChild(p);
n.AddChild(p, childNode);
}
}
return (val, deleted);
}
private static void TryShrink(ref ISubjectTreeNode<T> np, byte[] parentPrefix)
{
var shrunk = np.Shrink();
if (shrunk == null) return;
if (shrunk.IsLeaf)
{
var shrunkLeaf = (SubjectTreeLeaf<T>)shrunk;
if (parentPrefix.Length > 0)
shrunkLeaf.Suffix = [.. parentPrefix, .. shrunkLeaf.Suffix];
}
else if (parentPrefix.Length > 0)
{
((SubjectTreeMeta<T>)shrunk).SetPrefix([.. parentPrefix, .. shrunk.Prefix]);
}
np = shrunk;
}
// -------------------------------------------------------------------------
// Internal recursive wildcard match
// -------------------------------------------------------------------------
private static bool MatchNode(ISubjectTreeNode<T> n, byte[][] parts, byte[] pre, Func<byte[], T, bool> fn)
{
var hasFwc = parts.Length > 0 && parts[^1].Length == 1 && parts[^1][0] == SubjectTreeParts.Fwc;
while (n != null!)
{
var (nparts, matched) = n.MatchParts(parts);
if (!matched) return true;
if (n.IsLeaf)
{
if (nparts.Length == 0 || (hasFwc && nparts.Length == 1))
{
var ln = (SubjectTreeLeaf<T>)n;
if (!fn(ConcatBytes(pre, ln.Suffix), ln.Value)) return false;
}
return true;
}
// Append this node's prefix to the running accumulator.
var prefix = n.Prefix;
if (prefix.Length > 0)
pre = ConcatBytes(pre, prefix);
if (nparts.Length == 0 && !hasFwc)
{
// No parts remaining and no fwc — look for terminal matches.
var hasTermPwc = parts.Length > 0 && parts[^1].Length == 1 && parts[^1][0] == SubjectTreeParts.Pwc;
var termParts = hasTermPwc ? parts[^1..] : Array.Empty<byte[]>();
foreach (var cn in n.Children())
{
if (cn == null!) continue;
if (cn.IsLeaf)
{
var ln = (SubjectTreeLeaf<T>)cn;
if (ln.Suffix.Length == 0)
{
if (!fn(ConcatBytes(pre, ln.Suffix), ln.Value)) return false;
}
else if (hasTermPwc && Array.IndexOf(ln.Suffix, SubjectTreeParts.TSep) < 0)
{
if (!fn(ConcatBytes(pre, ln.Suffix), ln.Value)) return false;
}
}
else if (hasTermPwc)
{
if (!MatchNode(cn, termParts, pre, fn)) return false;
}
}
return true;
}
// Re-put the terminal fwc if nparts was exhausted by matching.
if (hasFwc && nparts.Length == 0)
nparts = parts[^1..];
var fp = nparts[0];
var pByte = SubjectTreeParts.Pivot(fp, 0);
if (fp.Length == 1 && (pByte == SubjectTreeParts.Pwc || pByte == SubjectTreeParts.Fwc))
{
// Wildcard part — iterate all children.
foreach (var cn in n.Children())
{
if (cn != null!)
{
if (!MatchNode(cn, nparts, pre, fn)) return false;
}
}
return true;
}
// Literal part — find specific child and loop.
var nextNode = n.FindChild(pByte);
if (nextNode == null) return true;
n = nextNode;
parts = nparts;
}
return true;
}
// -------------------------------------------------------------------------
// Internal iteration
// -------------------------------------------------------------------------
private static bool IterNode(ISubjectTreeNode<T> n, byte[] pre, bool ordered, Func<byte[], T, bool> fn)
{
if (n.IsLeaf)
{
var ln = (SubjectTreeLeaf<T>)n;
return fn(ConcatBytes(pre, ln.Suffix), ln.Value);
}
pre = ConcatBytes(pre, n.Prefix);
if (!ordered)
{
foreach (var cn in n.Children())
{
if (cn == null!) continue;
if (!IterNode(cn, pre, false, fn)) return false;
}
return true;
}
// Ordered: sort children by their path bytes lexicographically.
