nodemailer 10.1.4 → 10.1.6

package/dist/cjs/addressparser/index.js

@@ -1,5 +1,83 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
+/* eslint-disable no-bitwise -- the kind of a token is a set of bit flags packed into one slot of a typed array */
+/** The token is an operator, a text run otherwise */
+const TOKEN_OPERATOR = 1;
+/** The next character continues the same run without whitespace in between */
+const TOKEN_NO_BREAK = 2;
+/** The text run holds characters its value leaves out or changes, so it is no plain slice */
+const TOKEN_DIRTY = 4;
+/** A backslash escapes the next character in the text run, inside a quoted string or a comment */
+const TOKEN_ESCAPES = 8;
+/** The text run opens on a character an escaping backslash in front of it has escaped */
+const TOKEN_STARTS_ESCAPED = 16;
+/**
+ * The tokens of an address field, kept as offsets into it.
+ *
+ * Every token used to be an object of its own, an operator for every ',' and a value string
+ * for every run of text, and a list of one letter names, 'a,a,a,...', took some 65 times its
+ * size in tokens before any address was read. A token is now three numbers in one typed
+ * array, its kind and the range of the field it covers, and its value is sliced off the field
+ * when it is read. The few text runs whose value differs from their slice, as they hold an
+ * escaping backslash, a line break or a control character, are rebuilt once and kept aside.
+ */
+class TokenList {
+    constructor(str) {
+        this.str = str;
+        this.slots = new Int32Array(48);
+        this.length = 0;
+        this.cleaned = new Map();
+    }
+    /**
+     * Appends a token
+     *
+     * @param kind Kind and flags of the token
+     * @param start Offset of its first character
+     * @param end Offset just past its last character
+     * @return Index of the token
+     */
+    push(kind, start, end) {
+        if ((this.length + 1) * 3 > this.slots.length) {
+            const grown = new Int32Array(this.slots.length * 2);
+            grown.set(this.slots);
+            this.slots = grown;
+        }
+        const at = this.length * 3;
+        this.slots[at] = kind;
+        this.slots[at + 1] = start;
+        this.slots[at + 2] = end;
+        return this.length++;
+    }
+    /**
+     * The character of an operator token, an empty string for a text run
+     */
+    operatorAt(index) {
+        const at = index * 3;
+        return this.slots[at] & TOKEN_OPERATOR ? this.str.charAt(this.slots[at + 1]) : '';
+    }
+    /**
+     * Whether a token is an operator
+     */
+    isOperator(index) {
+        return (this.slots[index * 3] & TOKEN_OPERATOR) !== 0;
+    }
+    /**
+     * Whether the next character continues the run of a token without whitespace in between
+     */
+    noBreak(index) {
+        return (this.slots[index * 3] & TOKEN_NO_BREAK) !== 0;
+    }
+    /**
+     * Value of a token
+     */
+    value(index) {
+        const cleaned = this.cleaned.get(index);
+        if (cleaned !== undefined) {
+            return cleaned;
+        }
+        return this.str.slice(this.slots[index * 3 + 1], this.slots[index * 3 + 2]);
+    }
+}
 /**
  * A domain literal that ends an address. dtext includes '@' (RFC 5322 3.4.1), so the '@'
  * in front of the literal is the one that splits the domain off, not the last one: reading
@@ -454,20 +532,19 @@ function _stripAddressComments(address) {
     return { address: result.trim(), comments: comments.filter(text => text) };
 }
 /**
- * Converts tokens for a single address into an address object
+ * Walks the tokens of one entry of an address list and sorts them into the runs of display
+ * text, angle-addr and comment they belong to
  *
- * @param tokens Tokens object
- * @param depth Current recursion depth for nested group protection
- * @return Address object
+ * @param tokens Tokens of the whole address list
+ * @param from Index of the first token of the entry
+ * @param to Index just past the last token of the entry
+ * @return The collected runs, and whether the display text came out of a comment
  */
-function _handleAddress(tokens, depth) {
-    let isGroup = false;
+function _collectAddressParts(tokens, from, to) {
     let state = 'text';
-    const addresses = [];
     const data = {
         address: [],
         comment: [],
-        group: [],
         text: [],
         textWasQuoted: [],
         quotedLocalAt: []
@@ -481,13 +558,15 @@ function _handleAddress(tokens, depth) {
     // every token, which is quadratic over an address built from many comment-joined atoms
     // (GHSA-prgh-xp8r-p3m5). A run only ever grows by the token appended below, so the
     // character is carried along instead of re-read.
