@stacksjs/zig-dtsx 0.9.12 → 0.9.14

package/src/scanner.zig

@@ -8,6 +8,23 @@ const Declaration = types.Declaration;
 const DeclarationKind = types.DeclarationKind;
 const Allocator = std.mem.Allocator;
 
+/// Comptime-optimized word comparison: uses integer casts for power-of-2 sizes (1/2/4/8 bytes).
+inline fn comptime_match(comptime N: usize, src: *const [N]u8, comptime word: []const u8) bool {
+    // Use integer comparison for power-of-2 sizes (u8, u16, u32, u64)
+    if (N == 1 or N == 2 or N == 4 or N == 8) {
+        const T = std.meta.Int(.unsigned, N * 8);
+        const expected: T = comptime blk: {
+            var val: T = 0;
+            for (word, 0..) |b, i| {
+                val |= @as(T, b) << @intCast(i * 8);
+            }
+            break :blk val;
+        };
+        return @as(*align(1) const T, @ptrCast(src)).* == expected;
+    }
+    return std.mem.eql(u8, src, word);
+}
+
 /// Result of scanning: declarations + non-exported types
 pub const ScanResult = struct {
     declarations: std.array_list.Managed(Declaration),
@@ -21,8 +38,12 @@ pub const Scanner = struct {
     len: usize,
     allocator: Allocator,
     declarations: std.array_list.Managed(Declaration),
+    /// Lazy-initialized: only allocated on first insertion to avoid overhead for
+    /// files that have no non-exported types or overloaded functions.
     non_exported_types: std.StringHashMap(Declaration),
     func_body_indices: std.AutoHashMap(usize, void),
+    non_exported_types_inited: bool,
+    func_body_indices_inited: bool,
     keep_comments: bool,
     isolated_declarations: bool,
 
@@ -38,6 +59,8 @@ pub const Scanner = struct {
             .declarations = declarations,
             .non_exported_types = std.StringHashMap(Declaration).init(allocator),
             .func_body_indices = std.AutoHashMap(usize, void).init(allocator),
+            .non_exported_types_inited = false,
+            .func_body_indices_inited = false,
             .keep_comments = keep_comments,
             .isolated_declarations = isolated_declarations,
         };
@@ -45,8 +68,24 @@ pub const Scanner = struct {
 
     pub fn deinit(self: *Scanner) void {
         self.declarations.deinit();
-        self.non_exported_types.deinit();
-        self.func_body_indices.deinit();
+        if (self.non_exported_types_inited) self.non_exported_types.deinit();
+        if (self.func_body_indices_inited) self.func_body_indices.deinit();
+    }
+
+    /// Lazy-put into non_exported_types (initializes the map on first use)
+    pub fn putNonExportedType(self: *Scanner, name: []const u8, decl: Declaration) void {
+        if (!self.non_exported_types_inited) {
+            self.non_exported_types_inited = true;
+        }
+        self.non_exported_types.put(name, decl) catch {};
+    }
+
+    /// Lazy-put into func_body_indices (initializes on first use)
+    pub fn putFuncBodyIndex(self: *Scanner, idx: usize) void {
+        if (!self.func_body_indices_inited) {
+            self.func_body_indices_inited = true;
+        }
+        self.func_body_indices.put(idx, {}) catch {};
     }
 
     // ========================================================================
@@ -78,21 +117,30 @@ pub const Scanner = struct {
             }
             if (self.pos >= self.len) break;
 