var children = n.Children().Where(c => c != null!).ToArray();
Array.Sort(children, static (a, b) => a.Path.AsSpan().SequenceCompareTo(b.Path.AsSpan()));
foreach (var cn in children)
{
if (!IterNode(cn, pre, true, fn)) return false;
}
return true;
}
// -------------------------------------------------------------------------
// Byte array helpers
// -------------------------------------------------------------------------
internal static byte[] ConcatBytes(byte[] a, byte[] b)
{
if (a.Length == 0) return b.Length == 0 ? Array.Empty<byte>() : b;
if (b.Length == 0) return a;
var result = new byte[a.Length + b.Length];
a.CopyTo(result, 0);
b.CopyTo(result, a.Length);
return result;
}
}

View File

@@ -0,0 +1,483 @@
// Copyright 2023-2025 The NATS Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
namespace ZB.MOM.NatsNet.Server.Internal.DataStructures;
// Internal node interface for the adaptive radix trie.
internal interface ISubjectTreeNode<T>
{
bool IsLeaf { get; }
byte[] Prefix { get; }
void AddChild(byte key, ISubjectTreeNode<T> child);
ISubjectTreeNode<T>? FindChild(byte key);
void DeleteChild(byte key);
bool IsFull { get; }
ISubjectTreeNode<T> Grow();
ISubjectTreeNode<T>? Shrink();
ISubjectTreeNode<T>[] Children();
int NumChildren { get; }
byte[] Path { get; }
(byte[][] remainingParts, bool matched) MatchParts(byte[][] parts);
string Kind { get; }
}
// Base class for non-leaf nodes, holding prefix and child count.
internal abstract class SubjectTreeMeta<T> : ISubjectTreeNode<T>
{
protected byte[] _prefix;
protected int _size;
protected SubjectTreeMeta(byte[] prefix)
{
_prefix = SubjectTreeParts.CopyBytes(prefix);
}
public bool IsLeaf => false;
public byte[] Prefix => _prefix;
public int NumChildren => _size;
public byte[] Path => _prefix;
public void SetPrefix(byte[] prefix)
{
_prefix = SubjectTreeParts.CopyBytes(prefix);
}
public (byte[][] remainingParts, bool matched) MatchParts(byte[][] parts)
=> SubjectTreeParts.MatchParts(parts, _prefix);
public abstract void AddChild(byte key, ISubjectTreeNode<T> child);
public abstract ISubjectTreeNode<T>? FindChild(byte key);
public abstract void DeleteChild(byte key);
public abstract bool IsFull { get; }
public abstract ISubjectTreeNode<T> Grow();
public abstract ISubjectTreeNode<T>? Shrink();
public abstract ISubjectTreeNode<T>[] Children();
public abstract string Kind { get; }
}
// Leaf node storing the terminal value plus a suffix byte[].
internal sealed class SubjectTreeLeaf<T> : ISubjectTreeNode<T>
{
public T Value;
public byte[] Suffix;
public SubjectTreeLeaf(byte[] suffix, T value)
{
Suffix = SubjectTreeParts.CopyBytes(suffix);
Value = value;
}
public bool IsLeaf => true;
public byte[] Prefix => Array.Empty<byte>();
public int NumChildren => 0;
public byte[] Path => Suffix;
public string Kind => "LEAF";
public bool Match(ReadOnlySpan<byte> subject)
=> subject.SequenceEqual(Suffix);
public void SetSuffix(byte[] suffix)
=> Suffix = SubjectTreeParts.CopyBytes(suffix);
public bool IsFull => true;
public (byte[][] remainingParts, bool matched) MatchParts(byte[][] parts)
=> SubjectTreeParts.MatchParts(parts, Suffix);
// Leaf nodes do not support child operations.