-    const lastChars = { address: '', comment: '', group: '', text: '' };
+    const lastChars = { address: '', comment: '', text: '' };
     // Filter out <addresses>, (comments) and regular text
-    for (let i = 0, len = tokens.length; i < len; i++) {
-        const token = tokens[i];
-        const prevToken = i ? tokens[i - 1] : null;
-        if (token.type === 'operator') {
-            switch (token.value) {
+    for (let i = from; i < to; i++) {
+        const value = tokens.value(i);
+        // the operator in front of this token, if there is one, and whether it runs on into it
+        const prevOperator = i > from ? tokens.operatorAt(i - 1) : '';
+        const prevNoBreak = i > from && tokens.noBreak(i - 1);
+        if (tokens.isOperator(i)) {
+            switch (value) {
                 case '<':
                     state = 'address';
                     insideQuotes = false;
@@ -496,11 +575,6 @@ function _handleAddress(tokens, depth) {
                     state = 'comment';
                     insideQuotes = false;
                     break;
-                case ':':
-                    state = 'group';
-                    isGroup = true;
-                    insideQuotes = false;
-                    break;
                 case '"':
                     insideQuotes = !insideQuotes;
                     state = 'text';
@@ -511,7 +585,7 @@ function _handleAddress(tokens, depth) {
                     break;
             }
         }
-        else if (token.value) {
+        else {
             // An empty quoted string is dropped by the tokenizer, leaving no text token of its
             // own for textWasQuoted to be recorded on, so the pair of quote operators right in
             // front of this token is all that is left of it and the run it opens carries the
@@ -519,44 +593,34 @@ function _handleAddress(tokens, depth) {
             // the quotes never go back on, and the value is no longer an addr-spec a trailing
             // comment can be peeled off of. It only ever opens a run: a run that already holds
             // material collected outside the quotes is not a quoted string, whatever follows it
-            const prevPrevToken = i > 1 ? tokens[i - 2] : null;
-            const opensAfterEmptyQuotedString = prevToken?.type === 'operator' &&
-                prevToken.value === '"' &&
-                !!prevToken.noBreak &&
-                prevPrevToken?.type === 'operator' &&
-                prevPrevToken.value === '"';
+            const opensAfterEmptyQuotedString = prevOperator === '"' && prevNoBreak && i > from + 1 && tokens.operatorAt(i - 2) === '"';
             // A comment is folding whitespace. It may sit inside an addr-spec, on either side
             // of the '@', but it cannot join two atoms into one: gluing across it would read
             // 'user@example.com(x)evil.com' as the single domain 'example.comevil.com' and
             // deliver to a domain the sender never named.
             const parts = data[state];
-            const joins = prevToken &&
-                prevToken.noBreak &&
-                parts.length &&
-                (prevToken.value !== ')' || lastChars[state] === '@' || token.value.charAt(0) === '@');
+            const joins = prevNoBreak && parts.length && (prevOperator !== ')' || lastChars[state] === '@' || value.charAt(0) === '@');
             if (joins) {
                 const last = data[state].length - 1;
                 if (state === 'text' && !insideQuotes) {
-                    if (lastTextPureQuoted && token.value.charAt(0) === '@') {
+                    if (lastTextPureQuoted && value.charAt(0) === '@') {
                         data.quotedLocalAt[last] = data.text[last].length;
                     }
                     lastTextPureQuoted = false;
                 }
-                data[state][last] += token.value;
-                if (token.value) {
-                    lastChars[state] = token.value.charAt(token.value.length - 1);
-                }
+                data[state][last] += value;
+                lastChars[state] = value.charAt(value.length - 1);
                 if (state === 'text' && insideQuotes) {
                     data.textWasQuoted[last] = true;
                 }
             }
             else {
-                data[state].push(token.value);
-                lastChars[state] = token.value.charAt(token.value.length - 1);
+                data[state].push(value);
+                lastChars[state] = value.charAt(value.length - 1);
                 if (state === 'text') {
                     data.textWasQuoted.push(insideQuotes || opensAfterEmptyQuotedString);
                     // an empty quoted string leaves no text of its own, so the '@' opens the run
-                    data.quotedLocalAt.push(opensAfterEmptyQuotedString && token.value.charAt(0) === '@' ? 0 : -1);
+                    data.quotedLocalAt.push(opensAfterEmptyQuotedString && value.charAt(0) === '@' ? 0 : -1);
                     lastTextPureQuoted = insideQuotes;
                 }
             }
@@ -571,154 +635,158 @@ function _handleAddress(tokens, depth) {
         data.comment = [];
         textFromComment = true;
     }
-    if (isGroup) {
-        // http://tools.ietf.org/html/rfc2822#appendix-A.1.3
-        data.text = data.text.join(' ');
-        // Parse group members, but flatten any nested groups (RFC 5322 doesn't allow nesting)
-        let groupMembers = [];
-        if (data.group.length) {
-            const parsedGroup = _parseAddressList(data.group.join(','), depth + 1);
-            parsedGroup.forEach(member => {
-                if (member.group) {
-                    groupMembers = groupMembers.concat(member.group);
-                }
-                else {
-                    groupMembers.push(member);
-                }
-            });
+    return { data, textFromComment };
+}
+/**
+ * Name of a group, from the tokens in front of its ':'. A comment names it when no text does,
+ * as it names a mailbox.