+            // Scalar: consume single whitespace bytes
             const c = self.source[self.pos];
             if (c == ch.CH_SPACE or c == ch.CH_TAB or c == ch.CH_LF or c == ch.CH_CR) {
                 self.pos += 1;
+                // Fast scalar drain: consume consecutive whitespace without re-entering the outer loop
+                while (self.pos < self.len) {
+                    const c2 = self.source[self.pos];
+                    if (c2 == ch.CH_SPACE or c2 == ch.CH_TAB or c2 == ch.CH_LF or c2 == ch.CH_CR) {
+                        self.pos += 1;
+                    } else break;
+                }
                 continue;
             }
             if (c == ch.CH_SLASH and self.pos + 1 < self.len) {
                 const next = self.source[self.pos + 1];
                 if (next == ch.CH_SLASH) {
-                    // Line comment
+                    // Line comment — SIMD scan via indexOfChar.
                     const nl = ch.indexOfChar(self.source, ch.CH_LF, self.pos + 2);
                     self.pos = if (nl) |n| n + 1 else self.len;
                     continue;
                 }
                 if (next == ch.CH_STAR) {
-                    // Block comment
+                    // Block comment — `*/` is two specific bytes; the SIMD
+                    // first-byte scan in indexOf handles this efficiently.
                     const end_idx = ch.indexOf(self.source, "*/", self.pos + 2);
                     self.pos = if (end_idx) |e| e + 2 else self.len;
                     continue;
@@ -143,33 +191,70 @@ pub const Scanner = struct {
         self.pos = len;
     }
 
-    /// Skip past a template literal (backtick string with ${} interpolation)
+    /// Skip past a template literal (backtick string with ${} interpolation).
+    /// Uses SIMD to scan 16 bytes at a time for structural characters.
     pub fn skipTemplateLiteral(self: *Scanner) void {
         self.pos += 1; // skip opening backtick
+        const src = self.source;
+        const len = self.len;
+        var pos = self.pos;
         var depth: usize = 0;
-        while (self.pos < self.len) {
-            const c = self.source[self.pos];
+
+        while (pos < len) {
+            // SIMD fast-skip when at depth 0: scan for backtick, backslash, or $
+            // At depth > 0: also scan for }
+            if (depth == 0) {
+                while (pos + 16 <= len) {
+                    const chunk: @Vector(16, u8) = src[pos..][0..16].*;
+                    const match_mask = (chunk == @as(@Vector(16, u8), @splat(ch.CH_BACKTICK))) |
+                        (chunk == @as(@Vector(16, u8), @splat(ch.CH_BACKSLASH))) |
+                        (chunk == @as(@Vector(16, u8), @splat(ch.CH_DOLLAR)));
+                    if (@reduce(.Or, match_mask)) {
+                        const bits: u16 = @bitCast(match_mask);
+                        pos += @ctz(bits);
+                        break;
+                    }
+                    pos += 16;
+                }
+            } else {
+                while (pos + 16 <= len) {
+                    const chunk: @Vector(16, u8) = src[pos..][0..16].*;
+                    const match_mask = (chunk == @as(@Vector(16, u8), @splat(ch.CH_BACKTICK))) |
+                        (chunk == @as(@Vector(16, u8), @splat(ch.CH_BACKSLASH))) |
+                        (chunk == @as(@Vector(16, u8), @splat(ch.CH_DOLLAR))) |
+                        (chunk == @as(@Vector(16, u8), @splat(ch.CH_RBRACE)));
+                    if (@reduce(.Or, match_mask)) {
+                        const bits: u16 = @bitCast(match_mask);
+                        pos += @ctz(bits);
+                        break;
+                    }
+                    pos += 16;
+                }
+            }
+
+            if (pos >= len) break;
+            const c = src[pos];
             if (c == ch.CH_BACKSLASH) {
-                self.pos += 1;
-                if (self.pos < self.len) self.pos += 1;
+                pos += 2; // skip escaped char
                 continue;
             }
             if (c == ch.CH_BACKTICK and depth == 0) {
-                self.pos += 1;
+                self.pos = pos + 1;
                 return;
             }
-            if (c == ch.CH_DOLLAR and self.pos + 1 < self.len and self.source[self.pos + 1] == ch.CH_LBRACE) {
-                self.pos += 2;
+            if (c == ch.CH_DOLLAR and pos + 1 < len and src[pos + 1] == ch.CH_LBRACE) {
+                pos += 2;
                 depth += 1;
                 continue;
             }
             if (c == ch.CH_RBRACE and depth > 0) {
                 depth -= 1;
-                self.pos += 1;
+                pos += 1;
                 continue;
             }
-            self.pos += 1;
+            pos += 1;
         }
+        self.pos = len;
     }
 
