@suveren/gateway 0.17.0 → 0.18.1

package/node_modules/htmlparser2/src/Tokenizer.ts

@@ -0,0 +1,1150 @@
+import {
+    DecodingMode,
+    EntityDecoder,
+    htmlDecodeTree,
+    xmlDecodeTree,
+} from "entities/decode";
+
+const enum CharCodes {
+    Tab = 0x9, // "\t"
+    NewLine = 0xa, // "\n"
+    FormFeed = 0xc, // "\f"
+    CarriageReturn = 0xd, // "\r"
+    Space = 0x20, // " "
+    ExclamationMark = 0x21, // "!"
+    Number = 0x23, // "#"
+    Amp = 0x26, // "&"
+    SingleQuote = 0x27, // "'"
+    DoubleQuote = 0x22, // '"'
+    Dash = 0x2d, // "-"
+    Slash = 0x2f, // "/"
+    Zero = 0x30, // "0"
+    Nine = 0x39, // "9"
+    Semi = 0x3b, // ";"
+    Lt = 0x3c, // "<"
+    Eq = 0x3d, // "="
+    Gt = 0x3e, // ">"
+    Questionmark = 0x3f, // "?"
+    UpperA = 0x41, // "A"
+    LowerA = 0x61, // "a"
+    UpperF = 0x46, // "F"
+    LowerF = 0x66, // "f"
+    UpperZ = 0x5a, // "Z"
+    LowerZ = 0x7a, // "z"
+    LowerX = 0x78, // "x"
+    OpeningSquareBracket = 0x5b, // "["
+}
+
+/** All the states the tokenizer can be in. */
+const enum State {
+    Text = 1,
+    BeforeTagName, // After <
+    InTagName,
+    InSelfClosingTag,
+    BeforeClosingTagName,
+    InClosingTagName,
+    AfterClosingTagName,
+
+    // Attributes
+    BeforeAttributeName,
+    InAttributeName,
+    AfterAttributeName,
+    BeforeAttributeValue,
+    InAttributeValueDq, // "
+    InAttributeValueSq, // '
+    InAttributeValueNq,
+
+    // Declarations
+    BeforeDeclaration, // !
+    InDeclaration,
+
+    // Processing instructions
+    InProcessingInstruction, // ?
+
+    // Comments & CDATA
+    BeforeComment,
+    CDATASequence,
+    DeclarationSequence,
+    InSpecialComment,
+    InCommentLike,
+
+    // Special tags
+    SpecialStartSequence,
+    InSpecialTag,
+    InPlainText,
+
+    InEntity,
+}
+
+function isWhitespace(c: number): boolean {
+    return (
+        c === CharCodes.Space ||
+        c === CharCodes.NewLine ||
+        c === CharCodes.Tab ||
+        c === CharCodes.FormFeed ||
+        c === CharCodes.CarriageReturn
+    );
+}
+
+function isEndOfTagSection(c: number): boolean {
+    return c === CharCodes.Slash || c === CharCodes.Gt || isWhitespace(c);
+}
+
+function isASCIIAlpha(c: number): boolean {
+    return (
+        (c >= CharCodes.LowerA && c <= CharCodes.LowerZ) ||
+        (c >= CharCodes.UpperA && c <= CharCodes.UpperZ)
+    );
+}
+
+/**
+ * Quote style used for parsed attributes.
+ */
+export enum QuoteType {
+    NoValue = 0,
+    Unquoted = 1,
+    Single = 2,
+    Double = 3,
+}
+
+/**
+ * Low-level tokenizer callback interface.
+ */
+export interface Callbacks {
+    onattribdata(start: number, endIndex: number): void;
+    onattribentity(codepoint: number): void;
+    onattribend(quote: QuoteType, endIndex: number): void;
+    onattribname(start: number, endIndex: number): void;
+    oncdata(start: number, endIndex: number, endOffset: number): void;
+    onclosetag(start: number, endIndex: number): void;
+    oncomment(start: number, endIndex: number, endOffset: number): void;
+    ondeclaration(start: number, endIndex: number): void;
+    onend(): void;
+    onopentagend(endIndex: number): void;
+    onopentagname(start: number, endIndex: number): void;
+    onprocessinginstruction(start: number, endIndex: number): void;
+    onselfclosingtag(endIndex: number): void;
+    ontext(start: number, endIndex: number): void;
+    ontextentity(codepoint: number, endIndex: number): void;
+    isInForeignContext?(): boolean;
+}
+
+/**
+ * Sequences used to match longer strings.
+ *
+ * We don't have `Script`, `Style`, or `Title` here. Instead, we re-use the *End
+ * sequences with an increased offset.