+ *
+ * @param tokens Tokens of the whole address list
+ * @param from Index of the first token in front of the ':'
+ * @param to Index of the ':'
+ * @return Name of the group
+ */
+function _groupName(tokens, from, to) {
+    // http://tools.ietf.org/html/rfc2822#appendix-A.1.3
+    return _collectAddressParts(tokens, from, to).data.text.join(' ');
+}
+/**
+ * Converts tokens for a single address into an address object
+ *
+ * @param tokens Tokens of the whole address list
+ * @param from Index of the first token of this address
+ * @param to Index just past the last token of this address
+ * @return Address object
+ */
+function _handleAddress(tokens, from, to) {
+    const collected = _collectAddressParts(tokens, from, to);
+    const data = collected.data;
+    let textFromComment = collected.textFromComment;
+    // Comments come out of the angle-addr before anything asks whether one was found, so
+    // that brackets holding nothing but a comment read the same as empty ones
+    const addressComments = [];
+    const addressParts = [];
+    for (const part of data.address) {
+        const stripped = _stripAddressComments(part);
+        for (const comment of stripped.comments) {
+            addressComments.push(comment);
+        }
+        if (stripped.address) {
+            addressParts.push(stripped.address);
         }
-        addresses.push({
-            name: data.text || '',
-            group: groupMembers
-        });
     }
-    else {
-        // Comments come out of the angle-addr before anything asks whether one was found, so
-        // that brackets holding nothing but a comment read the same as empty ones
-        const addressComments = [];
-        const addressParts = [];
-        for (const part of data.address) {
-            const stripped = _stripAddressComments(part);
-            for (const comment of stripped.comments) {
-                addressComments.push(comment);
-            }
-            if (stripped.address) {
-                addressParts.push(stripped.address);
+    data.address = addressParts;
+    // If no address was found, try to detect one from regular text
+    if (!data.address.length && data.text.length && !textFromComment) {
+        // Whitespace beside the '@' is CFWS, so 'user @ good.com' is read as one run. Both
+        // tiers read the collapsed runs, and only a run an address comes out of is replaced
+        // by its collapsed form, the display text keeps its spacing
+        const candidates = data.text.map((part) => _collapseAtSpacing(part));
+        // The first addr-spec in the text is the mailbox, as it is for a conformant parser
+        // reading as far as it can. Taking the last one let a comment part two of them and
+        // pick the second, 'user@good.com (x) evil@evil.com' went to evil.com, while the
+        // same two without the comment, one run, went to good.com
+        for (let i = 0; i < data.text.length; i++) {
+            // Security: Do not extract email addresses from quoted strings.
+            // RFC 5321 allows @ inside quoted local-parts like "user@domain"@example.com.
+            // Extracting emails from quoted text leads to misrouting vulnerabilities.
+            // A run that is a quoted local part with its '@' and domain is an addr-spec of
+            // its own, with the quoting put back on, and takes its turn like any other
+            const local = data.quotedLocalAt[i];
+            const quotedLocal = data.textWasQuoted[i] && local >= 0 && QUOTED_LOCAL_DOMAIN.test(data.text[i].slice(local))
+                ? _quoteLocalPart(data.text[i])
+                : '';
+            if (quotedLocal || (!data.textWasQuoted[i] && _isAddrSpec(candidates[i]))) {
+                data.address = [quotedLocal || candidates[i]];
+                data.text.splice(i, 1);
+                data.textWasQuoted.splice(i, 1);
+                data.quotedLocalAt.splice(i, 1);
+                break;
             }
         }
-        data.address = addressParts;
-        // If no address was found, try to detect one from regular text
-        if (!data.address.length && data.text.length && !textFromComment) {
-            // Whitespace beside the '@' is CFWS, so 'user @ good.com' is read as one run. Both
-            // tiers read the collapsed runs, and only a run an address comes out of is replaced
-            // by its collapsed form, the display text keeps its spacing
-            const candidates = data.text.map((part) => _collapseAtSpacing(part));
-            // The first addr-spec in the text is the mailbox, as it is for a conformant parser
-            // reading as far as it can. Taking the last one let a comment part two of them and
-            // pick the second, 'user@good.com (x) evil@evil.com' went to evil.com, while the
-            // same two without the comment, one run, went to good.com
+        // Try a looser regex match if strict match found nothing
+        if (!data.address.length) {
+            let extracted = false;
             for (let i = 0; i < data.text.length; i++) {
-                // Security: Do not extract email addresses from quoted strings.
-                // RFC 5321 allows @ inside quoted local-parts like "user@domain"@example.com.
-                // Extracting emails from quoted text leads to misrouting vulnerabilities.