     /// Check if `/` at current pos starts a regex literal (not division)
@@ -196,10 +281,22 @@ pub const Scanner = struct {
             const word_start: usize = @intCast(wp + 1);
             const word_end: usize = @intCast(p + 1);
             const word = self.source[word_start..word_end];
-            const keywords = [_][]const u8{ "return", "typeof", "void", "delete", "throw", "new", "in", "of", "case", "instanceof", "yield", "await" };
-            for (keywords) |kw| {
-                if (std.mem.eql(u8, word, kw)) return true;
-            }
+            // First-char dispatch to avoid 12 std.mem.eql calls per slash. The
+            // hot path here runs on every '/' that might start division.
+            if (word.len < 2 or word.len > 10) return false;
+            return switch (word[0]) {
+                'r' => std.mem.eql(u8, word, "return"),
+                't' => std.mem.eql(u8, word, "typeof") or std.mem.eql(u8, word, "throw"),
+                'v' => std.mem.eql(u8, word, "void"),
+                'd' => std.mem.eql(u8, word, "delete"),
+                'n' => std.mem.eql(u8, word, "new"),
+                'i' => std.mem.eql(u8, word, "in") or std.mem.eql(u8, word, "instanceof"),
+                'o' => std.mem.eql(u8, word, "of"),
+                'c' => std.mem.eql(u8, word, "case"),
+                'y' => std.mem.eql(u8, word, "yield"),
+                'a' => std.mem.eql(u8, word, "await"),
+                else => false,
+            };
         }
         return false;
     }
@@ -268,18 +365,21 @@ pub const Scanner = struct {
     }
 
     /// Read an identifier at current position
-    pub fn readIdent(self: *Scanner) []const u8 {
+    pub inline fn readIdent(self: *Scanner) []const u8 {
         const start = self.pos;
         while (self.pos < self.len and ch.isIdentChar(self.source[self.pos])) self.pos += 1;
         return self.source[start..self.pos];
     }
 
-    /// Check if source matches a word at pos (followed by non-ident char)
-    pub fn matchWord(self: *const Scanner, word: []const u8) bool {
+    /// Check if source matches a word at pos (followed by non-ident char).
+    /// Uses integer comparison for short words to avoid byte-by-byte loop.
+    pub fn matchWord(self: *const Scanner, comptime word: []const u8) bool {
         if (self.pos + word.len > self.len) return false;
-        if (!std.mem.eql(u8, self.source[self.pos .. self.pos + word.len], word)) return false;
+        // Boundary check: word must not be followed by an identifier char
         if (self.pos + word.len < self.len and ch.isIdentChar(self.source[self.pos + word.len])) return false;
-        return true;
+        // Integer comparison: compare 1-8 bytes as a single integer (comptime-optimized)
+        const src_slice = self.source[self.pos..][0..word.len];
+        return comptime_match(word.len, src_slice, word);
     }
 
     /// Check if current position is at a top-level statement-starting keyword
@@ -367,10 +467,17 @@ pub const Scanner = struct {
             self.pos = saved;
             return false;
         }
-        // Type continuation keywords
-        if (self.matchWord("extends") or self.matchWord("keyof") or self.matchWord("typeof") or
-            self.matchWord("infer") or self.matchWord("is") or self.matchWord("as") or self.matchWord("in"))
-        {
+        // Type continuation keywords — first-byte dispatch avoids running 7
+        // sequential matchWord calls on every newline at member depth 0.
+        const is_type_kw = switch (nc) {
+            'e' => self.matchWord("extends"),
+            'k' => self.matchWord("keyof"),
+            't' => self.matchWord("typeof"),
+            'i' => self.matchWord("infer") or self.matchWord("is") or self.matchWord("in"),
+            'a' => self.matchWord("as"),
+            else => false,
+        };
+        if (is_type_kw) {
             self.pos = saved;
             return false;
         }
@@ -581,8 +688,11 @@ pub const Scanner = struct {
         if (self.pos < self.len and self.source[self.pos] == ch.CH_SEMI) self.pos += 1;
     }
 