+ */
+const Sequences = {
+    Empty: new Uint8Array(0),
+    Cdata: new Uint8Array([0x43, 0x44, 0x41, 0x54, 0x41, 0x5b]), // CDATA[
+    CdataEnd: new Uint8Array([0x5d, 0x5d, 0x3e]), // ]]>
+    CommentEnd: new Uint8Array([0x2d, 0x2d, 0x21, 0x3e]), // `--!>`
+    Doctype: new Uint8Array([0x64, 0x6f, 0x63, 0x74, 0x79, 0x70, 0x65]), // `doctype`
+    IframeEnd: new Uint8Array([0x3c, 0x2f, 0x69, 0x66, 0x72, 0x61, 0x6d, 0x65]), // `</iframe`
+    NoembedEnd: new Uint8Array([
+        0x3c, 0x2f, 0x6e, 0x6f, 0x65, 0x6d, 0x62, 0x65, 0x64,
+    ]), // `</noembed`
+    NoframesEnd: new Uint8Array([
+        0x3c, 0x2f, 0x6e, 0x6f, 0x66, 0x72, 0x61, 0x6d, 0x65, 0x73,
+    ]), // `</noframes`
+    Plaintext: new Uint8Array([
+        0x3c, 0x2f, 0x70, 0x6c, 0x61, 0x69, 0x6e, 0x74, 0x65, 0x78, 0x74,
+    ]), // `</plaintext`
+    ScriptEnd: new Uint8Array([0x3c, 0x2f, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74]), // `</script`
+    StyleEnd: new Uint8Array([0x3c, 0x2f, 0x73, 0x74, 0x79, 0x6c, 0x65]), // `</style`
+    TitleEnd: new Uint8Array([0x3c, 0x2f, 0x74, 0x69, 0x74, 0x6c, 0x65]), // `</title`
+    TextareaEnd: new Uint8Array([
+        0x3c, 0x2f, 0x74, 0x65, 0x78, 0x74, 0x61, 0x72, 0x65, 0x61,
+    ]), // `</textarea`
+    XmpEnd: new Uint8Array([0x3c, 0x2f, 0x78, 0x6d, 0x70]), // `</xmp`
+};
+
+/**
+ * Maps the first lowercase character of an HTML tag name to the sequence
+ * used for special-tag detection.  All sequences share a common layout
+ * where index 2 is the first tag-name character, so matching always
+ * continues from offset 3.
+ */
+const specialStartSequences = new Map<number, Uint8Array>([
+    [Sequences.IframeEnd[2], Sequences.IframeEnd],
+    [Sequences.NoembedEnd[2], Sequences.NoembedEnd],
+    [Sequences.Plaintext[2], Sequences.Plaintext],
+    [Sequences.ScriptEnd[2], Sequences.ScriptEnd],
+    [Sequences.TitleEnd[2], Sequences.TitleEnd],
+    [Sequences.XmpEnd[2], Sequences.XmpEnd],
+]);
+
+/**
+ * Tokenizer implementation used by `Parser`.
+ */
+export default class Tokenizer {
+    /** The current state the tokenizer is in. */
+    private state = State.Text;
+    /** The read buffer. */
+    private buffer = "";
+    /** The beginning of the section that is currently being read. */
+    private sectionStart = 0;
+    /** The index within the buffer that we are currently looking at. */
+    private index = 0;
+    /** The start of the last entity. */
+    private entityStart = 0;
+    /** Some behavior, eg. when decoding entities, is done while we are in another state. This keeps track of the other state type. */
+    private baseState = State.Text;
+    /** For special parsing behavior inside of script and style tags. */
+    private isSpecial = false;
+    /** Indicates whether the tokenizer has been paused. */
+    running = true;
+    /** The offset of the current buffer. */
+    private offset = 0;
+
+    private readonly xmlMode: boolean;
+    private readonly decodeEntities: boolean;
+    private readonly recognizeSelfClosing: boolean;
+    private readonly entityDecoder: EntityDecoder;
+
+    constructor(
+        {
+            xmlMode = false,
+            decodeEntities = true,
+            recognizeSelfClosing = xmlMode,
+        }: {
+            xmlMode?: boolean;
+            decodeEntities?: boolean;
+            recognizeSelfClosing?: boolean;
+        },
+        private readonly cbs: Callbacks,
+    ) {
+        this.xmlMode = xmlMode;
+        this.decodeEntities = decodeEntities;
+        this.recognizeSelfClosing = recognizeSelfClosing;
+        this.entityDecoder = new EntityDecoder(
+            xmlMode ? xmlDecodeTree : htmlDecodeTree,
+            (cp, consumed) => this.emitCodePoint(cp, consumed),
+        );
+    }
+
+    reset(): void {
+        this.state = State.Text;
+        this.buffer = "";
+        this.sectionStart = 0;
+        this.index = 0;
+        this.baseState = State.Text;
+        this.isSpecial = false;
+        this.currentSequence = Sequences.Empty;
+        this.sequenceIndex = 0;
+        this.running = true;
+        this.offset = 0;
+    }
+
+    write(chunk: string): void {
+        this.offset += this.buffer.length;
+        this.buffer = chunk;
+        this.parse();
+    }
+
+    end(): void {
+        if (this.running) this.finish();
+    }
+
+    pause(): void {
+        this.running = false;
+    }
+
+    resume(): void {
+        this.running = true;
+        if (this.index < this.buffer.length + this.offset) {
+            this.parse();
+        }
+    }
+
+    private stateText(c: number): void {
+        if (
+            c === CharCodes.Lt ||
+            (!this.decodeEntities && this.fastForwardTo(CharCodes.Lt))
+        ) {
+            if (this.index > this.sectionStart) {
+                this.cbs.ontext(this.sectionStart, this.index);
+            }
+            this.state = State.BeforeTagName;
+            this.sectionStart = this.index;
+        } else if (this.decodeEntities && c === CharCodes.Amp) {
+            this.startEntity();
+        }
+    }
+
+    private currentSequence: Uint8Array = Sequences.Empty;
+    private sequenceIndex = 0;
+
+    private enterTagBody(): void {
+        if (this.currentSequence === Sequences.Plaintext) {
+            this.currentSequence = Sequences.Empty;
+            this.state = State.InPlainText;
+        } else if (this.isSpecial) {
+            this.state = State.InSpecialTag;
+            this.sequenceIndex = 0;
+        } else {
+            this.state = State.Text;
+        }
+    }
+
+    /**
+     * Match the opening tag name against an HTML text-only tag sequence.