public void AddChild(byte key, ISubjectTreeNode<T> child)
=> throw new InvalidOperationException("AddChild called on leaf");
public ISubjectTreeNode<T>? FindChild(byte key)
=> throw new InvalidOperationException("FindChild called on leaf");
public void DeleteChild(byte key)
=> throw new InvalidOperationException("DeleteChild called on leaf");
public ISubjectTreeNode<T> Grow()
=> throw new InvalidOperationException("Grow called on leaf");
public ISubjectTreeNode<T>? Shrink()
=> throw new InvalidOperationException("Shrink called on leaf");
public ISubjectTreeNode<T>[] Children()
=> Array.Empty<ISubjectTreeNode<T>>();
}
// Node with up to 4 children (keys + children arrays, unsorted).
internal sealed class SubjectTreeNode4<T> : SubjectTreeMeta<T>
{
private readonly byte[] _keys = new byte[4];
private readonly ISubjectTreeNode<T>?[] _children = new ISubjectTreeNode<T>?[4];
public SubjectTreeNode4(byte[] prefix) : base(prefix) { }
public override string Kind => "NODE4";
public override void AddChild(byte key, ISubjectTreeNode<T> child)
{
if (_size >= 4) throw new InvalidOperationException("node4 full!");
_keys[_size] = key;
_children[_size] = child;
_size++;
}
public override ISubjectTreeNode<T>? FindChild(byte key)
{
for (var i = 0; i < _size; i++)
{
if (_keys[i] == key) return _children[i];
}
return null;
}
public override void DeleteChild(byte key)
{
for (var i = 0; i < _size; i++)
{
if (_keys[i] == key)
{
var last = _size - 1;
if (i < last)
{
_keys[i] = _keys[last];
_children[i] = _children[last];
}
_keys[last] = 0;
_children[last] = null;
_size--;
return;
}
}
}
public override bool IsFull => _size >= 4;
public override ISubjectTreeNode<T> Grow()
{
var nn = new SubjectTreeNode10<T>(_prefix);
for (var i = 0; i < 4; i++)
nn.AddChild(_keys[i], _children[i]!);
return nn;
}
public override ISubjectTreeNode<T>? Shrink()
{
if (_size == 1) return _children[0];
return null;
}
public override ISubjectTreeNode<T>[] Children()
{
var result = new ISubjectTreeNode<T>[_size];
for (var i = 0; i < _size; i++)
result[i] = _children[i]!;
return result;
}
// Internal access for tests.
internal byte GetKey(int index) => _keys[index];
internal ISubjectTreeNode<T>? GetChild(int index) => _children[index];
}
// Node with up to 10 children (for numeric token segments).
internal sealed class SubjectTreeNode10<T> : SubjectTreeMeta<T>
{
private readonly byte[] _keys = new byte[10];
private readonly ISubjectTreeNode<T>?[] _children = new ISubjectTreeNode<T>?[10];
public SubjectTreeNode10(byte[] prefix) : base(prefix) { }
public override string Kind => "NODE10";
public override void AddChild(byte key, ISubjectTreeNode<T> child)
{
if (_size >= 10) throw new InvalidOperationException("node10 full!");
_keys[_size] = key;
_children[_size] = child;
_size++;
}
public override ISubjectTreeNode<T>? FindChild(byte key)
{
for (var i = 0; i < _size; i++)
{
if (_keys[i] == key) return _children[i];
}
return null;
}
public override void DeleteChild(byte key)
{
for (var i = 0; i < _size; i++)
{
if (_keys[i] == key)
{
var last = _size - 1;
if (i < last)
{
_keys[i] = _keys[last];
_children[i] = _children[last];
}
_keys[last] = 0;
_children[last] = null;
_size--;
return;
}
}
}
public override bool IsFull => _size >= 10;
public override ISubjectTreeNode<T> Grow()
{
var nn = new SubjectTreeNode16<T>(_prefix);
for (var i = 0; i < _size; i++)
nn.AddChild(_keys[i], _children[i]!);
return nn;
}
public override ISubjectTreeNode<T>? Shrink()
{
if (_size > 4) return null;
var nn = new SubjectTreeNode4<T>(Array.Empty<byte>());
for (var i = 0; i < _size; i++)
nn.AddChild(_keys[i], _children[i]!);
return nn;
}
public override ISubjectTreeNode<T>[] Children()
{
var result = new ISubjectTreeNode<T>[_size];
for (var i = 0; i < _size; i++)
result[i] = _children[i]!;
return result;
}
}
// Node with up to 16 children.
