Bug 1439855 - Introduce BinAST multipart tokenizer;r=arai
☠☠ backed out by 868d65e22482 ☠ ☠
authorDavid Teller <dteller@mozilla.com>
Thu, 05 Apr 2018 14:31:40 +0200
changeset 467769 cf13314fb3bfe7600e15a68281b5cdc1f0b2df78
parent 467768 6809950e6c1b3005bc85f2a541a27386e9766ad6
child 467770 1878e7f3c0e74d70a850a0c5ab31b59464ef6dd2
push id9165
push userasasaki@mozilla.com
push dateThu, 26 Apr 2018 21:04:54 +0000
treeherdermozilla-beta@064c3804de2e [default view] [failures only]
perfherder[talos] [build metrics] [platform microbench] (compared to previous push)
reviewersarai
bugs1439855
milestone61.0a1
first release with
nightly linux32
nightly linux64
nightly mac
nightly win32
nightly win64
last release without
nightly linux32
nightly linux64
nightly mac
nightly win32
nightly win64
Bug 1439855 - Introduce BinAST multipart tokenizer;r=arai This introduces a new tokenizer for the BinAST parser: the multipart tokenizer. The format used by the multipart tokenizer is described here: https://binast.github.io/binjs-ref/binjs_io/multipart/index.html While the details of the format may still change, this is the tokenizer we intend to ship. We only implement the `identity; ` compression scheme. For the moment, there are no plans to implement other compression schemes. MozReview-Commit-ID: 8WUa3QX5GWt
js/src/frontend/BinTokenReaderMultipart.cpp
js/src/frontend/BinTokenReaderMultipart.h
js/src/moz.build
new file mode 100644
--- /dev/null
+++ b/js/src/frontend/BinTokenReaderMultipart.cpp
@@ -0,0 +1,440 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
+ * vim: set ts=8 sts=4 et sw=4 tw=99:
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#include "frontend/BinTokenReaderMultipart.h"
+
+#include "mozilla/EndianUtils.h"
+#include "mozilla/Maybe.h"
+
+#include "frontend/BinSource-macros.h"
+#include "frontend/BinSourceRuntimeSupport.h"
+
+#include "gc/Zone.h"
+
+namespace js {
+namespace frontend {
+
+// The magic header, at the start of every binjs file.
+const char MAGIC_HEADER[] = "BINJS";
+// The latest format version understood by this tokenizer.
+const uint32_t MAGIC_FORMAT_VERSION = 0;
+
+// The headers at the start of each section of the binjs file.
+const char SECTION_HEADER_GRAMMAR[] = "[GRAMMAR]";
+const char SECTION_HEADER_STRINGS[] = "[STRINGS]";
+const char SECTION_HEADER_TREE[] = "[TREE]";
+
+// The (only) internal compression mechanism understood by this parser.
+const char COMPRESSION_IDENTITY[] = "identity;";
+
+
+// The maximal number of distinct strings that may be declared in a
+// single file.
+const uint32_t MAX_NUMBER_OF_STRINGS = 32768;
+
+using AutoList = BinTokenReaderMultipart::AutoList;
+using AutoTaggedTuple = BinTokenReaderMultipart::AutoTaggedTuple;
+using AutoTuple = BinTokenReaderMultipart::AutoTuple;
+using CharSlice = BinaryASTSupport::CharSlice;
+using Chars = BinTokenReaderMultipart::Chars;
+
+BinTokenReaderMultipart::BinTokenReaderMultipart(JSContext* cx, const uint8_t* start, const size_t length)
+  : BinTokenReaderBase(cx, start, length)
+  , grammarTable_(cx)
+  , atomsTable_(cx, AtomVector(cx))
+  , slicesTable_(cx)
+  , posBeforeTree_(nullptr)
+{ }
+
+JS::Result<Ok>
+BinTokenReaderMultipart::readHeader()
+{
+    // Check that we don't call this function twice.
+    MOZ_ASSERT(!posBeforeTree_);
+
+    // Read global headers.
+    MOZ_TRY(readConst(MAGIC_HEADER));
+    BINJS_MOZ_TRY_DECL(version, readInternalUint32());
+
+    // For the moment, MAGIC_FORMAT_VERSION is 0. Once we have a story
+    // on backwards compatibility of the binary container, we will
+    // probably want to change this to `if (version > MAGIC_FORMAT_VERSION)`.