+     *
+     * Some tags share an initial prefix (`script`/`style`, `title`/`textarea`,
+     * `noembed`/`noframes`), so we may switch to an alternate sequence at the
+     * first distinguishing byte.  On a successful full match we fall back to
+     * the normal tag-name state; a later `>` will enter raw-text, RCDATA, or
+     * plaintext mode based on `currentSequence` / `isSpecial`.
+     * @param c Current character code point.
+     */
+    private stateSpecialStartSequence(c: number): void {
+        const lower = c | 0x20;
+
+        // Still matching — check for an alternate sequence at branch points.
+        if (this.sequenceIndex < this.currentSequence.length) {
+            if (lower === this.currentSequence[this.sequenceIndex]) {
+                this.sequenceIndex++;
+                return;
+            }
+
+            if (this.sequenceIndex === 3) {
+                if (
+                    this.currentSequence === Sequences.ScriptEnd &&
+                    lower === Sequences.StyleEnd[3]
+                ) {
+                    this.currentSequence = Sequences.StyleEnd;
+                    this.sequenceIndex = 4;
+                    return;
+                }
+
+                if (
+                    this.currentSequence === Sequences.TitleEnd &&
+                    lower === Sequences.TextareaEnd[3]
+                ) {
+                    this.currentSequence = Sequences.TextareaEnd;
+                    this.sequenceIndex = 4;
+                    return;
+                }
+            } else if (
+                this.sequenceIndex === 4 &&
+                this.currentSequence === Sequences.NoembedEnd &&
+                lower === Sequences.NoframesEnd[4]
+            ) {
+                this.currentSequence = Sequences.NoframesEnd;
+                this.sequenceIndex = 5;
+                return;
+            }
+        } else if (isEndOfTagSection(c)) {
+            // Full match on a valid tag boundary — keep the sequence.
+            this.sequenceIndex = 0;
+            this.state = State.InTagName;
+            this.stateInTagName(c);
+            return;
+        }
+
+        // No match — abandon special-tag detection.
+        this.isSpecial = false;
+        this.currentSequence = Sequences.Empty;
+        this.sequenceIndex = 0;
+        this.state = State.InTagName;
+        this.stateInTagName(c);
+    }
+
+    private stateCDATASequence(c: number): void {
+        if (c === Sequences.Cdata[this.sequenceIndex]) {
+            if (++this.sequenceIndex === Sequences.Cdata.length) {
+                this.state = State.InCommentLike;
+                this.currentSequence = Sequences.CdataEnd;
+                this.sequenceIndex = 0;
+                this.sectionStart = this.index + 1;
+            }
+        } else {
+            this.sequenceIndex = 0;
+            if (this.xmlMode) {
+                this.state = State.InDeclaration;
+                this.stateInDeclaration(c); // Reconsume the character
+            } else {
+                this.state = State.InSpecialComment;
+                this.stateInSpecialComment(c); // Reconsume the character
+            }
+        }
+    }
+
+    /**
+     * When we wait for one specific character, we can speed things up
+     * by skipping through the buffer until we find it.
+     * @param c Current character code point.
+     * @returns Whether the character was found.
+     */
+    private fastForwardTo(c: number): boolean {
+        while (++this.index < this.buffer.length + this.offset) {
+            if (this.buffer.charCodeAt(this.index - this.offset) === c) {
+                return true;
+            }
+        }
+
+        /*
+         * We increment the index at the end of the `parse` loop,
+         * so set it to `buffer.length - 1` here.
+         *
+         * TODO: Refactor `parse` to increment index before calling states.
+         */
+        this.index = this.buffer.length + this.offset - 1;
+
+        return false;
+    }
+
+    /**
+     * Emit a comment token and return to the text state.
+     * @param offset Number of characters in the end sequence that have already been matched.
+     */
+    private emitComment(offset: number): void {
+        this.cbs.oncomment(this.sectionStart, this.index, offset);
+        this.sequenceIndex = 0;
+        this.sectionStart = this.index + 1;
+        this.state = State.Text;
+    }
+
+    /**
+     * Comments and CDATA end with `-->` and `]]>`.
+     *
+     * Their common qualities are:
+     * - Their end sequences have a distinct character they start with.