internal sealed class SubjectTreeNode16<T> : SubjectTreeMeta<T>
{
private readonly byte[] _keys = new byte[16];
private readonly ISubjectTreeNode<T>?[] _children = new ISubjectTreeNode<T>?[16];
public SubjectTreeNode16(byte[] prefix) : base(prefix) { }
public override string Kind => "NODE16";
public override void AddChild(byte key, ISubjectTreeNode<T> child)
{
if (_size >= 16) throw new InvalidOperationException("node16 full!");
_keys[_size] = key;
_children[_size] = child;
_size++;
}
public override ISubjectTreeNode<T>? FindChild(byte key)
{
for (var i = 0; i < _size; i++)
{
if (_keys[i] == key) return _children[i];
}
return null;
}
public override void DeleteChild(byte key)
{
for (var i = 0; i < _size; i++)
{
if (_keys[i] == key)
{
var last = _size - 1;
if (i < last)
{
_keys[i] = _keys[last];
_children[i] = _children[last];
}
_keys[last] = 0;
_children[last] = null;
_size--;
return;
}
}
}
public override bool IsFull => _size >= 16;
public override ISubjectTreeNode<T> Grow()
{
var nn = new SubjectTreeNode48<T>(_prefix);
for (var i = 0; i < _size; i++)
nn.AddChild(_keys[i], _children[i]!);
return nn;
}
public override ISubjectTreeNode<T>? Shrink()
{
if (_size > 10) return null;
var nn = new SubjectTreeNode10<T>(Array.Empty<byte>());
for (var i = 0; i < _size; i++)
nn.AddChild(_keys[i], _children[i]!);
return nn;
}
public override ISubjectTreeNode<T>[] Children()
{
var result = new ISubjectTreeNode<T>[_size];
for (var i = 0; i < _size; i++)
result[i] = _children[i]!;
return result;
}
}
// Node with up to 48 children, using a 256-byte key index (1-indexed, 0 means empty).
internal sealed class SubjectTreeNode48<T> : SubjectTreeMeta<T>
{
// _keyIndex[byte] = 1-based index into _children; 0 means no entry.
private readonly byte[] _keyIndex = new byte[256];
private readonly ISubjectTreeNode<T>?[] _children = new ISubjectTreeNode<T>?[48];
public SubjectTreeNode48(byte[] prefix) : base(prefix) { }
public override string Kind => "NODE48";
public override void AddChild(byte key, ISubjectTreeNode<T> child)
{
if (_size >= 48) throw new InvalidOperationException("node48 full!");
_children[_size] = child;
_keyIndex[key] = (byte)(_size + 1); // 1-indexed
_size++;
}
public override ISubjectTreeNode<T>? FindChild(byte key)
{
var i = _keyIndex[key];
if (i == 0) return null;
return _children[i - 1];
}
public override void DeleteChild(byte key)
{
var i = _keyIndex[key];
if (i == 0) return;
i--; // Convert from 1-indexed
var last = _size - 1;
if (i < last)
{
_children[i] = _children[last];
// Find which key index points to 'last' and redirect it to 'i'.
for (var ic = 0; ic < 256; ic++)
{
if (_keyIndex[ic] == last + 1)
{
_keyIndex[ic] = (byte)(i + 1);
break;
}
}
}
_children[last] = null;
_keyIndex[key] = 0;
_size--;
}
public override bool IsFull => _size >= 48;
public override ISubjectTreeNode<T> Grow()
{
var nn = new SubjectTreeNode256<T>(_prefix);
for (var c = 0; c < 256; c++)
{
var i = _keyIndex[c];
if (i > 0)
nn.AddChild((byte)c, _children[i - 1]!);
}
return nn;
}
public override ISubjectTreeNode<T>? Shrink()
{
if (_size > 16) return null;
var nn = new SubjectTreeNode16<T>(Array.Empty<byte>());
for (var c = 0; c < 256; c++)
{
var i = _keyIndex[c];
if (i > 0)
nn.AddChild((byte)c, _children[i - 1]!);
}
return nn;
}
public override ISubjectTreeNode<T>[] Children()
{
var result = new ISubjectTreeNode<T>[_size];
var idx = 0;
for (var i = 0; i < _size; i++)
{
if (_children[i] != null)
result[idx++] = _children[i]!;
}
return result[..idx];
}
// Internal access for tests.