+    if (version != MAGIC_FORMAT_VERSION)
+        return raiseError("Format version not implemented");
+
+    // Start reading grammar.
+    MOZ_TRY(readConst(SECTION_HEADER_GRAMMAR));
+    MOZ_TRY(readConst(COMPRESSION_IDENTITY)); // For the moment, we only support identity compression.
+    BINJS_MOZ_TRY_DECL(grammarByteLen, readInternalUint32());
+    const auto posBeforeGrammar = current_;
+
+    if (posBeforeGrammar + grammarByteLen > stop_ || posBeforeGrammar + grammarByteLen < current_) // Sanity check.
+        return raiseError("Invalid byte length in grammar table");
+
+    BINJS_MOZ_TRY_DECL(grammarNumberOfEntries, readInternalUint32());
+    if (grammarNumberOfEntries > BINKIND_LIMIT) // Sanity check.
+        return raiseError("Invalid number of entries in grammar table");
+
+    // This table maps BinKind index -> BinKind.
+    // Initialize and populate.
+    if (!grammarTable_.reserve(grammarNumberOfEntries))
+        return raiseOOM();
+
+    for (uint32_t i = 0; i < grammarNumberOfEntries; ++i) {
+        BINJS_MOZ_TRY_DECL(byteLen, readInternalUint32());
+        if (current_ + byteLen > stop_)
+            return raiseError("Invalid byte length in grammar table");
+        if (current_ + byteLen < current_) // Overflow.
+            return raiseError("Invalid byte length in grammar table");
+        CharSlice name((const char*)current_, byteLen);
+        current_ += byteLen;
+
+        BINJS_MOZ_TRY_DECL(kind, cx_->runtime()->binast().binKind(cx_, name));
+        if (!kind)
+            return raiseError("Invalid entry in grammar table");
+
+        grammarTable_.infallibleAppend(*kind); // We called `reserve` before the loop.
+    }
+    if (current_ != grammarByteLen + posBeforeGrammar)
+        return raiseError("The length of the grammar table didn't match its contents.");
+
+    // Start reading strings
+    MOZ_TRY(readConst(SECTION_HEADER_STRINGS));
+    MOZ_TRY(readConst(COMPRESSION_IDENTITY)); // For the moment, we only support identity compression.
+    BINJS_MOZ_TRY_DECL(stringsByteLen, readInternalUint32());
+    const auto posBeforeStrings = current_;
+
+    if (posBeforeStrings + stringsByteLen > stop_ || posBeforeStrings + stringsByteLen < current_) // Sanity check.
+        return raiseError("Invalid byte length in strings table");
+
+    BINJS_MOZ_TRY_DECL(stringsNumberOfEntries, readInternalUint32());
+    if (stringsNumberOfEntries > MAX_NUMBER_OF_STRINGS) // Sanity check.
+        return raiseError("Too many entries in strings table");
+
+    // This table maps String index -> String.
+    // Initialize and populate.
+    if (!atomsTable_.reserve(stringsNumberOfEntries))
+        return raiseOOM();
+    if (!slicesTable_.reserve(stringsNumberOfEntries))
+        return raiseOOM();
+    if (!variantsTable_.init())
+        return cx_->alreadyReportedError();
+
+    RootedAtom atom(cx_);
+    for (uint32_t i = 0; i < stringsNumberOfEntries; ++i) {
+        BINJS_MOZ_TRY_DECL(byteLen, readInternalUint32());
+        if (current_ + byteLen > stop_ || current_ + byteLen < current_)
+            return raiseError("Invalid byte length in individual string");
+
+        // Check null string.
+        if (byteLen == 2 && *current_ == 255 && *(current_ + 1) == 0) {
+            atom = nullptr;
+        } else {
+            BINJS_TRY_VAR(atom, Atomize(cx_, (const char*)current_, byteLen));
+        }
+
+        // Populate `atomsTable_`: i => atom.
+        atomsTable_.infallibleAppend(atom); // We have reserved before entering the loop.
+
+        // Populate `slicesTable_`: i => slice
+        Chars slice((const char*)current_, byteLen);
+        slicesTable_.infallibleAppend(Move(slice)); // We have reserved before entering the loop.
+
+        current_ += byteLen;
+    }
+
+    if (posBeforeStrings + stringsByteLen != current_)
+        return raiseError("The length of the strings table didn't match its contents.");
+
+    // Start reading AST.