+     * - That character is then repeated, so we have to check multiple repeats.
+     * - All characters but the start character of the sequence can be skipped.
+     * @param c Current character code point.
+     */
+    private stateInCommentLike(c: number): void {
+        if (
+            !this.xmlMode &&
+            this.currentSequence === Sequences.CommentEnd &&
+            this.sequenceIndex <= 1 &&
+            /*
+             * We're still at the very start of the comment: the only
+             * characters consumed since `<!--` are the dashes that
+             * advanced sequenceIndex (0 for `<!-->`, 1 for `<!--->`).
+             */
+            this.index === this.sectionStart + this.sequenceIndex &&
+            c === CharCodes.Gt
+        ) {
+            // Abruptly closed empty HTML comment.
+            this.emitComment(this.sequenceIndex);
+        } else if (
+            this.currentSequence === Sequences.CommentEnd &&
+            this.sequenceIndex === 2 &&
+            c === CharCodes.Gt
+        ) {
+            // `!` is optional here, so the same sequence also accepts `-->`.
+            this.emitComment(2);
+        } else if (
+            this.currentSequence === Sequences.CommentEnd &&
+            this.sequenceIndex === this.currentSequence.length - 1 &&
+            c !== CharCodes.Gt
+        ) {
+            this.sequenceIndex = Number(c === CharCodes.Dash);
+        } else if (c === this.currentSequence[this.sequenceIndex]) {
+            if (++this.sequenceIndex === this.currentSequence.length) {
+                if (this.currentSequence === Sequences.CdataEnd) {
+                    this.cbs.oncdata(this.sectionStart, this.index, 2);
+                } else {
+                    this.cbs.oncomment(this.sectionStart, this.index, 3);
+                }
+
+                this.sequenceIndex = 0;
+                this.sectionStart = this.index + 1;
+                this.state = State.Text;
+            }
+        } else if (this.sequenceIndex === 0) {
+            // Fast-forward to the first character of the sequence
+            if (this.fastForwardTo(this.currentSequence[0])) {
+                this.sequenceIndex = 1;
+            }
+        } else if (c !== this.currentSequence[this.sequenceIndex - 1]) {
+            // Allow long sequences, eg. --->, ]]]>
+            this.sequenceIndex = 0;
+        }
+    }
+
+    /**
+     * HTML only allows ASCII alpha characters (a-z and A-Z) at the beginning of a tag name.
+     *
+     * XML allows a lot more characters here (@see https://www.w3.org/TR/REC-xml/#NT-NameStartChar).
+     * We allow anything that wouldn't end the tag.
+     * @param c Current character code point.
+     */
+    private isTagStartChar(c: number) {
+        return this.xmlMode ? !isEndOfTagSection(c) : isASCIIAlpha(c);
+    }
+
+    /**
+     * Scan raw-text / RCDATA content for the matching end tag.
+     *
+     * For RCDATA tags (`<title>`, `<textarea>`) entities are decoded inline.
+     * For raw-text tags (`<script>`, `<style>`, etc.) we fast-forward to `<`.
+     * @param c Current character code point.
+     */
+    private stateInSpecialTag(c: number): void {
+        if (this.sequenceIndex === this.currentSequence.length) {
+            if (isEndOfTagSection(c)) {
+                const endOfText = this.index - this.currentSequence.length;
+
+                if (this.sectionStart < endOfText) {
+                    // Spoof the index so that reported locations match up.
+                    const actualIndex = this.index;
+                    this.index = endOfText;
+                    this.cbs.ontext(this.sectionStart, endOfText);
+                    this.index = actualIndex;
+                }
+
+                this.isSpecial = false;
+                this.sectionStart = endOfText + 2; // Skip over the `</`
+                this.stateInClosingTagName(c);
+                return; // We are done; skip the rest of the function.
+            }
+
+            this.sequenceIndex = 0;
+        }
+
+        if ((c | 0x20) === this.currentSequence[this.sequenceIndex]) {
+            this.sequenceIndex += 1;
+        } else if (this.sequenceIndex === 0) {
+            if (
+                this.currentSequence === Sequences.TitleEnd ||
+                this.currentSequence === Sequences.TextareaEnd
+            ) {
+                // RCDATA tags have to parse entities while still looking for their end tag.
+                if (this.decodeEntities && c === CharCodes.Amp) {
+                    this.startEntity();
+                }
+            } else if (this.fastForwardTo(CharCodes.Lt)) {
+                // Outside of RCDATA tags, we can fast-forward.
+                this.sequenceIndex = 1;
+            }
+        } else {
+            // If we see a `<`, set the sequence index to 1; useful for eg. `<</script>`.