-    /// Peek at what char comes after a word (skipping whitespace)
-    pub fn peekAfterWord(self: *const Scanner, word: []const u8) u8 {
+    /// Peek at what char comes after a word (skipping whitespace).
+    /// Defensive bounds check on pos+word.len handles the edge where the
+    /// caller positions at end-of-source.
+    pub inline fn peekAfterWord(self: *const Scanner, word: []const u8) u8 {
+        if (self.pos + word.len > self.len) return 0;
         var p = self.pos + word.len;
         while (p < self.len and ch.isWhitespace(self.source[p])) p += 1;
         return if (p < self.len) self.source[p] else 0;
@@ -592,6 +702,9 @@ pub const Scanner = struct {
     pub fn peekAfterKeyword(self: *const Scanner, word1: []const u8, word2: []const u8) bool {
         var p = self.pos + word1.len;
         while (p < self.len and ch.isWhitespace(self.source[p])) p += 1;
+        // First-byte fast-fail: avoid std.mem.eql call when first char doesn't match.
+        if (p >= self.len or word2.len == 0) return false;
+        if (self.source[p] != word2[0]) return false;
         if (p + word2.len > self.len) return false;
         if (!std.mem.eql(u8, self.source[p .. p + word2.len], word2)) return false;
         return p + word2.len >= self.len or !ch.isIdentChar(self.source[p + word2.len]);
@@ -614,13 +727,12 @@ pub const Scanner = struct {
             return self.source[start..self.pos];
         }
         if (c == ch.CH_HASH) {
+            // Zero-copy: keep the leading `#` as part of the source slice instead
+            // of allocating a new buffer to prepend it.
+            const start = self.pos;
             self.pos += 1;
-            const ident = self.readIdent();
-            // Return "#name" — we need to allocate this
-            const result = self.allocator.alloc(u8, 1 + ident.len) catch return "";
-            result[0] = '#';
-            @memcpy(result[1..], ident);
-            return result;
+            _ = self.readIdent();
+            return self.source[start..self.pos];
         }
         return self.readIdent();
     }
@@ -651,24 +763,23 @@ pub const Scanner = struct {
 
     /// Scan TypeScript source and extract all declarations
     pub fn scan(self: *Scanner) !ScanResult {
-        // Skip BOM
-        if (self.pos < self.len and self.source[0] >= 0xEF) {
-            // UTF-8 BOM is EF BB BF
-            if (self.len >= 3 and self.source[0] == 0xEF and self.source[1] == 0xBB and self.source[2] == 0xBF) {
-                self.pos = 3;
-            }
+        // Skip UTF-8 BOM (EF BB BF). Check exact bytes — the previous `>= 0xEF`
+        // guard was over-permissive, accepting non-BOM high bytes only to fall
+        // through the strict triple-check.
+        if (self.len >= 3 and self.source[0] == 0xEF and self.source[1] == 0xBB and self.source[2] == 0xBF) {
+            self.pos = 3;
         }
 
         // Main scan loop — delegate to scan_loop module
         try @import("scan_loop.zig").scanMainLoop(self);
 
-        // Post-process: resolve referenced non-exported types
-        if (self.non_exported_types.count() > 0) {
+        // Post-process: resolve referenced non-exported types (only if map was used)
+        if (self.non_exported_types_inited and self.non_exported_types.count() > 0) {
             resolveReferencedTypes(&self.declarations, &self.non_exported_types);
         }
 
-        // Post-process: remove implementation signatures of overloaded functions
-        if (self.func_body_indices.count() > 0) {
+        // Post-process: remove implementation signatures of overloaded functions (only if map was used)
+        if (self.func_body_indices_inited and self.func_body_indices.count() > 0) {
             removeOverloadImplementations(self);
         }
 