+    MOZ_TRY(readConst(SECTION_HEADER_TREE));
+    MOZ_TRY(readConst(COMPRESSION_IDENTITY)); // For the moment, we only support identity compression.
+    posBeforeTree_ = current_;
+
+    BINJS_MOZ_TRY_DECL(treeByteLen, readInternalUint32());
+
+    if (posBeforeTree_ + treeByteLen > stop_ || posBeforeTree_ + treeByteLen < posBeforeTree_) // Sanity check.
+        return raiseError("Invalid byte length in tree table");
+
+    // At this stage, we're ready to start reading the tree.
+    return Ok();
+}
+
+JS::Result<bool>
+BinTokenReaderMultipart::readBool()
+{
+    updateLatestKnownGood();
+    BINJS_MOZ_TRY_DECL(byte, readByte());
+
+    switch (byte) {
+      case 0:
+        return false;
+      case 1:
+        return true;
+      case 2:
+        return raiseError("Not implemented: null boolean value");
+      default:
+        return raiseError("Invalid boolean value");
+    }
+}
+
+// Nullable doubles (little-endian)
+//
+// NULL_FLOAT_REPRESENTATION (signaling NaN) => null
+// anything other 64 bit sequence => IEEE-764 64-bit floating point number
+JS::Result<double>
+BinTokenReaderMultipart::readDouble()
+{
+    updateLatestKnownGood();
+
+    uint8_t bytes[8];
+    MOZ_ASSERT(sizeof(bytes) == sizeof(double));
+    MOZ_TRY(readBuf(reinterpret_cast<uint8_t*>(bytes), ArrayLength(bytes)));
+
+    // Decode little-endian.
+    const uint64_t asInt = LittleEndian::readUint64(bytes);
+
+    if (asInt == NULL_FLOAT_REPRESENTATION)
+        return raiseError("Not implemented: null double value");
+
+    // Canonicalize NaN, just to make sure another form of signalling NaN
+    // doesn't slip past us.
+    const double asDouble = CanonicalizeNaN(BitwiseCast<double>(asInt));
+    return asDouble;
+}
+
+
+// A single atom is represented as an index into the table of strings.
+JS::Result<JSAtom*>
+BinTokenReaderMultipart::readMaybeAtom()
+{
+    updateLatestKnownGood();
+    BINJS_MOZ_TRY_DECL(index, readInternalUint32());
+
+    if (index >= atomsTable_.length())
+        return raiseError("Invalid index to strings table");
+    return atomsTable_[index].get();
+}
+
+JS::Result<JSAtom*>
+BinTokenReaderMultipart::readAtom()
+{
+    BINJS_MOZ_TRY_DECL(maybe, readMaybeAtom());
+
+    if (!maybe)
+        return raiseError("Empty string");
+
+    return maybe;
+}
+
+JS::Result<Ok>
+BinTokenReaderMultipart::readChars(Chars& out)
+{
+    updateLatestKnownGood();
+    BINJS_MOZ_TRY_DECL(index, readInternalUint32());
+
+    if (index >= slicesTable_.length())
+        return raiseError("Invalid index to strings table for string enum");
+
+    out = slicesTable_[index];
+    return Ok();
+}
+
+JS::Result<BinVariant>
+BinTokenReaderMultipart::readVariant()
+{
+    updateLatestKnownGood();
+    BINJS_MOZ_TRY_DECL(index, readInternalUint32());
+
+    if (index >= slicesTable_.length())
+        return raiseError("Invalid index to strings table for string enum");
+
+    auto variantsPtr = variantsTable_.lookupForAdd(index);
+    if (variantsPtr)
+        return variantsPtr->value();
+
+    // Either we haven't cached the result yet or this is not a variant.
+    // Check in the slices table and, in case of success, cache the result.
+
+    // Note that we stop parsing if we attempt to readVariant() with an
+    // ill-formed variant, so we don't run the risk of feching an ill-variant
+    // more than once.
+    Chars slice = slicesTable_[index]; // We have checked `index` above.