+            this.sequenceIndex = Number(c === CharCodes.Lt);
+        }
+    }
+
+    private stateBeforeTagName(c: number): void {
+        if (c === CharCodes.ExclamationMark) {
+            this.state = State.BeforeDeclaration;
+            this.sectionStart = this.index + 1;
+        } else if (c === CharCodes.Questionmark) {
+            if (this.xmlMode) {
+                this.state = State.InProcessingInstruction;
+                this.sequenceIndex = 0;
+                this.sectionStart = this.index + 1;
+            } else {
+                this.state = State.InSpecialComment;
+                this.sectionStart = this.index;
+            }
+        } else if (this.isTagStartChar(c)) {
+            this.sectionStart = this.index;
+
+            const special =
+                this.xmlMode || this.cbs.isInForeignContext?.()
+                    ? undefined
+                    : specialStartSequences.get(c | 0x20);
+
+            if (special === undefined) {
+                this.state = State.InTagName;
+            } else {
+                this.isSpecial = true;
+                this.currentSequence = special;
+                this.sequenceIndex = 3;
+                this.state = State.SpecialStartSequence;
+            }
+        } else if (c === CharCodes.Slash) {
+            this.state = State.BeforeClosingTagName;
+        } else {
+            this.state = State.Text;
+            this.stateText(c);
+        }
+    }
+    private stateInTagName(c: number): void {
+        if (isEndOfTagSection(c)) {
+            this.cbs.onopentagname(this.sectionStart, this.index);
+            this.sectionStart = -1;
+            this.state = State.BeforeAttributeName;
+            this.stateBeforeAttributeName(c);
+        }
+    }
+    private stateBeforeClosingTagName(c: number): void {
+        if (isWhitespace(c)) {
+            if (this.xmlMode) {
+                // Ignore
+            } else {
+                this.state = State.InSpecialComment;
+                this.sectionStart = this.index;
+            }
+        } else if (c === CharCodes.Gt) {
+            this.state = State.Text;
+            if (!this.xmlMode) {
+                this.sectionStart = this.index + 1;
+            }
+        } else {
+            this.state = this.isTagStartChar(c)
+                ? State.InClosingTagName
+                : State.InSpecialComment;
+            this.sectionStart = this.index;
+        }
+    }
+    private stateInClosingTagName(c: number): void {
+        if (isEndOfTagSection(c)) {
+            this.cbs.onclosetag(this.sectionStart, this.index);
+            this.sectionStart = -1;
+            this.state = State.AfterClosingTagName;
+            this.stateAfterClosingTagName(c);
+        }
+    }
+    private stateAfterClosingTagName(c: number): void {
+        // Skip everything until ">"
+        if (c === CharCodes.Gt || this.fastForwardTo(CharCodes.Gt)) {
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        }
+    }
+    private stateBeforeAttributeName(c: number): void {
+        if (c === CharCodes.Gt) {
+            this.cbs.onopentagend(this.index);
+            this.enterTagBody();
+            this.sectionStart = this.index + 1;
+        } else if (c === CharCodes.Slash) {
+            this.state = State.InSelfClosingTag;
+        } else if (!isWhitespace(c)) {
+            this.state = State.InAttributeName;
+            this.sectionStart = this.index;
+        }
+    }
+    /**
+     * Handle `/` before `>` in an opening tag.
+     *
+     * In HTML mode, text-only tags ignore the self-closing flag and still enter
+     * their raw-text/RCDATA/plaintext state unless self-closing tags are being
+     * recognized. In XML mode, or for ordinary tags, the tokenizer returns to
+     * regular text parsing after emitting the self-closing callback.
+     * @param c Current character code point.
+     */
+    private stateInSelfClosingTag(c: number): void {
+        if (c === CharCodes.Gt) {
+            this.cbs.onselfclosingtag(this.index);
+            this.sectionStart = this.index + 1;
+
+            if (!this.recognizeSelfClosing) {
+                this.enterTagBody();
+                return;
+            }
+
+            this.state = State.Text;
+            this.isSpecial = false; // Reset special state, in case of self-closing special tags
+            this.currentSequence = Sequences.Empty;
+        } else if (!isWhitespace(c)) {
+            this.state = State.BeforeAttributeName;
+            this.stateBeforeAttributeName(c);
+        }
+    }
+    private stateInAttributeName(c: number): void {
+        if (c === CharCodes.Eq || isEndOfTagSection(c)) {
+            this.cbs.onattribname(this.sectionStart, this.index);
+            this.sectionStart = this.index;
+            this.state = State.AfterAttributeName;
+            this.stateAfterAttributeName(c);
+        }
+    }
+    private stateAfterAttributeName(c: number): void {
+        if (c === CharCodes.Eq) {
+            this.state = State.BeforeAttributeValue;
+        } else if (c === CharCodes.Slash || c === CharCodes.Gt) {
+            this.cbs.onattribend(QuoteType.NoValue, this.sectionStart);
+            this.sectionStart = -1;
+            this.state = State.BeforeAttributeName;
+            this.stateBeforeAttributeName(c);
+        } else if (!isWhitespace(c)) {