@@ -685,6 +796,7 @@ pub const Scanner = struct {
 
 /// Check if name appears as a whole word in text (fast indexOf + boundary check)
 pub fn isWordInText(name: []const u8, text: []const u8) bool {
+    if (name.len == 0 or name.len > text.len) return false;
     var search_from: usize = 0;
     while (search_from < text.len) {
         const idx = ch.indexOf(text, name, search_from) orelse return false;
@@ -693,29 +805,14 @@ pub fn isWordInText(name: []const u8, text: []const u8) bool {
         const before_ok = !ch.isIdentChar(before);
         const after_ok = !ch.isIdentChar(after);
         if (before_ok and after_ok) return true;
-        search_from = idx + 1;
+        // When the right boundary fails (we matched a longer identifier), skip past
+        // the entire identifier so we don't re-scan its prefix. When only the left
+        // boundary fails, advance just by 1 to allow overlapping self-similar names.
+        search_from = if (!after_ok) idx + name.len else idx + 1;
     }
     return false;
 }
 
-/// Extract all identifier words from text into a HashSet (single pass, O(n))
-fn extractWordsFromText(alloc: std.mem.Allocator, text: []const u8) std.StringHashMap(void) {
-    var words = std.StringHashMap(void).init(alloc);
-    var i: usize = 0;
-    while (i < text.len) {
-        const c = text[i];
-        if (ch.isIdentStart(c)) {
-            const start = i;
-            i += 1;
-            while (i < text.len and ch.isIdentChar(text[i])) i += 1;
-            words.put(text[start..i], {}) catch {};
-        } else {
-            i += 1;
-        }
-    }
-    return words;
-}
-
 /// Resolve non-exported types that are referenced by exported declarations
 fn resolveReferencedTypes(declarations: *std.array_list.Managed(Declaration), non_exported_types: *std.StringHashMap(Declaration)) void {
     var resolved = std.StringHashMap(void).init(declarations.allocator);
@@ -743,6 +840,15 @@ fn resolveReferencedTypes(declarations: *std.array_list.Managed(Declaration), no
     // Track how far we've extracted words — only process new text_parts each iteration
     var words_extracted_up_to: usize = 0;
 
+    // Move outside loop — reuse across iterations
+    var to_insert = std.array_list.Managed(Declaration).init(declarations.allocator);
+    to_insert.ensureTotalCapacity(non_exported_types.count()) catch {};
+    defer to_insert.deinit();
+
+    var merged = std.array_list.Managed(Declaration).init(declarations.allocator);
+    merged.ensureTotalCapacity(declarations.items.len + non_exported_types.count()) catch {};
+    defer merged.deinit();
+
     while (true) {
         // Incrementally extract words from only the NEW text parts
         for (text_parts.items[words_extracted_up_to..]) |part| {
@@ -760,9 +866,7 @@ fn resolveReferencedTypes(declarations: *std.array_list.Managed(Declaration), no
         }
         words_extracted_up_to = text_parts.items.len;
 
-        var to_insert = std.array_list.Managed(Declaration).init(declarations.allocator);
-        to_insert.ensureTotalCapacity(non_exported_types.count()) catch {};
-        defer to_insert.deinit();
+        to_insert.clearRetainingCapacity();
 
         var it = non_exported_types.iterator();
         while (it.next()) |entry| {
@@ -789,7 +893,7 @@ fn resolveReferencedTypes(declarations: *std.array_list.Managed(Declaration), no
         }.cmp);
 
         // Merge at correct source positions
-        var merged = std.array_list.Managed(Declaration).init(declarations.allocator);
+        merged.clearRetainingCapacity();
         merged.ensureTotalCapacity(declarations.items.len + to_insert.items.len) catch {};
         var ti: usize = 0;
         for (declarations.items) |d| {
@@ -806,7 +910,6 @@ fn resolveReferencedTypes(declarations: *std.array_list.Managed(Declaration), no
 
         declarations.clearRetainingCapacity();
         declarations.appendSlice(merged.items) catch {};
-        merged.deinit();
 
         // Add new texts to search
         for (to_insert.items) |d| {
@@ -818,59 +921,71 @@ fn resolveReferencedTypes(declarations: *std.array_list.Managed(Declaration), no
 }
 