internal byte GetKeyIndex(int key) => _keyIndex[key];
internal ISubjectTreeNode<T>? GetChildAt(int index) => _children[index];
}
// Node with 256 children, indexed directly by byte value.
internal sealed class SubjectTreeNode256<T> : SubjectTreeMeta<T>
{
private readonly ISubjectTreeNode<T>?[] _children = new ISubjectTreeNode<T>?[256];
public SubjectTreeNode256(byte[] prefix) : base(prefix) { }
public override string Kind => "NODE256";
public override void AddChild(byte key, ISubjectTreeNode<T> child)
{
_children[key] = child;
_size++;
}
public override ISubjectTreeNode<T>? FindChild(byte key)
=> _children[key];
public override void DeleteChild(byte key)
{
if (_children[key] != null)
{
_children[key] = null;
_size--;
}
}
public override bool IsFull => false;
public override ISubjectTreeNode<T> Grow()
=> throw new InvalidOperationException("Grow cannot be called on node256");
public override ISubjectTreeNode<T>? Shrink()
{
if (_size > 48) return null;
var nn = new SubjectTreeNode48<T>(Array.Empty<byte>());
for (var c = 0; c < 256; c++)
{
if (_children[c] != null)
nn.AddChild((byte)c, _children[c]!);
}
return nn;
}
public override ISubjectTreeNode<T>[] Children()
=> _children.Where(c => c != null).Select(c => c!).ToArray();
}

View File

@@ -0,0 +1,242 @@
// Copyright 2023-2025 The NATS Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
namespace ZB.MOM.NatsNet.Server.Internal.DataStructures;
/// <summary>
/// Utility methods for NATS subject matching, wildcard part decomposition,
/// common prefix computation, and byte manipulation used by SubjectTree.
/// </summary>
internal static class SubjectTreeParts
{
// NATS subject special bytes.
internal const byte Pwc = (byte)'*'; // single-token wildcard
internal const byte Fwc = (byte)'>'; // full wildcard (terminal)
internal const byte TSep = (byte)'.'; // token separator
// Sentinel pivot returned when subject position is past end.
internal const byte NoPivot = 127;
/// <summary>
/// Returns the pivot byte at <paramref name="pos"/> in <paramref name="subject"/>,
/// or <see cref="NoPivot"/> if the position is at or beyond the end.
/// </summary>
internal static byte Pivot(ReadOnlySpan<byte> subject, int pos)
=> pos >= subject.Length ? NoPivot : subject[pos];
/// <summary>
/// Returns the pivot byte at <paramref name="pos"/> in <paramref name="subject"/>,
/// or <see cref="NoPivot"/> if the position is at or beyond the end.
/// </summary>
internal static byte Pivot(byte[] subject, int pos)
=> pos >= subject.Length ? NoPivot : subject[pos];
/// <summary>
/// Computes the number of leading bytes that are equal between two spans.
/// </summary>
internal static int CommonPrefixLen(ReadOnlySpan<byte> s1, ReadOnlySpan<byte> s2)
{
var limit = Math.Min(s1.Length, s2.Length);
var i = 0;
while (i < limit && s1[i] == s2[i])
i++;
return i;
}
/// <summary>
/// Returns a copy of <paramref name="src"/>, or an empty array if src is empty.
/// </summary>
internal static byte[] CopyBytes(ReadOnlySpan<byte> src)
{
if (src.IsEmpty) return Array.Empty<byte>();
return src.ToArray();
}
/// <summary>
/// Returns a copy of <paramref name="src"/>, or an empty array if src is null or empty.
/// </summary>
internal static byte[] CopyBytes(byte[]? src)
{
if (src == null || src.Length == 0) return Array.Empty<byte>();
var dst = new byte[src.Length];
src.CopyTo(dst, 0);
return dst;
}
/// <summary>
/// Converts a byte array to a string using Latin-1 (ISO-8859-1) encoding,
/// which preserves a 1:1 byte-to-char mapping for all byte values 0-255.