+    BINJS_MOZ_TRY_DECL(variant, cx_->runtime()->binast().binVariant(cx_, slice));
+
+    if (!variant)
+        return raiseError("Invalid string enum variant");
+
+    if (!variantsTable_.add(variantsPtr, index, *variant))
+        return raiseOOM();
+
+    return *variant;
+}
+
+// Untagged tuple:
+// - contents (specified by the higher-level grammar);
+JS::Result<Ok>
+BinTokenReaderMultipart::enterUntaggedTuple(AutoTuple& guard)
+{
+    guard.init();
+    return Ok();
+}
+
+
+// Tagged tuples:
+// - uint32_t index in table [grammar];
+// - content (specified by the higher-level grammar);
+JS::Result<Ok>
+BinTokenReaderMultipart::enterTaggedTuple(BinKind& tag, BinTokenReaderMultipart::BinFields&, AutoTaggedTuple& guard)
+{
+    BINJS_MOZ_TRY_DECL(index, readInternalUint32());
+    if (index >= grammarTable_.length())
+        return raiseError("Invalid index to grammar table");
+
+    tag = grammarTable_[index];
+
+    // Enter the body.
+    guard.init();
+    return Ok();
+}
+
+// List:
+//
+// - uint32_t number of items;
+// - contents (specified by higher-level grammar);
+//
+// The total byte length of `number of items` + `contents` must be `byte length`.
+JS::Result<Ok>
+BinTokenReaderMultipart::enterList(uint32_t& items, AutoList& guard)
+{
+    guard.init();
+
+    MOZ_TRY_VAR(items, readInternalUint32());
+
+    return Ok();
+}
+
+void
+BinTokenReaderMultipart::AutoBase::init()
+{
+    initialized_ = true;
+}
+
+BinTokenReaderMultipart::AutoBase::AutoBase(BinTokenReaderMultipart& reader)
+    : reader_(reader)
+{ }
+
+BinTokenReaderMultipart::AutoBase::~AutoBase()
+{
+    // By now, the `AutoBase` must have been deinitialized by calling `done()`.
+    // The only case in which we can accept not calling `done()` is if we have
+    // bailed out because of an error.
+    MOZ_ASSERT_IF(initialized_, reader_.cx_->isExceptionPending());
+}
+
+JS::Result<Ok>
+BinTokenReaderMultipart::AutoBase::checkPosition(const uint8_t* expectedEnd)
+{
+    if (reader_.current_ != expectedEnd)
+        return reader_.raiseError("Caller did not consume the expected set of bytes");
+
+    return Ok();
+}
+
+
+
+BinTokenReaderMultipart::AutoList::AutoList(BinTokenReaderMultipart& reader)
+    : AutoBase(reader)
+{ }
+
+void
+BinTokenReaderMultipart::AutoList::init()
+{
+    AutoBase::init();
+}
+
+JS::Result<Ok>
+BinTokenReaderMultipart::AutoList::done()
+{
+    MOZ_ASSERT(initialized_);
+    initialized_ = false;
+    if (reader_.cx_->isExceptionPending()) {
+        // Already errored, no need to check further.
+        return reader_.cx_->alreadyReportedError();
+    }
+
+    return Ok();
+}
+
+
+// Internal uint32_t
+//
+// Encoded as variable length number.
+
+MOZ_MUST_USE JS::Result<uint32_t>
+BinTokenReaderMultipart::readInternalUint32()
+{
+    uint32_t result = 0;
+    uint32_t shift  = 0;
+    while (true) {
+        MOZ_ASSERT(shift < 32);
+        uint32_t byte;
+        MOZ_TRY_VAR(byte, readByte());
+
+        const uint32_t newResult = result | (byte >> 1) << shift;
+        if (newResult < result)
+            return raiseError("Overflow during readInternalUint32");
+
+        result = newResult;
+        shift += 7;
+
+        if ((byte & 1) == 0) {
+            return result;
+        }
+    }
+}
+
+
+BinTokenReaderMultipart::AutoTaggedTuple::AutoTaggedTuple(BinTokenReaderMultipart& reader)
+    : AutoBase(reader)
+{ }
+
+JS::Result<Ok>
+BinTokenReaderMultipart::AutoTaggedTuple::done()
+{
+    MOZ_ASSERT(initialized_);
+    initialized_ = false;
+    if (reader_.cx_->isExceptionPending()) {
+        // Already errored, no need to check further.
+        return reader_.cx_->alreadyReportedError();
+    }
+
+    return Ok();
+}
+
+BinTokenReaderMultipart::AutoTuple::AutoTuple(BinTokenReaderMultipart& reader)
+    : AutoBase(reader)
+{ }
+
+JS::Result<Ok>
+BinTokenReaderMultipart::AutoTuple::done()
+{
+    MOZ_ASSERT(initialized_);
+    initialized_ = false;
+    if (reader_.cx_->isExceptionPending()) {
+        // Already errored, no need to check further.