 /// Remove implementation signatures of overloaded functions
+/// Remove implementation signatures of overloaded functions.
+/// Single-HashMap approach: count function names, then find and remove body-bearing
+/// implementations in one backward pass.
 fn removeOverloadImplementations(scanner: *Scanner) void {
-    // Count function names
-    var func_name_counts = std.StringHashMap(usize).init(scanner.allocator);
-    defer func_name_counts.deinit();
+    // Single HashMap: count function declarations by name. Track overloads inline
+    // so we don't need a second iteration pass to count them.
+    var func_counts = std.StringHashMap(usize).init(scanner.allocator);
+    defer func_counts.deinit();
+    var overload_count: usize = 0;
 
     for (scanner.declarations.items) |d| {
         if (d.kind == .function_decl) {
-            const entry = func_name_counts.getOrPut(d.name) catch continue;
+            const entry = func_counts.getOrPut(d.name) catch continue;
             if (!entry.found_existing) {
                 entry.value_ptr.* = 1;
             } else {
                 entry.value_ptr.* += 1;
+                if (entry.value_ptr.* == 2) overload_count += 1;
             }
         }
     }
-
-    // Find overloaded names (count > 1)
-    var overloaded = std.StringHashMap(void).init(scanner.allocator);
-    defer overloaded.deinit();
-    var it = func_name_counts.iterator();
-    while (it.next()) |entry| {
-        if (entry.value_ptr.* > 1) {
-            overloaded.put(entry.key_ptr.*, {}) catch {};
-        }
-    }
-
-    if (overloaded.count() == 0) return;
-
-    // Find last body-bearing index for each overloaded name and remove them
-    var to_remove = std.AutoHashMap(usize, void).init(scanner.allocator);
-    defer to_remove.deinit();
-
-    var oit = overloaded.iterator();
-    while (oit.next()) |entry| {
-        const name = entry.key_ptr.*;
-        // Walk backwards
-        var i: usize = scanner.declarations.items.len;
-        while (i > 0) {
-            i -= 1;
-            const d = scanner.declarations.items[i];
-            if (d.kind == .function_decl and std.mem.eql(u8, d.name, name) and scanner.func_body_indices.contains(i)) {
-                to_remove.put(i, {}) catch {};
-                break;
-            }
+    if (overload_count == 0) return;
+
+    // Single backward pass: for each overloaded name (count > 1),
+    // find and mark the last body-bearing declaration for removal.
+    // Use a bitset (array of bools) instead of a HashMap for to_remove.
+    const len = scanner.declarations.items.len;
+    const remove_flags = scanner.allocator.alloc(bool, len) catch return;
+    defer scanner.allocator.free(remove_flags);
+    @memset(remove_flags, false);
+
+    var found_count: usize = 0;
+
+    // Walk backward: for each overloaded function, mark its last body-bearing impl
+    var names_found = std.StringHashMap(void).init(scanner.allocator);
+    defer names_found.deinit();
+
+    var i: usize = len;
+    while (i > 0) {
+        i -= 1;
+        const d = scanner.declarations.items[i];
+        if (d.kind != .function_decl) continue;
+
+        // Check if this function is overloaded
+        const count_entry = func_counts.get(d.name) orelse continue;
+        if (count_entry <= 1) continue;
+
+        // Already found the impl for this name?
+        if (names_found.contains(d.name)) continue;
+
+        // Is this the implementation (has body)?
+        if (scanner.func_body_indices.contains(i)) {
+            remove_flags[i] = true;
+            found_count += 1;
+            names_found.put(d.name, {}) catch {};
+            if (found_count == overload_count) break; // All found
         }
     }
 
-    if (to_remove.count() == 0) return;
+    if (found_count == 0) return;
 
