gecko-dev/parser/html/nsHtml5StreamParser.h

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* 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 nsHtml5StreamParser_h
#define nsHtml5StreamParser_h

#include "nsAutoPtr.h"
#include "nsCOMPtr.h"
#include "nsICharsetDetectionObserver.h"
#include "nsHtml5MetaScanner.h"
#include "mozilla/Encoding.h"
#include "mozilla/JapaneseDetector.h"
#include "nsHtml5TreeOpExecutor.h"
#include "nsHtml5OwningUTF16Buffer.h"
#include "nsIInputStream.h"
#include "mozilla/Mutex.h"
#include "mozilla/UniquePtr.h"
#include "nsHtml5AtomTable.h"
#include "nsHtml5Speculation.h"
#include "nsISerialEventTarget.h"
#include "nsITimer.h"
#include "nsICharsetDetector.h"
#include "mozilla/dom/DocGroup.h"
#include "mozilla/Buffer.h"

class nsHtml5Parser;

enum eParserMode {
  /**
   * Parse a document normally as HTML.
   */
  NORMAL,

  /**
   * View document as HTML source.
   */
  VIEW_SOURCE_HTML,

  /**
   * View document as XML source
   */
  VIEW_SOURCE_XML,

  /**
   * View document as plain text source
   */
  VIEW_SOURCE_PLAIN,

  /**
   * View document as plain text
   */
  PLAIN_TEXT,

  /**
   * Load as data (XHR)
   */
  LOAD_AS_DATA
};

enum eBomState {
  /**
   * BOM sniffing hasn't started.
   */
  BOM_SNIFFING_NOT_STARTED = 0,

  /**
   * BOM sniffing is ongoing, and the first byte of an UTF-16LE BOM has been
   * seen.
   */
  SEEN_UTF_16_LE_FIRST_BYTE = 1,

  /**
   * BOM sniffing is ongoing, and the first byte of an UTF-16BE BOM has been
   * seen.
   */
  SEEN_UTF_16_BE_FIRST_BYTE = 2,

  /**
   * BOM sniffing is ongoing, and the first byte of an UTF-8 BOM has been
   * seen.
   */
  SEEN_UTF_8_FIRST_BYTE = 3,

  /**
   * BOM sniffing is ongoing, and the first and second bytes of an UTF-8 BOM
   * have been seen.
   */
  SEEN_UTF_8_SECOND_BYTE = 4,

  /**
   * BOM sniffing was started but is now over for whatever reason.
   */
  BOM_SNIFFING_OVER = 5
};

enum eHtml5StreamState {
  STREAM_NOT_STARTED = 0,
  STREAM_BEING_READ = 1,
  STREAM_ENDED = 2
};

class nsHtml5StreamParser final : public nsICharsetDetectionObserver {
  template <typename T>
  using NotNull = mozilla::NotNull<T>;
  using Encoding = mozilla::Encoding;

  const uint32_t SNIFFING_BUFFER_SIZE = 1024;
  const uint32_t READ_BUFFER_SIZE = 1024;
  const uint32_t LOCAL_FILE_UTF_8_BUFFER_SIZE = 1024 * 1024 * 4;  // 4 MB

  friend class nsHtml5RequestStopper;
  friend class nsHtml5DataAvailable;
  friend class nsHtml5StreamParserContinuation;
  friend class nsHtml5TimerKungFu;
  friend class nsHtml5StreamParserPtr;

 public:
  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
  NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(nsHtml5StreamParser,
                                           nsICharsetDetectionObserver)

  nsHtml5StreamParser(nsHtml5TreeOpExecutor* aExecutor, nsHtml5Parser* aOwner,
                      eParserMode aMode);

  // Methods that nsHtml5StreamListener calls
  nsresult CheckListenerChain();

  nsresult OnStartRequest(nsIRequest* aRequest);

  nsresult OnDataAvailable(nsIRequest* aRequest, nsIInputStream* aInStream,
                           uint64_t aSourceOffset, uint32_t aLength);

  nsresult OnStopRequest(nsIRequest* aRequest, nsresult status);

  // nsICharsetDetectionObserver
  /**
   * Chardet calls this to report the detection result
   */
  NS_IMETHOD Notify(const char* aCharset, nsDetectionConfident aConf) override;