+        return reader_.cx_->alreadyReportedError();
+    }
+
+    // Check suffix.
+    return Ok();
+}
+
+} // namespace frontend
+
+
+} // namespace js
+
new file mode 100644
--- /dev/null
+++ b/js/src/frontend/BinTokenReaderMultipart.h
@@ -0,0 +1,299 @@
+/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*-
+ * vim: set ts=8 sts=4 et sw=4 tw=99:
+ * This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+#ifndef frontend_BinTokenReaderMultipart_h
+#define frontend_BinTokenReaderMultipart_h
+
+#include "mozilla/Maybe.h"
+
+#include "frontend/BinToken.h"
+#include "frontend/BinTokenReaderBase.h"
+
+namespace js {
+namespace frontend {
+
+using namespace mozilla;
+using namespace JS;
+
+/**
+ * A token reader implementing the "multipart" serialization format for BinAST.
+ *
+ * This serialization format, which is also supported by the reference
+ * implementation of the BinAST compression suite, is designed to be
+ * space- and time-efficient.
+ *
+ * As other token readers for the BinAST:
+ *
+ * - the reader does not support error recovery;
+ * - the reader does not support lookahead or pushback.
+ */
+class MOZ_STACK_CLASS BinTokenReaderMultipart: public BinTokenReaderBase
+{
+  public:
+    class AutoList;
+    class AutoTuple;
+    class AutoTaggedTuple;
+
+    using CharSlice = BinaryASTSupport::CharSlice;
+
+    // This implementation of `BinFields` is effectively `void`, as the format
+    // does not embed field information.
+    class BinFields {
+      public:
+        explicit BinFields(JSContext*) {}
+    };
+    struct Chars: public CharSlice {
+        explicit Chars(JSContext*)
+          : CharSlice(nullptr, 0)
+        { }
+        Chars(const char* start, const uint32_t byteLen)
+          : CharSlice(start, byteLen)
+        { }
+        Chars(const Chars& other) = default;
+    };
+
+  public:
+    /**
+     * Construct a token reader.
+     *
+     * Does NOT copy the buffer.
+     */
+    BinTokenReaderMultipart(JSContext* cx, const uint8_t* start, const size_t length);
+
+    /**
+     * Construct a token reader.
+     *
+     * Does NOT copy the buffer.
+     */
+    BinTokenReaderMultipart(JSContext* cx, const Vector<uint8_t>& chars);
+
+    /**
+     * Read the header of the file.
+     */
+    MOZ_MUST_USE JS::Result<Ok> readHeader();
+
+    // --- Primitive values.
+    //
+    // Note that the underlying format allows for a `null` value for primitive
+    // values.
+    //
+    // Reading will return an error either in case of I/O error or in case of
+    // a format problem. Reading if an exception in pending is an error and
+    // will cause assertion failures. Do NOT attempt to read once an exception
+    // has been cleared: the token reader does NOT support recovery, by design.
+
+    /**
+     * Read a single `true | false` value.
+     */
+    MOZ_MUST_USE JS::Result<bool> readBool();
+
+    /**
+     * Read a single `number` value.
+     */
+    MOZ_MUST_USE JS::Result<double> readDouble();
+
+    /**
+     * Read a single `string | null` value.
+     *
+     * Fails if that string is not valid UTF-8.
+     */
+    MOZ_MUST_USE JS::Result<JSAtom*> readMaybeAtom();
+    MOZ_MUST_USE JS::Result<JSAtom*> readAtom();
+
+
+    /**
+     * Read a single `string | null` value.
+     *
+     * MAY check if that string is not valid UTF-8.
+     */
+    MOZ_MUST_USE JS::Result<Ok> readChars(Chars&);
+
+    /**
+     * Read a single `BinVariant | null` value.
+     */
+    MOZ_MUST_USE JS::Result<Maybe<BinVariant>> readMaybeVariant();
+    MOZ_MUST_USE JS::Result<BinVariant> readVariant();
+
+    // --- Composite values.
+    //
+    // The underlying format does NOT allows for a `null` composite value.
+    //
+    // Reading will return an error either in case of I/O error or in case of
+    // a format problem. Reading from a poisoned tokenizer is an error and
+    // will cause assertion failures.