-                // A run that is a quoted local part with its '@' and domain is an addr-spec of
-                // its own, with the quoting put back on, and takes its turn like any other
-                const local = data.quotedLocalAt[i];
-                const quotedLocal = data.textWasQuoted[i] && local >= 0 && QUOTED_LOCAL_DOMAIN.test(data.text[i].slice(local))
-                    ? _quoteLocalPart(data.text[i])
-                    : '';
-                if (quotedLocal || (!data.textWasQuoted[i] && _isAddrSpec(candidates[i]))) {
-                    data.address = [quotedLocal || candidates[i]];
-                    data.text.splice(i, 1);
-                    data.textWasQuoted.splice(i, 1);
-                    data.quotedLocalAt.splice(i, 1);
-                    break;
-                }
-            }
-            // Try a looser regex match if strict match found nothing
-            if (!data.address.length) {
-                let extracted = false;
-                for (let i = 0; i < data.text.length; i++) {
-                    // Security: Do not extract email addresses from quoted strings
-                    if (!data.textWasQuoted[i]) {
-                        // the strict tier found nothing, so it spliced nothing and the indexes still line up
-                        const part = candidates[i];
-                        let remainder = data.text[i];
-                        const at = _looseAddressStart(part);
-                        if (at >= 0) {
-                            LOOSE_TEXT_ADDR.lastIndex = at;
-                            const match = LOOSE_TEXT_ADDR.exec(part);
-                            if (match) {
-                                data.address = [match[0].trim()];
-                                extracted = true;
-                                remainder = part.slice(0, at) + ' ' + part.slice(at + match[0].length);
-                            }
-                        }
-                        data.text[i] = remainder.trim();
-                        if (extracted) {
-                            break;
+                // Security: Do not extract email addresses from quoted strings
+                if (!data.textWasQuoted[i]) {
+                    // the strict tier found nothing, so it spliced nothing and the indexes still line up
+                    const part = candidates[i];
+                    let remainder = data.text[i];
+                    const at = _looseAddressStart(part);
+                    if (at >= 0) {
+                        LOOSE_TEXT_ADDR.lastIndex = at;
+                        const match = LOOSE_TEXT_ADDR.exec(part);
+                        if (match) {
+                            data.address = [match[0].trim()];
+                            extracted = true;
+                            remainder = part.slice(0, at) + ' ' + part.slice(at + match[0].length);
                         }
                     }
+                    data.text[i] = remainder.trim();
+                    if (extracted) {
+                        break;
+                    }
                 }
             }
         }
-        // If there's still no text but a comment exists, replace the two
-        if (!data.text.length && data.comment.length) {
-            data.text = data.comment;
-            data.comment = [];
-        }
-        // Keep only the first address occurrence, push others to regular text
-        if (data.address.length > 1) {
-            data.text = data.text.concat(data.address.splice(1));
-        }
-        // An address is only taken from unquoted text, so anything left in the text at this
-        // point that still has to serve as the address carries its quoting in this flag
-        const addressFromQuotedText = !data.address.length && data.textWasQuoted.some(wasQuoted => wasQuoted);
-        // Join values with spaces
-        data.text = data.text.join(' ');
-        data.address = data.address.join(' ');
-        if (addressFromQuotedText && data.text.indexOf('@') >= 0) {
-            // A quoted run with no '@' in it is a display name and nothing else, so it stays
-            // where it is. Moving it over anyway made a comment the mailbox was read from:
-            // '"Display Name" <(a comment)>' handed on 'Display Name' as the address with the
-            // comment as the name, and a quoted name holding a ',' or a ';' became an address
-            // that reads as two recipients once a consumer writes it back into a header.
-            //
-            // The mailbox is still sitting in the text, so it moves over here and is quoted
-            // before the recovery below rather than after it. Anything else the text holds
-            // came along with it: a comment ends the domain but leaves the atoms behind it in
-            // the same text, and '"user"@example.com(x)evil.com' was handed on as the address
-            // 'user@example.com evil.com', a second domain riding into the envelope recipient
-            // on a value that is no addr-spec at all (GHSA-g57g-f23g-4646). Putting the quotes
-            // back first is what lets the recovery tell the whitespace an addr-spec may carry
-            // from the wreckage trailing one, as only a quoted local part may hold whitespace
-            // The quotes are gone at this point, so the '@' the domain splits off at is the
-            // last one, not the single one _collapseAtSpacing insists on
-            data.address = _quoteLocalPart(_joinAtSpacing(data.text, _domainAt(data.text)));
-            data.text = '';
-        }
-        _recoverAddrSpec(data);
-        // a comment names the mailbox only when nothing else does, as one outside the brackets
-        if (!data.text && addressComments.length) {
-            data.text = addressComments.join(' ');
-            textFromComment = true;
+    }
+    // If there's still no text but a comment exists, replace the two
+    if (!data.text.length && data.comment.length) {
+        data.text = data.comment;
+        data.comment = [];
+    }
+    // Keep only the first address occurrence, push others to regular text
+    if (data.address.length > 1) {
+        data.text = data.text.concat(data.address.splice(1));
+    }
+    // An address is only taken from unquoted text, so anything left in the text at this
+    // point that still has to serve as the address carries its quoting in this flag
+    const addressFromQuotedText = !data.address.length && data.textWasQuoted.some(wasQuoted => wasQuoted);
+    // Join values with spaces
+    data.text = data.text.join(' ');
+    data.address = data.address.join(' ');
+    if (addressFromQuotedText && data.text.indexOf('@') >= 0) {
+        // A quoted run with no '@' in it is a display name and nothing else, so it stays
+        // where it is. Moving it over anyway made a comment the mailbox was read from:
+        // '"Display Name" <(a comment)>' handed on 'Display Name' as the address with the
+        // comment as the name, and a quoted name holding a ',' or a ';' became an address
+        // that reads as two recipients once a consumer writes it back into a header.