+            this.cbs.onattribend(QuoteType.NoValue, this.sectionStart);
+            this.state = State.InAttributeName;
+            this.sectionStart = this.index;
+        }
+    }
+    private stateBeforeAttributeValue(c: number): void {
+        if (c === CharCodes.DoubleQuote) {
+            this.state = State.InAttributeValueDq;
+            this.sectionStart = this.index + 1;
+        } else if (c === CharCodes.SingleQuote) {
+            this.state = State.InAttributeValueSq;
+            this.sectionStart = this.index + 1;
+        } else if (!isWhitespace(c)) {
+            this.sectionStart = this.index;
+            this.state = State.InAttributeValueNq;
+            this.stateInAttributeValueNoQuotes(c); // Reconsume token
+        }
+    }
+    private handleInAttributeValue(c: number, quote: number) {
+        if (
+            c === quote ||
+            (!this.decodeEntities && this.fastForwardTo(quote))
+        ) {
+            this.cbs.onattribdata(this.sectionStart, this.index);
+            this.sectionStart = -1;
+            this.cbs.onattribend(
+                quote === CharCodes.DoubleQuote
+                    ? QuoteType.Double
+                    : QuoteType.Single,
+                this.index + 1,
+            );
+            this.state = State.BeforeAttributeName;
+        } else if (this.decodeEntities && c === CharCodes.Amp) {
+            this.startEntity();
+        }
+    }
+    private stateInAttributeValueDoubleQuotes(c: number): void {
+        this.handleInAttributeValue(c, CharCodes.DoubleQuote);
+    }
+    private stateInAttributeValueSingleQuotes(c: number): void {
+        this.handleInAttributeValue(c, CharCodes.SingleQuote);
+    }
+    private stateInAttributeValueNoQuotes(c: number): void {
+        if (isWhitespace(c) || c === CharCodes.Gt) {
+            this.cbs.onattribdata(this.sectionStart, this.index);
+            this.sectionStart = -1;
+            this.cbs.onattribend(QuoteType.Unquoted, this.index);
+            this.state = State.BeforeAttributeName;
+            this.stateBeforeAttributeName(c);
+        } else if (this.decodeEntities && c === CharCodes.Amp) {
+            this.startEntity();
+        }
+    }
+    /**
+     * Distinguish between CDATA, declarations, HTML comments, and HTML bogus
+     * comments after `<!`.
+     *
+     * In HTML mode, only real comments and doctypes stay on declaration paths;
+     * everything else becomes a bogus comment terminated by the next `>`.
+     * @param c Current character code point.
+     */
+    private stateBeforeDeclaration(c: number): void {
+        if (c === CharCodes.OpeningSquareBracket) {
+            this.state = State.CDATASequence;
+            this.sequenceIndex = 0;
+        } else if (this.xmlMode) {
+            this.state =
+                c === CharCodes.Dash
+                    ? State.BeforeComment
+                    : State.InDeclaration;
+        } else if ((c | 0x20) === Sequences.Doctype[0]) {
+            this.state = State.DeclarationSequence;
+            this.currentSequence = Sequences.Doctype;
+            this.sequenceIndex = 1;
+        } else if (c === CharCodes.Gt) {
+            this.cbs.oncomment(this.sectionStart, this.index, 0);
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        } else if (c === CharCodes.Dash) {
+            this.state = State.BeforeComment;
+        } else {
+            this.state = State.InSpecialComment;
+        }
+    }
+    /**
+     * Continue matching `doctype` after `<!d`.
+     *
+     * A full `doctype` match stays on the declaration path; any other name falls
+     * back to an HTML bogus comment, which matches browser behavior for
+     * non-doctype `<!...>` constructs.
+     * @param c Current character code point.
+     */
+    private stateDeclarationSequence(c: number): void {
+        if (this.sequenceIndex === this.currentSequence.length) {
+            this.state = State.InDeclaration;
+            this.stateInDeclaration(c);
+        } else if ((c | 0x20) === this.currentSequence[this.sequenceIndex]) {
+            this.sequenceIndex += 1;
+        } else if (c === CharCodes.Gt) {
+            this.cbs.oncomment(this.sectionStart, this.index, 0);
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        } else {
+            this.state = State.InSpecialComment;
+        }
+    }
+    private stateInDeclaration(c: number): void {
+        if (c === CharCodes.Gt || this.fastForwardTo(CharCodes.Gt)) {
+            this.cbs.ondeclaration(this.sectionStart, this.index);
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        }
+    }
+    /**
+     * XML processing instructions (`<?...?>`).
+     *
+     * In HTML mode `<?` is routed to `InSpecialComment` instead, so this
+     * state is only reachable in XML mode.
+     * @param c Current character code point.
+     */
+    private stateInProcessingInstruction(c: number): void {
+        if (c === CharCodes.Questionmark) {
+            // Remember that we just consumed `?`, so the next `>` closes the PI.
+            this.sequenceIndex = 1;
+        } else if (c === CharCodes.Gt && this.sequenceIndex === 1) {
+            this.cbs.onprocessinginstruction(this.sectionStart, this.index - 1);
+            this.sequenceIndex = 0;
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        } else {
+            // Keep scanning for the next `?`, which can start a closing `?>`.
