diff --git a/src-tools/lib/unicode/chars.js b/src-tools/lib/unicode/chars.js
new file mode 100644
index 00000000..a5be6f69
--- /dev/null
+++ b/src-tools/lib/unicode/chars.js
@@ -0,0 +1,155 @@
+/*
+ *  Select a set of Unicode characters (based on included/excluded categories)
+ *  and write out a compact bitstream for matching a character against
+ *  the set at runtime.  This is for the slow path, where we're especially
+ *  concerned with compactness.
+ *
+ *  Unicode categories (such as 'Z') and codepoints (such as 0x200d) can be
+ *  used.  Two pseudo-categories are also available for exclusion only: 'ASCII'
+ *  and 'NONBMP'.  'ASCII' category excludes ASCII codepoints which is useful
+ *  because C code typically contains an ASCII fast path so ASCII characters
+ *  don't need to be considered in the Unicode tables.  'NONBMP' excludes
+ *  codepoints above U+FFFF which is useful because such codepoints don't need
+ *  to be supported in standard ECMAScript.
+ */
+
+'use strict';
+
+const { createBareObject } = require('../util/bare');
+const { BitEncoder } = require('../util/bitencoder');
+const { assert } = require('../util/assert');
+
+// Filter cpMap by first including all categories listed in 'includeList' and
+// then excluding anything in 'excludeList'.
+function filterCpMap(cpMap, includeList, excludeList) {
+    var filterAscii = excludeList.indexOf('ASCII') >= 0;
+    var filterNonBmp = excludeList.indexOf('NONBMP') >= 0;
+    var includeCatMap = createBareObject({});
+    var excludeCatMap = createBareObject({});
+    var includeCpMap = createBareObject({});
+    var excludeCpMap = createBareObject({});
+    var filteredCpMap;
+
+    // Helper lookups to speed up processing.
+    includeList.forEach((cat) => {
+        if (typeof cat === 'number') {
+            includeCpMap[cat] = true;
+        } else if (typeof cat === 'string') {
+            includeCatMap[cat] = true;
+        } else {
+            throw new TypeError('invalid includeList entry: ' + cat);
+        }
+    });
+    excludeList.forEach((cat) => {
+        if (typeof cat === 'number') {
+            excludeCpMap[cat] = true;
+        } else if (typeof cat === 'string') {
+            excludeCatMap[cat] = true;  // includes ASCII and NONBMP, does not matter
+        } else {
+            throw new TypeError('invalid excludeList entry: ' + cat);
+        }
+    });
+
+    // Filter codepoint map according to our criteria.
+    filteredCpMap = cpMap.filter((ent) => {
+        if (!ent) {
+            return false;
+        }
+        let cp = ent.cp;
+        let gc = ent.gc;
+        if ((filterAscii && cp <= 0x7f) || (filterNonBmp && cp >= 0x10000)) {
+            return false;
+        }
+        if ((includeCatMap[gc] || includeCpMap[cp]) &&
+            !(excludeCatMap[gc] || excludeCpMap[cp])) {
+            // Included in one or more categories/codepoints, and not
+            // excluded by any categories/codepoints.
+            return true;
+        }
+        return false;
+    });
+
+    return filteredCpMap;
+}
+exports.filterCpMap = filterCpMap;
+
+// Pack match ranges into a varint encoding.  For previous unused encoding
+// variants, see old Python tooling.
+function generateMatchTable3(ranges) {
+    var be = new BitEncoder();
+    var freq = [];  // informative
+    while (freq.length < 0x110000) {
+        freq.push(0);
+    }
+
+    function encCustom(x) {
+        freq[x]++;
+
+        if (x <= 0x0e) {
+            // 4-bit encoding
+            be.bits(x, 4);
+            return;
+        }
+        x -= 0x0e + 1;
+
+        if (x <= 0xfd) {
+            // 12-bit encoding
+            be.bits(0x0f, 4);
+            be.bits(x, 8);
+            return;
+        }
+        x -= 0xfd + 1;
+
+        if (x <= 0xfff) {
+            // 24-bit encoding
+            be.bits(0x0f, 4);
+            be.bits(0xfe, 8);
+            be.bits(x, 12);
+            return;
+        }
+        x -= 0xfff + 1;
+
+        // 36-bit encoding
+        be.bits(0x0f, 4);
+        be.bits(0xff, 8);
+        be.bits(x, 24);
+    }
+
+    function encVaruint(x) {
+        be.varuint(x);
+    }
+    void encVaruint;
+
+    var enc = encCustom;
+
+    var prevRangeEnd = 0;
+    for (let i = 0; i < ranges.length; i++) {
+        let rangeStart = ranges[i][0];
+        let rangeEnd = ranges[i][1];
+        let r1 = rangeStart - prevRangeEnd;  // 1 or above (no unjoined ranges)
+        assert(r1 >= 1);
+        let r2 = rangeEnd - rangeStart;  // 0 or above
+        assert(r2 >= 0);
+
+        // r1 is >= 1, so r1 == 0 is used as an end marker.  Encoding an
+        // explicit count and (r1 - 1) here improves total output size
+        // by about 30 bytes.
+        //
+        // Encoding using BitEncoder varuint is more efficient and shares
+        // code so maybe switch to that.
+
+        enc(r1);
+        enc(r2);
+        prevRangeEnd = rangeEnd;
+    }
+
+    // End marker (r1 can never be 0).
+    enc(0);
+
+    return {
+        data: be.getBytes(),
+        nbits: be.getNumBits(),
+        freq: freq
+    };
+}
+exports.generateMatchTable3 = generateMatchTable3;