+        //
+        // The mailbox is still sitting in the text, so it moves over here and is quoted
+        // before the recovery below rather than after it. Anything else the text holds
+        // came along with it: a comment ends the domain but leaves the atoms behind it in
+        // the same text, and '"user"@example.com(x)evil.com' was handed on as the address
+        // 'user@example.com evil.com', a second domain riding into the envelope recipient
+        // on a value that is no addr-spec at all (GHSA-g57g-f23g-4646). Putting the quotes
+        // back first is what lets the recovery tell the whitespace an addr-spec may carry
+        // from the wreckage trailing one, as only a quoted local part may hold whitespace
+        // The quotes are gone at this point, so the '@' the domain splits off at is the
+        // last one, not the single one _collapseAtSpacing insists on
+        data.address = _quoteLocalPart(_joinAtSpacing(data.text, _domainAt(data.text)));
+        data.text = '';
+    }
+    _recoverAddrSpec(data);
+    // a comment names the mailbox only when nothing else does, as one outside the brackets
+    if (!data.text && addressComments.length) {
+        data.text = addressComments.join(' ');
+        textFromComment = true;
+    }
+    const address = {
+        address: data.address || (textFromComment ? '' : data.text) || '',
+        name: data.text || data.address || ''
+    };
+    if (address.address === address.name) {
+        if (/@/.test(address.address || '')) {
+            address.name = '';
         }
-        const address = {
-            address: data.address || (textFromComment ? '' : data.text) || '',
-            name: data.text || data.address || ''
-        };
-        if (address.address === address.name) {
-            if (/@/.test(address.address || '')) {
-                address.name = '';
-            }
-            else {
-                address.address = '';
-            }
+        else {
+            address.address = '';
         }
-        addresses.push(address);
     }
-    return addresses;
+    return address;
 }
+/* eslint-disable no-bitwise -- see TokenList */
 /**
  * Creates a Tokenizer object for tokenizing address field strings
  *
@@ -730,15 +798,16 @@ class Tokenizer {
         this.str = (str || '').toString();
         this.operatorCurrent = '';
         this.operatorExpecting = '';
-        this.node = null;
+        this.node = -1;
         this.escaped = false;
         this.inDomainLiteral = false;
         this.afterAt = false;
+        this.inGroup = false;
         this.innerQuoted = false;
         this.innerComment = 0;
         this.innerLiteral = false;
         this.innerEscaped = false;
-        this.list = [];
+        this.list = new TokenList(this.str);
         /**
          * Operator tokens and which tokens are expected to end the sequence
          */
@@ -747,7 +816,9 @@ class Tokenizer {
             '(': ')',
             '<': '>',
             ',': '',
-            ':': ';',
+            // A group is read in the same pass as the rest of the list, so its ':' only marks
+            // where the name ends and its members are tokenized like any other address
+            ':': '',
             // Semicolons are not a legal delimiter per the RFC2822 grammar other
             // than for terminating a group, but they are also not valid for any
             // other use in this context.  Given that some mail clients have
@@ -763,27 +834,96 @@ class Tokenizer {
      * @return An array of operator|text tokens
      */
     tokenize() {
-        const list = [];
-        for (let i = 0, len = this.str.length; i < len; i++) {
-            const chr = this.str.charAt(i);
-            const nextChr = i < len - 1 ? this.str.charAt(i + 1) : null;
-            this.checkChar(chr, nextChr);
+        const str = this.str;
+        const len = str.length;
+        for (let i = 0; i < len; i++) {
+            this.checkChar(str.charAt(i), i);
         }
-        this.list.forEach(node => {
-            node.value = (node.value || '').toString().trim();
-            if (node.value) {
-                list.push(node);
+        this._closeText(len);
+        // Compacted in place: a text run is trimmed to its value and one left empty is
+        // dropped, and the value of a run that is no plain slice is rebuilt and kept aside
+        const list = this.list;
+        const slots = list.slots;
+        let kept = 0;
+        for (let i = 0, count = list.length; i < count; i++) {
+            const kind = slots[i * 3];
+            let start = slots[i * 3 + 1];
+            let end = slots[i * 3 + 2];
+            if (kind & TOKEN_DIRTY) {
+                const value = this._cleanValue(start, end, kind).trim();
+                if (!value) {
+                    continue;
+                }
+                list.cleaned.set(kept, value);
+            }
+            else if (!(kind & TOKEN_OPERATOR)) {
+                while (start < end && _isSpaceCode(str.charCodeAt(start))) {
+                    start++;
+                }
+                while (end > start && _isSpaceCode(str.charCodeAt(end - 1))) {
+                    end--;
+                }
+                if (start === end) {
+                    continue;
+                }
             }
-        });
+            slots[kept * 3] = kind;
+            slots[kept * 3 + 1] = start;
+            slots[kept * 3 + 2] = end;
+            kept++;
+        }
+        list.length = kept;
         return list;
     }
+    /**
+     * Ends the text run being collected, if there is one
+     *
+     * @param end Offset just past its last character
+     */
+    _closeText(end) {
+        if (this.node >= 0) {
+            this.list.slots[this.node * 3 + 2] = end;
+            this.node = -1;
+        }
+    }
+    /**
+     * Rebuilds the value of a text run that is no plain slice of the field, applying the
+     * rules checkChar collects characters by: an escaping backslash is dropped, a line break
+     * reads as a space and any other control character is left out
+     *