+            this.sequenceIndex = Number(
+                this.fastForwardTo(CharCodes.Questionmark),
+            );
+        }
+    }
+    private stateBeforeComment(c: number): void {
+        if (c === CharCodes.Dash) {
+            this.state = State.InCommentLike;
+            this.currentSequence = Sequences.CommentEnd;
+            this.sequenceIndex = 0;
+            this.sectionStart = this.index + 1;
+        } else if (this.xmlMode) {
+            this.state = State.InDeclaration;
+        } else if (c === CharCodes.Gt) {
+            this.cbs.oncomment(this.sectionStart, this.index, 0);
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        } else {
+            this.state = State.InSpecialComment;
+        }
+    }
+    private stateInSpecialComment(c: number): void {
+        if (c === CharCodes.Gt || this.fastForwardTo(CharCodes.Gt)) {
+            this.cbs.oncomment(this.sectionStart, this.index, 0);
+            this.state = State.Text;
+            this.sectionStart = this.index + 1;
+        }
+    }
+
+    private startEntity() {
+        this.baseState = this.state;
+        this.state = State.InEntity;
+        this.entityStart = this.index;
+        this.entityDecoder.startEntity(
+            this.xmlMode
+                ? DecodingMode.Strict
+                : this.baseState === State.Text ||
+                    this.baseState === State.InSpecialTag
+                  ? DecodingMode.Legacy
+                  : DecodingMode.Attribute,
+        );
+    }
+
+    private stateInEntity(): void {
+        const indexInBuffer = this.index - this.offset;
+        const length = this.entityDecoder.write(this.buffer, indexInBuffer);
+
+        // If `length` is positive, we are done with the entity.
+        if (length >= 0) {
+            this.state = this.baseState;
+
+            if (length === 0) {
+                this.index -= 1;
+            }
+        } else {
+            if (
+                indexInBuffer < this.buffer.length &&
+                this.buffer.charCodeAt(indexInBuffer) === CharCodes.Amp
+            ) {
+                this.state = this.baseState;
+                this.index -= 1;
+                return;
+            }
+
+            // Mark buffer as consumed.
+            this.index = this.offset + this.buffer.length - 1;
+        }
+    }
+
+    /**
+     * Remove data that has already been consumed from the buffer.
+     */
+    private cleanup() {
+        // If we are inside of text or attributes, emit what we already have.
+        if (this.running && this.sectionStart !== this.index) {
+            if (
+                this.state === State.Text ||
+                this.state === State.InPlainText ||
+                (this.state === State.InSpecialTag && this.sequenceIndex === 0)
+            ) {
+                this.cbs.ontext(this.sectionStart, this.index);
+                this.sectionStart = this.index;
+            } else if (
+                this.state === State.InAttributeValueDq ||
+                this.state === State.InAttributeValueSq ||
+                this.state === State.InAttributeValueNq
+            ) {
+                this.cbs.onattribdata(this.sectionStart, this.index);
+                this.sectionStart = this.index;
+            }
+        }
+    }
+
+    private shouldContinue() {
+        return this.index < this.buffer.length + this.offset && this.running;
+    }
+
+    /**
+     * Iterates through the buffer, calling the function corresponding to the current state.
+     *
+     * States that are more likely to be hit are higher up, as a performance improvement.
+     */
+    private parse() {
+        while (this.shouldContinue()) {
+            const c = this.buffer.charCodeAt(this.index - this.offset);
+            switch (this.state) {
+                case State.Text: {
+                    this.stateText(c);
+                    break;
+                }
+                case State.InPlainText: {
+                    // Skip to end of buffer; cleanup() emits the text.
+                    this.index = this.buffer.length + this.offset - 1;
+                    break;
+                }
+                case State.SpecialStartSequence: {
+                    this.stateSpecialStartSequence(c);
+                    break;
+                }
+                case State.InSpecialTag: {
+                    this.stateInSpecialTag(c);
+                    break;
+                }
+                case State.CDATASequence: {
+                    this.stateCDATASequence(c);
+                    break;
+                }
+                case State.DeclarationSequence: {
+                    this.stateDeclarationSequence(c);
+                    break;
+                }
+                case State.InAttributeValueDq: {
+                    this.stateInAttributeValueDoubleQuotes(c);
+                    break;
+                }
+                case State.InAttributeName: {
+                    this.stateInAttributeName(c);
+                    break;
+                }
+                case State.InCommentLike: {
+                    this.stateInCommentLike(c);
+                    break;
+                }
+                case State.InSpecialComment: {
+                    this.stateInSpecialComment(c);
+                    break;
+                }
+                case State.BeforeAttributeName: {
+                    this.stateBeforeAttributeName(c);
+                    break;
+                }
+                case State.InTagName: {
+                    this.stateInTagName(c);
+                    break;
+                }
+                case State.InClosingTagName: {
+                    this.stateInClosingTagName(c);
+                    break;
+                }
+                case State.BeforeTagName: {
+                    this.stateBeforeTagName(c);
+                    break;
+                }
+                case State.AfterAttributeName: {
+                    this.stateAfterAttributeName(c);
+                    break;
+                }
+                case State.InAttributeValueSq: {
+                    this.stateInAttributeValueSingleQuotes(c);
+                    break;