+
+    /**
+     * Start reading a list.
+     *
+     * @param length (OUT) The number of elements in the list.
+     * @param guard (OUT) A guard, ensuring that we read the list correctly.
+     *
+     * The `guard` is dedicated to ensuring that reading the list has consumed
+     * exactly all the bytes from that list. The `guard` MUST therefore be
+     * destroyed at the point where the caller has reached the end of the list.
+     * If the caller has consumed too few/too many bytes, this will be reported
+     * in the call go `guard.done()`.
+     */
+    MOZ_MUST_USE JS::Result<Ok> enterList(uint32_t& length, AutoList& guard);
+
+    /**
+     * Start reading a tagged tuple.
+     *
+     * @param tag (OUT) The tag of the tuple.
+     * @param fields Ignored, provided for API compatibility.
+     * @param guard (OUT) A guard, ensuring that we read the tagged tuple correctly.
+     *
+     * The `guard` is dedicated to ensuring that reading the list has consumed
+     * exactly all the bytes from that tuple. The `guard` MUST therefore be
+     * destroyed at the point where the caller has reached the end of the tuple.
+     * If the caller has consumed too few/too many bytes, this will be reported
+     * in the call go `guard.done()`.
+     *
+     * @return out If the header of the tuple is invalid.
+     */
+    MOZ_MUST_USE JS::Result<Ok> enterTaggedTuple(BinKind& tag, BinTokenReaderMultipart::BinFields& fields, AutoTaggedTuple& guard);
+
+    /**
+     * Start reading an untagged tuple.
+     *
+     * @param guard (OUT) A guard, ensuring that we read the tuple correctly.
+     *
+     * The `guard` is dedicated to ensuring that reading the list has consumed
+     * exactly all the bytes from that tuple. The `guard` MUST therefore be
+     * destroyed at the point where the caller has reached the end of the tuple.
+     * If the caller has consumed too few/too many bytes, this will be reported
+     * in the call go `guard.done()`.
+     */
+    MOZ_MUST_USE JS::Result<Ok> enterUntaggedTuple(AutoTuple& guard);
+
+  private:
+    /**
+     * Read a single uint32_t.
+     */
+    MOZ_MUST_USE JS::Result<uint32_t> readInternalUint32();
+
+  private:
+    // A mapping grammar index => BinKind, as defined by the [GRAMMAR]
+    // section of the file. Populated during readHeader().
+    Vector<BinKind> grammarTable_;
+
+    // A mapping string index => BinVariant as extracted from the [STRINGS]
+    // section of the file. Populated lazily.
+    js::HashMap<uint32_t, BinVariant, DefaultHasher<uint32_t>, SystemAllocPolicy> variantsTable_;
+
+    // A mapping index => JSAtom, as defined by the [STRINGS]
+    // section of the file. Populated during readHeader().
+    using AtomVector = GCVector<JSAtom*>;
+    JS::Rooted<AtomVector> atomsTable_;
+
+    // The same mapping, but as CharSlice. Populated during readHeader().
+    // The slices are into the source buffer.
+    Vector<Chars> slicesTable_;
+
+    const uint8_t* posBeforeTree_;
+
+    BinTokenReaderMultipart(const BinTokenReaderMultipart&) = delete;
+    BinTokenReaderMultipart(BinTokenReaderMultipart&&) = delete;
+    BinTokenReaderMultipart& operator=(BinTokenReaderMultipart&) = delete;
+
+  public:
+    // The following classes are used whenever we encounter a tuple/tagged tuple/list
+    // to make sure that:
+    //
+    // - if the construct "knows" its byte length, we have exactly consumed all
+    //   the bytes (otherwise, this means that the file is corrupted, perhaps on
+    //   purpose, so we need to reject the stream);
+    // - if the construct has a footer, once we are done reading it, we have
+    //   reached the footer (this is to aid with debugging).
+    //
+    // In either case, the caller MUST call method `done()` of the guard once
+    // it is done reading the tuple/tagged tuple/list, to report any pending error.
+
+    // Base class used by other Auto* classes.
+    class MOZ_STACK_CLASS AutoBase
+    {
+      protected:
+        explicit AutoBase(BinTokenReaderMultipart& reader);
+        ~AutoBase();
+
+        // Raise an error if we are not in the expected position.