+     * @param start Offset of the first character of the run
+     * @param end Offset just past its last character
+     * @param kind Kind and flags of the run
+     * @return Value of the run
+     */
+    _cleanValue(start, end, kind) {
+        const str = this.str;
+        const escapes = (kind & TOKEN_ESCAPES) !== 0;
+        let escaped = (kind & TOKEN_STARTS_ESCAPED) !== 0;
+        let value = '';
+        for (let i = start; i < end; i++) {
+            const chr = str.charAt(i);
+            if (escapes && !escaped && chr === '\\') {
+                escaped = true;
+                continue;
+            }
+            escaped = false;
+            if (chr === '\n') {
+                value += ' ';
+            }
+            else if (chr.charCodeAt(0) >= 0x21 || chr === ' ' || chr === '\t') {
+                value += chr;
+            }
+        }
+        return value;
+    }
     /**
      * Checks if a character is an operator or text and acts accordingly
      *
      * @param chr Character from the address field
-     * @param nextChr Character following chr, null at the end of the field
+     * @param i Offset of the character in the field
      */
-    checkChar(chr, nextChr) {
+    checkChar(chr, i) {
         // Track RFC 5322 domain-literals ("[" *dtext "]"). The ":" of an IPv6
         // address-literal (user@[IPv6:2001:db8::1]) is dtext and must not be treated as
         // the group delimiter while inside the brackets. That is the only operator the
@@ -804,15 +944,23 @@ class Tokenizer {
                 this.inDomainLiteral = false;
             }
         }
-        // The closer of an angle-addr or a group is only read outside the quoted strings,
-        // comments and domain literals it holds. Looking for nothing but the closer let a '>'
-        // inside a quoted local part end the angle-addr, '<"a@evil.com>"@good.com>' was read
-        // as an address at evil.com, and a ';' inside a quoted display name or a comment end
-        // the group and release the rest of that text as recipients. The characters are still
-        // collected as they are, the angle-addr and the group content are read again later
+        // The closer of an angle-addr is only read outside the quoted strings, comments and
+        // domain literals it holds. Looking for nothing but the closer let a '>' inside a
+        // quoted local part end the angle-addr, '<"a@evil.com>"@good.com>' was read as an
+        // address at evil.com. The characters are still collected as they are, the angle-addr
+        // is read again later
         let nested = false;
-        if (this.operatorExpecting === '>' || this.operatorExpecting === ';') {
+        if (this.operatorExpecting === '>') {
             nested = this._trackInner(chr);
+            if (!nested && chr === ';' && this.inGroup) {
+                // The ';' of a group ends an angle-addr left open inside it, as it always did
+                // when a group was cut out of the field at its ';' and parsed on its own. Read
+                // as text instead, an unclosed '<' swallowed the end of the group and every
+                // recipient after it
+                this._closeText(i);
+                this.operatorExpecting = '';
+                this._resetInner();
+            }
         }
         else if (this.operatorExpecting === ')' && !this.escaped) {
             // Comments nest (RFC 5322 3.2.2). Ending one at the first ')' let the ')' of a
@@ -830,15 +978,9 @@ class Tokenizer {
             // ignore next condition blocks
         }
         else if (chr === this.operatorExpecting && !nested) {
-            this.node = {
-                type: 'operator',
-                value: chr
-            };
-            if (nextChr && ![' ', '\t', '\r', '\n', ',', ';'].includes(nextChr)) {
-                this.node.noBreak = true;
-            }
-            this.list.push(this.node);
-            this.node = null;
+            const nextChr = this.str.charAt(i + 1);
+            this._closeText(i);
+            this.list.push(nextChr && ![' ', '\t', '\r', '\n', ',', ';'].includes(nextChr) ? TOKEN_OPERATOR | TOKEN_NO_BREAK : TOKEN_OPERATOR, i, i + 1);
             if (chr !== ')') {
                 // a comment is folding whitespace, it does not part the "@" from the "["
                 this.afterAt = false;
@@ -849,12 +991,14 @@ class Tokenizer {
             return;
         }
         else if (!this.operatorExpecting && !(this.inDomainLiteral && chr === ':') && chr in this.operators) {
-            this.node = {
-                type: 'operator',
-                value: chr
-            };
-            this.list.push(this.node);
-            this.node = null;
+            this._closeText(i);
+            this.list.push(TOKEN_OPERATOR, i, i + 1);
+            if (chr === ':') {
+                this.inGroup = true;
+            }
+            else if (chr === ';') {
+                this.inGroup = false;
+            }
             if (chr !== '(') {
                 this.afterAt = false;
             }
@@ -869,39 +1013,42 @@ class Tokenizer {
             // only in quoted strings let an escaped parenthesis close the comment, which
             // released the rest of it, an address included, into the header
             this.escaped = true;
+            if (this.node >= 0) {
+                // the backslash is left out of the value
+                this.list.slots[this.node * 3] |= TOKEN_DIRTY;
+            }
             return;
         }
-        if (!this.node) {
-            this.node = {
-                type: 'text',
-                value: ''
-            };
-            this.list.push(this.node);
+        if (this.node < 0) {
+            this.node = this.list.push((this.operatorExpecting === '"' || this.operatorExpecting === ')' ? TOKEN_ESCAPES : 0) |
+                (this.escaped ? TOKEN_STARTS_ESCAPED : 0), i, i);
         }
         if (chr === '\n') {
             // Convert newlines to spaces. Carriage return is ignored as \r and \n usually
             // go together anyway and there already is a WS for \n. Lone \r means something is fishy.