+                }
+                case State.BeforeAttributeValue: {
+                    this.stateBeforeAttributeValue(c);
+                    break;
+                }
+                case State.BeforeClosingTagName: {
+                    this.stateBeforeClosingTagName(c);
+                    break;
+                }
+                case State.AfterClosingTagName: {
+                    this.stateAfterClosingTagName(c);
+                    break;
+                }
+                case State.InAttributeValueNq: {
+                    this.stateInAttributeValueNoQuotes(c);
+                    break;
+                }
+                case State.InSelfClosingTag: {
+                    this.stateInSelfClosingTag(c);
+                    break;
+                }
+                case State.InDeclaration: {
+                    this.stateInDeclaration(c);
+                    break;
+                }
+                case State.BeforeDeclaration: {
+                    this.stateBeforeDeclaration(c);
+                    break;
+                }
+                case State.BeforeComment: {
+                    this.stateBeforeComment(c);
+                    break;
+                }
+                case State.InProcessingInstruction: {
+                    this.stateInProcessingInstruction(c);
+                    break;
+                }
+                case State.InEntity: {
+                    this.stateInEntity();
+                    break;
+                }
+            }
+            this.index++;
+        }
+        this.cleanup();
+    }
+
+    private finish() {
+        if (this.state === State.InEntity) {
+            this.entityDecoder.end();
+            this.state = this.baseState;
+        }
+
+        this.handleTrailingData();
+
+        this.cbs.onend();
+    }
+
+    private handleTrailingCommentLikeData(endIndex: number): boolean {
+        if (this.state !== State.InCommentLike) {
+            return false;
+        }
+
+        if (this.currentSequence === Sequences.CdataEnd) {
+            if (this.xmlMode) {
+                if (this.sectionStart < endIndex) {
+                    this.cbs.oncdata(this.sectionStart, endIndex, 0);
+                }
+            } else {
+                /* In HTML mode, unclosed CDATA is a bogus comment. */
+                const cdataStart =
+                    this.sectionStart - Sequences.Cdata.length - 1;
+                this.cbs.oncomment(cdataStart, endIndex, 0);
+            }
+        } else {
+            const offset = this.xmlMode
+                ? 0
+                : Math.min(this.sequenceIndex, Sequences.CommentEnd.length - 1);
+            this.cbs.oncomment(this.sectionStart, endIndex, offset);
+        }
+
+        return true;
+    }
+
+    private handleTrailingMarkupDeclaration(endIndex: number): boolean {
+        if (this.xmlMode) {
+            switch (this.state) {
+                case State.InSpecialComment:
+                case State.BeforeComment:
+                case State.CDATASequence:
+                case State.DeclarationSequence:
+                case State.InDeclaration: {
+                    this.cbs.ontext(this.sectionStart, endIndex);
+                    return true;
+                }
+                default: {
+                    return false;
+                }
+            }
+        }
+
+        switch (this.state) {
+            case State.BeforeDeclaration:
+            case State.InSpecialComment:
+            case State.BeforeComment:
+            case State.CDATASequence: {
+                this.cbs.oncomment(this.sectionStart, endIndex, 0);
+                return true;
+            }
+            case State.DeclarationSequence: {
+                if (this.sequenceIndex !== Sequences.Doctype.length) {
+                    this.cbs.oncomment(this.sectionStart, endIndex, 0);
+                }
+                return true;
+            }
+            case State.InDeclaration: {
+                return true;
+            }
+            default: {
+                return false;
+            }
+        }
+    }
+
+    /** Handle any trailing data. */
+    private handleTrailingData() {
+        const endIndex = this.buffer.length + this.offset;
+
+        if (
+            this.handleTrailingCommentLikeData(endIndex) ||
+            this.handleTrailingMarkupDeclaration(endIndex)
+        ) {
+            return;
+        }
+
+        // If there is no remaining data, we are done.
+        if (this.sectionStart >= endIndex) {
+            return;
+        }
+
+        switch (this.state) {
+            case State.InTagName:
+            case State.BeforeAttributeName:
+            case State.BeforeAttributeValue:
+            case State.AfterAttributeName:
+            case State.InAttributeName:
+            case State.InAttributeValueSq:
+            case State.InAttributeValueDq:
+            case State.InAttributeValueNq:
+            case State.InClosingTagName: {
+                /*
+                 * If we are currently in an opening or closing tag, us not calling the
+                 * respective callback signals that the tag should be ignored.
+                 */
+                break;
+            }
+            default: {
+                this.cbs.ontext(this.sectionStart, endIndex);
+            }
+        }
+    }
+
+    private emitCodePoint(cp: number, consumed: number): void {
+        if (
+            this.baseState !== State.Text &&
+            this.baseState !== State.InSpecialTag
+        ) {
+            if (this.sectionStart < this.entityStart) {
+                this.cbs.onattribdata(this.sectionStart, this.entityStart);
+            }
+            this.sectionStart = this.entityStart + consumed;
+            this.index = this.sectionStart - 1;
+
+            this.cbs.onattribentity(cp);
+        } else {
+            if (this.sectionStart < this.entityStart) {
+                this.cbs.ontext(this.sectionStart, this.entityStart);
+            }
+            this.sectionStart = this.entityStart + consumed;
+            this.index = this.sectionStart - 1;
+
+            this.cbs.ontextentity(cp, this.sectionStart);
+        }
+    }
+}