+        MOZ_MUST_USE JS::Result<Ok> checkPosition(const uint8_t* expectedPosition);
+
+        friend BinTokenReaderMultipart;
+        void init();
+
+        // Set to `true` if `init()` has been called. Reset to `false` once
+        // all conditions have been checked.
+        bool initialized_;
+        BinTokenReaderMultipart& reader_;
+    };
+
+    // Guard class used to ensure that `enterList` is used properly.
+    class MOZ_STACK_CLASS AutoList : public AutoBase
+    {
+      public:
+        explicit AutoList(BinTokenReaderMultipart& reader);
+
+        // Check that we have properly read to the end of the list.
+        MOZ_MUST_USE JS::Result<Ok> done();
+
+      protected:
+        friend BinTokenReaderMultipart;
+        void init();
+    };
+
+    // Guard class used to ensure that `enterTaggedTuple` is used properly.
+    class MOZ_STACK_CLASS AutoTaggedTuple : public AutoBase
+    {
+      public:
+        explicit AutoTaggedTuple(BinTokenReaderMultipart& reader);
+
+        // Check that we have properly read to the end of the tuple.
+        MOZ_MUST_USE JS::Result<Ok> done();
+    };
+
+    // Guard class used to ensure that `readTuple` is used properly.
+    class MOZ_STACK_CLASS AutoTuple : public AutoBase
+    {
+      public:
+        explicit AutoTuple(BinTokenReaderMultipart& reader);
+
+        // Check that we have properly read to the end of the tuple.
+        MOZ_MUST_USE JS::Result<Ok> done();
+    };
+
+    // Compare a `Chars` and a string literal (ONLY a string literal).
+    template <size_t N>
+    static bool equals(const Chars& left, const char (&right)[N]) {
+        MOZ_ASSERT(N > 0);
+        MOZ_ASSERT(right[N - 1] == 0);
+        if (left.byteLen_ + 1 /* implicit NUL */ != N)
+            return false;
+
+        if (!std::equal(left.start_, left.start_ + left.byteLen_, right))
+          return false;
+
+        return true;
+    }
+
+    template<size_t N>
+    static JS::Result<Ok, JS::Error&>
+    checkFields(const BinKind kind, const BinFields& actual, const BinField (&expected)[N]) {
+      // Not implemented in this tokenizer.
+      return Ok();
+    }
+
+    static JS::Result<Ok, JS::Error&>
+    checkFields0(const BinKind kind, const BinFields& actual) {
+      // Not implemented in this tokenizer.
+      return Ok();
+    }
+};
+
+} // namespace frontend
+} // namespace js
+
+#endif // frontend_BinTokenReaderMultipart_h
--- a/js/src/moz.build
+++ b/js/src/moz.build
@@ -699,25 +699,27 @@ if CONFIG['NIGHTLY_BUILD']:
 # An experiment we want to run on Nightly and early Beta: Can we change the JS
 # representation of an exported global from the global's value to an instance
 # of WebAssembly.Global without breaking existing wasm content?
 #
 # Additionally guarded by EARLY_BETA_OR_EARLIER in the code.
 DEFINES['ENABLE_WASM_GLOBAL'] = True
 
 if CONFIG['JS_BUILD_BINAST']:
-    # Using SOURCES as UNIFIED_SOURCES causes mysterious bugs on 32-bit platforms.
+    # Using SOURCES, as UNIFIED_SOURCES causes mysterious bugs on 32-bit platforms.
     # These parts of BinAST are designed only to test evolutions of the
     # specification.
     SOURCES += ['frontend/BinTokenReaderTester.cpp']
     # These parts of BinAST should eventually move to release.
     SOURCES += [
         'frontend/BinSource-auto.cpp',
         'frontend/BinSource.cpp',
-        'frontend/BinToken.cpp'
+        'frontend/BinToken.cpp',
+        'frontend/BinTokenReaderBase.cpp',
+        'frontend/BinTokenReaderMultipart.cpp',
     ]
 
     # Instrument BinAST files for fuzzing as we have a fuzzing target for BinAST.
     if CONFIG['FUZZING_INTERFACES'] and CONFIG['LIBFUZZER']:
         SOURCES['frontend/BinSource.cpp'].flags += libfuzzer_flags_cmp
         SOURCES['frontend/BinToken.cpp'].flags += libfuzzer_flags_cmp
         SOURCES['frontend/BinTokenReaderTester.cpp'].flags += libfuzzer_flags_cmp