             chr = ' ';
+            this.list.slots[this.node * 3] |= TOKEN_DIRTY;
         }
-        if (chr.charCodeAt(0) >= 0x21 || [' ', '\t'].includes(chr)) {
-            // skip command bytes
-            this.node.value += chr;
+        if (chr.charCodeAt(0) >= 0x21 || chr === ' ' || chr === '\t') {
+            // the character is part of the value, which is read off the field later
             // text inside a quoted string or a comment is not part of an addr-spec. Tracked
             // as it goes rather than read back off the value (see lastChars in _handleAddress)
             if (!this.operatorExpecting && chr !== ' ' && chr !== '\t') {
                 this.afterAt = chr === '@';
             }
         }
+        else {
+            // skip command bytes
+            this.list.slots[this.node * 3] |= TOKEN_DIRTY;
+        }
         this.escaped = false;
     }
     /**
-     * Tracks the quoted strings, comments and domain literals inside an angle-addr or a
-     * group, which the closer of either must not be read from. A backslash escapes the next
-     * character inside a quoted string and a comment. A domain literal is only tracked in an
-     * angle-addr, where it opens right after an "@" and ends at the "]". dtext holds the list
-     * separators as well, and inside the brackets they never split the list anyway. In a group
-     * they do, so a literal is not tracked there and an unclosed "[" can not swallow the
-     * recipients after it.
+     * Tracks the quoted strings, comments and domain literals inside an angle-addr, which its
+     * closer must not be read from. A backslash escapes the next character inside a quoted
+     * string and a comment. A domain literal opens right after an "@" and ends at the "]".
+     * dtext holds the list separators as well, and inside the brackets they never split the
+     * list anyway.
      *
      * @param chr Character from the address field
      * @return true while the character sits inside one of them, the one that closes it included
@@ -946,7 +1093,7 @@ class Tokenizer {
             this.innerComment = 1;
             return true;
         }
-        if (chr === '[' && this.afterAt && this.operatorExpecting === '>') {
+        if (chr === '[' && this.afterAt) {
             this.innerLiteral = true;
             return true;
         }
@@ -967,79 +1114,94 @@ class Tokenizer {
         this.innerEscaped = false;
     }
 }
+/* eslint-enable no-bitwise */
 /**
- * Maximum recursion depth for parsing nested groups.
- * RFC 5322 doesn't allow nested groups, so this is a safeguard against
- * malicious input that could cause stack overflow. Every level parses what is left of the
- * value once more, so the limit is also the factor a crafted header multiplies the parse by.
+ * Folds the fragments an unquoted display name holding a comma is split into back together.
+ *
+ * "Joe Foo, PhD <joe@example.com>" is split on the comma into
+ * [{name:"Joe Foo", address:""}, {name:"PhD", address:"joe@example.com"}]. A name-only entry
+ * followed by an entry with both name and address is recombined. Walked back to front so
+ * that a run of fragments folds into one entry in a single pass. Splicing each fragment out
+ * of the list instead would cost O(n^2).
+ *
+ * @param list Parsed entries, in order
+ * @return The entries with the fragments folded in
  */
-const MAX_NESTED_GROUP_DEPTH = 10;
+function _mergeNameFragments(list) {
+    const merged = [];
+    for (let i = list.length - 1; i >= 0; i--) {
+        const current = list[i];
+        const next = merged.length ? merged[merged.length - 1] : null;
+        if (next && current.address === '' && current.name && !current.group && next.address && next.name) {
+            next.name = current.name + ', ' + next.name;
+        }
+        else {
+            merged.push(current);
+        }
+    }
+    return merged.reverse();
+}
 /**
- * Parses an address list, recursing into the groups it holds.