/// </summary>
internal static string BytesToString(byte[] bytes)
{
if (bytes.Length == 0) return string.Empty;
return System.Text.Encoding.Latin1.GetString(bytes);
}
/// <summary>
/// Breaks a filter subject into parts separated by wildcards ('*' and '>').
/// Each literal segment between wildcards becomes one part; each wildcard
/// becomes its own single-byte part.
/// </summary>
internal static byte[][] GenParts(byte[] filter)
{
var parts = new List<byte[]>(8);
var start = 0;
var e = filter.Length - 1;
for (var i = 0; i < filter.Length; i++)
{
if (filter[i] == TSep)
{
// Check if next token is pwc (internal or terminal).
if (i < e && filter[i + 1] == Pwc &&
((i + 2 <= e && filter[i + 2] == TSep) || i + 1 == e))
{
if (i > start)
parts.Add(filter[start..(i + 1)]);
parts.Add(filter[(i + 1)..(i + 2)]);
i++; // skip pwc
if (i + 2 <= e)
i++; // skip next tsep from next part
start = i + 1;
}
else if (i < e && filter[i + 1] == Fwc && i + 1 == e)
{
if (i > start)
parts.Add(filter[start..(i + 1)]);
parts.Add(filter[(i + 1)..(i + 2)]);
i++; // skip fwc
start = i + 1;
}
}
else if (filter[i] == Pwc || filter[i] == Fwc)
{
// Wildcard must be preceded by tsep (or be at start).
var prev = i - 1;
if (prev >= 0 && filter[prev] != TSep)
continue;
// Wildcard must be at end or followed by tsep.
var next = i + 1;
if (next == e || (next < e && filter[next] != TSep))
continue;
// Full wildcard must be terminal.
if (filter[i] == Fwc && i < e)
break;
// Leading wildcard.
parts.Add(filter[i..(i + 1)]);
if (i + 1 <= e)
i++; // skip next tsep
start = i + 1;
}
}
if (start < filter.Length)
{
// Eat leading tsep if present.
if (filter[start] == TSep)
start++;
if (start < filter.Length)
parts.Add(filter[start..]);
}
return parts.ToArray();
}
/// <summary>
/// Matches parts against a fragment (prefix or suffix).
/// Returns the remaining parts and whether matching succeeded.
/// </summary>
internal static (byte[][] remainingParts, bool matched) MatchParts(byte[][] parts, byte[] frag)
{
var lf = frag.Length;
if (lf == 0) return (parts, true);
var si = 0;
var lpi = parts.Length - 1;
for (var i = 0; i < parts.Length; i++)
{
if (si >= lf)
return (parts[i..], true);
var part = parts[i];
var lp = part.Length;
// Check for wildcard placeholders.
if (lp == 1)
{
if (part[0] == Pwc)
{
// Find the next token separator.
var index = Array.IndexOf(frag, TSep, si);
if (index < 0)
{
// No tsep found.
if (i == lpi)
return (Array.Empty<byte[]>(), true);
return (parts[i..], true);
}
si = index + 1;
continue;
}
else if (part[0] == Fwc)
{
return (Array.Empty<byte[]>(), true);
}
}
var end = Math.Min(si + lp, lf);
// If part is larger than the remaining fragment, adjust.
var comparePart = part;
if (si + lp > end)
comparePart = part[..(end - si)];
if (!frag.AsSpan(si, end - si).SequenceEqual(comparePart))
return (parts, false);
// Fragment still has bytes left.
if (end < lf)
{
si = end;
continue;
}
// We matched a partial part.
if (end < si + lp)
{
if (end >= lf)
{
// Create a copy of parts with the current part trimmed.
var newParts = new byte[parts.Length][];
parts.CopyTo(newParts, 0);
newParts[i] = parts[i][(lf - si)..];
return (newParts[i..], true);
}
else
{
return (parts[(i + 1)..], true);
}
}
if (i == lpi)
return (Array.Empty<byte[]>(), true);
si += part.Length;
}
return (parts, false);
}
}