-    // Filter in single pass — O(n) instead of O(k*n)
+    // Filter in single pass
     var write: usize = 0;
-    for (scanner.declarations.items, 0..) |d, i| {
-        if (!to_remove.contains(i)) {
+    for (scanner.declarations.items, 0..) |d, fi| {
+        if (!remove_flags[fi]) {
             scanner.declarations.items[write] = d;
             write += 1;
         }
@@ -906,3 +1021,75 @@ test "scanner skipWhitespaceAndComments" {
     s.skipWhitespaceAndComments();
     try std.testing.expectEqualStrings("hello", s.source[s.pos .. s.pos + 5]);
 }
+
+// --- Tests added for performance/fix patches ---
+
+// isWordInText: when a hit fails the right-boundary check (e.g. searching for
+// "Foo" inside "FooBar"), the new implementation skips past the matched
+// substring instead of advancing one byte. These cases verify both the
+// short-circuit and the normal advancement paths still report correctly.
+test "isWordInText skips past failing right boundary" {
+    // Multiple longer-ident matches followed by a real match — the optimization
+    // must not skip the legitimate later match.
+    try std.testing.expect(isWordInText("Foo", "FooBar FooBaz Foo;"));
+    try std.testing.expect(!isWordInText("Foo", "FooBar FooBaz FooQuux"));
+    // Left-boundary failure at an early position (`xxaa`) must advance by 1
+    // and still find a real later match (`aa`) — exercises the advance-by-1
+    // branch. (Without the trailing standalone `aa`, this string contains
+    // no whole-word match: the only "aa" is glued to `xx`.)
+    try std.testing.expect(isWordInText("aa", "xxaa aa"));
+    // Empty needle is rejected.
+    try std.testing.expect(!isWordInText("", "anything"));
+    // Needle longer than text is rejected without indexOf.
+    try std.testing.expect(!isWordInText("longer", "no"));
+}
+
+// peekAfterKeyword: the first-byte fast-fail must short-circuit cleanly when
+// the next ident byte after `word1` doesn't even match `word2[0]`.
+test "peekAfterKeyword first-byte fast-fail" {
+    var s1 = Scanner.init(std.testing.allocator, "async function foo", true, false);
+    defer s1.deinit();
+    try std.testing.expect(s1.peekAfterKeyword("async", "function"));
+
+    var s2 = Scanner.init(std.testing.allocator, "async const foo", true, false);
+    defer s2.deinit();
+    try std.testing.expect(!s2.peekAfterKeyword("async", "function"));
+
+    // Empty word2 must be rejected (avoids matching anything spuriously).
+    var s3 = Scanner.init(std.testing.allocator, "async ", true, false);
+    defer s3.deinit();
+    try std.testing.expect(!s3.peekAfterKeyword("async", ""));
+}
+
+// isRegexStart: the first-char dispatch must still recognize every keyword
+// previously checked, and must reject identifiers that share a prefix with
+// one of the regex-introducing keywords.
+test "isRegexStart keyword dispatch covers all branches" {
+    // Helper: position the scanner at the slash and ask isRegexStart.
+    const cases = [_]struct { src: []const u8, expected: bool }{
+        .{ .src = "return /a/", .expected = true },
+        .{ .src = "typeof /a/", .expected = true },
+        .{ .src = "throw /a/", .expected = true },
+        .{ .src = "void /a/", .expected = true },
+        .{ .src = "delete /a/", .expected = true },
+        .{ .src = "new /a/", .expected = true },
+        .{ .src = "in /a/", .expected = true },
+        .{ .src = "instanceof /a/", .expected = true },
+        .{ .src = "of /a/", .expected = true },
+        .{ .src = "case /a/", .expected = true },
+        .{ .src = "yield /a/", .expected = true },
+        .{ .src = "await /a/", .expected = true },
+        // Look-alike identifiers must NOT trigger regex-mode.
+        .{ .src = "myReturn /a/", .expected = false }, // "myReturn" is not "return"
+        .{ .src = "thrower /a/", .expected = false },
+        .{ .src = "x /a/", .expected = false }, // 1-char ident, length-bucket reject
+    };
+    for (cases) |c| {
+        var s = Scanner.init(std.testing.allocator, c.src, true, false);
+        defer s.deinit();
+        // Walk to the slash position
+        const slash_idx = std.mem.indexOfScalar(u8, c.src, '/').?;
+        s.pos = slash_idx;
+        try std.testing.expectEqual(c.expected, s.isRegexStart());
+    }
+}