+ * Parses an address list, groups included, in a single pass over its tokens.
  *
- * The depth is threaded through the calls rather than carried in an options object, so a
- * caller supplied one can not seed it and lift the recursion limit: `{ _depth: -1e9 }`
- * bought a billion levels of nesting and turned the guard below into the stack overflow it
- * is there to prevent.
+ * A group used to be cut out of the field at the ';' that ends it and parsed again as an
+ * address list of its own, recursively, so every level of nesting parsed what was left of
+ * the field once more and the depth had to be capped. RFC 5322 does not nest groups at all,
+ * and the nested ones were flattened anyway. The tokens are now walked once: a ':' opens a
+ * group, its members are read as they come, a ';' closes it. A ':' inside an open group
+ * stands for a group nested in it, whose name is dropped and whose members belong to the
+ * outer one, which is how they were flattened before. The first ';' ends the group, nested
+ * or not, as it did.
  *
  * @param str Address field
- * @param depth Current recursion depth for nested group protection
  * @return An array of address objects
  */
-function _parseAddressList(str, depth) {
-    // Prevent stack overflow from deeply nested groups (DoS protection)
-    if (depth > MAX_NESTED_GROUP_DEPTH) {
-        return [];
-    }
-    const tokenizer = new Tokenizer(str);
-    const tokens = tokenizer.tokenize();
-    const addresses = [];
-    let address = [];
+function _parseAddressList(str) {
+    const tokens = new Tokenizer(str).tokenize();
     const parsedAddresses = [];
-    tokens.forEach(token => {
-        if (token.type === 'operator' && (token.value === ',' || token.value === ';')) {
-            if (address.length) {
-                addresses.push(address);
-            }
-            address = [];
-        }
-        else {
-            address.push(token);
+    let group = null;
+    let members = [];
+    // Every address is handed over as the range of the token list it covers, as soon as the
+    // separator after it is seen. Copying the tokens of each address into an array of its own,
+    // and collecting those before handling any, held a second structure as large as the token
+    // list for the whole parse, and an array that receives a single push is allocated room for
+    // seventeen: a list of one letter names took some 80 times its size in that alone
+    let from = 0;
+    for (let i = 0, len = tokens.length; i <= len; i++) {
+        const operator = i < len ? tokens.operatorAt(i) : '';
+        if (i < len && operator !== ',' && operator !== ';' && operator !== ':') {
+            continue;
         }
-    });
-    if (address.length) {
-        addresses.push(address);
-    }
-    addresses.forEach(addr => {
-        const handled = _handleAddress(addr, depth);
-        // Appended in place. Rebuilding the accumulator with concat() would copy every
-        // entry collected so far on each address, making a flat list cost O(n^2).
-        for (let i = 0; i < handled.length; i++) {
-            parsedAddresses.push(handled[i]);
+        if (operator === ':') {
+            if (!group) {
+                group = { name: _groupName(tokens, from, i), group: [] };
+                members = [];
+            }
+            from = i + 1;
+            continue;
         }
-    });
-    // Merge fragments produced when unquoted display names contain commas.
-    // "Joe Foo, PhD <joe@example.com>" is split on the comma into
-    // [{name:"Joe Foo", address:""}, {name:"PhD", address:"joe@example.com"}].
-    // Recombine: a name-only entry followed by an entry with both name and address.
-    // Walked back to front so that a run of fragments folds into one entry in a single
-    // pass. Splicing each fragment out of the list instead would cost O(n^2).
-    const mergedAddresses = [];
-    for (let i = parsedAddresses.length - 1; i >= 0; i--) {
-        const current = parsedAddresses[i];
-        const next = mergedAddresses.length ? mergedAddresses[mergedAddresses.length - 1] : null;
-        if (next && current.address === '' && current.name && !current.group && next.address && next.name) {
-            next.name = current.name + ', ' + next.name;
+        if (i > from) {
+            // Appended in place. Rebuilding the accumulator with concat() would copy every
+            // entry collected so far on each address, making a flat list cost O(n^2).
+            const address = _handleAddress(tokens, from, i);
+            if (group) {
+                members.push(address);
+            }
+            else {
+                parsedAddresses.push(address);
+            }
         }
-        else {
-            mergedAddresses.push(current);
+        from = i + 1;
+        if (group && (operator === ';' || i === len)) {
+            group.group = _mergeNameFragments(members);
+            parsedAddresses.push(group);
+            group = null;
         }
     }
-    mergedAddresses.reverse();
-    return mergedAddresses;
+    return _mergeNameFragments(parsedAddresses);
 }
 function addressparser(str, options) {
-    const parsedAddresses = _parseAddressList(str, 0);
+    const parsedAddresses = _parseAddressList(str);
     if (options?.flatten) {
         const flatAddresses = [];
         const walkAddressList = (list) => {