  // EncodingDeclarationHandler
  // https://hg.mozilla.org/projects/htmlparser/file/tip/src/nu/validator/htmlparser/common/EncodingDeclarationHandler.java
  /**
   * Tree builder uses this to report a late <meta charset>
   */
  bool internalEncodingDeclaration(nsHtml5String aEncoding);

  // Not from an external interface

  /**
   * Pass a buffer to the JapaneseDetector.
   */
  void FeedJapaneseDetector(mozilla::Span<const uint8_t> aBuffer, bool aLast);

  /**
   * Pass a buffer to the Japanese or Cyrillic detector as appropriate.
   */
  void FeedDetector(mozilla::Span<const uint8_t> aBuffer, bool aLast);

  /**
   *  Call this method once you've created a parser, and want to instruct it
   *  about what charset to load
   *
   *  @param   aEncoding the charset of a document
   *  @param   aCharsetSource the source of the charset
   */
  inline void SetDocumentCharset(NotNull<const Encoding*> aEncoding,
                                 int32_t aSource) {
    MOZ_ASSERT(mStreamState == STREAM_NOT_STARTED,
               "SetDocumentCharset called too late.");
    NS_ASSERTION(NS_IsMainThread(), "Wrong thread!");
    mEncoding = aEncoding;
    mCharsetSource = aSource;
  }

  inline void SetObserver(nsIRequestObserver* aObserver) {
    NS_ASSERTION(NS_IsMainThread(), "Wrong thread!");
    mObserver = aObserver;
  }

  nsresult GetChannel(nsIChannel** aChannel);

  /**
   * The owner parser must call this after script execution
   * when no scripts are executing and the document.written
   * buffer has been exhausted.
   */
  void ContinueAfterScripts(nsHtml5Tokenizer* aTokenizer,
                            nsHtml5TreeBuilder* aTreeBuilder, bool aLastWasCR);

  /**
   * Continues the stream parser if the charset switch failed.
   */
  void ContinueAfterFailedCharsetSwitch();

  void Terminate() {
    mozilla::MutexAutoLock autoLock(mTerminatedMutex);
    mTerminated = true;
  }

  void DropTimer();

  /**
   * Sets mEncoding and mCharsetSource appropriately for the XML View Source
   * case if aEncoding names a supported rough ASCII superset and sets
   * the mEncoding and mCharsetSource to the UTF-8 default otherwise.
   */
  void SetEncodingFromExpat(const char16_t* aEncoding);

  /**
   * Sets the URL for View Source title in case this parser ends up being
   * used for View Source. If aURL is a view-source: URL, takes the inner
   * URL. data: URLs are shown with an ellipsis instead of the actual data.
   */
  void SetViewSourceTitle(nsIURI* aURL);

 private:
  virtual ~nsHtml5StreamParser();

#ifdef DEBUG
  bool IsParserThread() { return mEventTarget->IsOnCurrentThread(); }
#endif

  void MarkAsBroken(nsresult aRv);

  /**
   * Marks the stream parser as interrupted. If you ever add calls to this
   * method, be sure to review Uninterrupt usage very, very carefully to
   * avoid having a previous in-flight runnable cancel your Interrupt()
   * call on the other thread too soon.
   */
  void Interrupt() {
    mozilla::MutexAutoLock autoLock(mTerminatedMutex);
    mInterrupted = true;
  }

  void Uninterrupt() {
    NS_ASSERTION(IsParserThread(), "Wrong thread!");
    mTokenizerMutex.AssertCurrentThreadOwns();
    // Not acquiring mTerminatedMutex because mTokenizerMutex is already
    // held at this point and is already stronger.
    mInterrupted = false;
  }

  /**
   * Flushes the tree ops from the tree builder and disarms the flush
   * timer.
   */
  void FlushTreeOpsAndDisarmTimer();

  void ParseAvailableData();

  void DoStopRequest();

  void DoDataAvailableBuffer(mozilla::Buffer<uint8_t>&& aBuffer);

  void DoDataAvailable(mozilla::Span<const uint8_t> aBuffer);

  static nsresult CopySegmentsToParser(nsIInputStream* aInStream,
                                       void* aClosure, const char* aFromSegment,
                                       uint32_t aToOffset, uint32_t aCount,
                                       uint32_t* aWriteCount);

  bool IsTerminatedOrInterrupted() {
    mozilla::MutexAutoLock autoLock(mTerminatedMutex);
    return mTerminated || mInterrupted;
  }

  bool IsTerminated() {
    mozilla::MutexAutoLock autoLock(mTerminatedMutex);
    return mTerminated;
  }

  /**
   * True when there is a Unicode decoder already
   */
  inline bool HasDecoder() { return !!mUnicodeDecoder; }

  /**
   * Push bytes from network when there is no Unicode decoder yet
   */
  nsresult SniffStreamBytes(mozilla::Span<const uint8_t> aFromSegment);

  /**
   * Push bytes from network when there is a Unicode decoder already
   */
  nsresult WriteStreamBytes(mozilla::Span<const uint8_t> aFromSegment);

  /**
   * Check whether every other byte in the sniffing buffer is zero.
   */
  void SniffBOMlessUTF16BasicLatin(mozilla::Span<const uint8_t> aFromSegment);

  /**
   * Write the start of the stream to detector.
   */
  void FinalizeSniffingWithDetector(mozilla::Span<const uint8_t> aFromSegment,
                                    uint32_t aCountToSniffingLimit, bool aEof);

  /**
   * <meta charset> scan failed. Try chardet if applicable. After this, the
   * the parser will have some encoding even if a last resolt fallback.
   *
   * @param aFromSegment The current network buffer
   * @param aCountToSniffingLimit The number of unfilled slots in
   *                              mSniffingBuffer
   * @param aEof true iff called upon end of stream
   */
  nsresult FinalizeSniffing(mozilla::Span<const uint8_t> aFromSegment,
                            uint32_t aCountToSniffingLimit, bool aEof);

  /**
   * Set up the Unicode decoder and write the sniffing buffer into it
   * followed by the current network buffer.
   *
   * @param aFromSegment The current network buffer
   */
  nsresult SetupDecodingAndWriteSniffingBufferAndCurrentSegment(
      mozilla::Span<const uint8_t> aFromSegment);

  /**
   * Initialize the Unicode decoder, mark the BOM as the source and
   * drop the sniffer.
   *
   * @param aDecoderCharsetName The name for the decoder's charset
   *                            (UTF-16BE, UTF-16LE or UTF-8; the BOM has
   *                            been swallowed)
   */
  nsresult SetupDecodingFromBom(NotNull<const Encoding*> aEncoding);

  /**
   * When speculatively decoding from file: URL as UTF-8, commit
   * to UTF-8 as the non-speculative encoding and start processing
   * the decoded data.
   */
  void CommitLocalFileToUTF8();

  /**
   * When speculatively decoding from file: URL as UTF-8, redecode
   * using fallback and then continue normally with the fallback.
   */
  void ReDecodeLocalFile();

  /**
   * Become confident or resolve and encoding name to its preferred form.
   * @param aEncoding the value of an internal encoding decl. Acts as an
   *                  out param, too, when the method returns true.
   * @return true if the parser needs to start using the new value of
   *         aEncoding and false if the parser became confident or if
   *         the encoding name did not specify a usable encoding
   */
  const Encoding* PreferredForInternalEncodingDecl(const nsACString& aEncoding);

  /**
   * Callback for mFlushTimer.
   */
  static void TimerCallback(nsITimer* aTimer, void* aClosure);

  /**
   * Parser thread entry point for (maybe) flushing the ops and posting
   * a flush runnable back on the main thread.
   */
  void TimerFlush();

  /**
   * Called when speculation fails.
   */
  void MaybeDisableFutureSpeculation() { mSpeculationFailureCount++; }

  /**
   * Used to check whether we're getting too many speculation failures and
   * should just stop trying.  The 100 is picked pretty randomly to be not too
   * small (so most pages are not affected) but small enough that we don't end
   * up with failed speculations over and over in pathological cases.
   */
  bool IsSpeculationEnabled() { return mSpeculationFailureCount < 100; }

  /**
   * Dispatch an event to a Quantum DOM main thread-ish thread.
   * (Not the parser thread.)
   */
  nsresult DispatchToMain(already_AddRefed<nsIRunnable>&& aRunnable);

  /**
   * Notify any devtools listeners about content newly received for parsing.
   */
  inline void OnNewContent(mozilla::Span<const char16_t> aData);

  /**
   * Notify any devtools listeners after all parse content has been received.
   */
  inline void OnContentComplete();

  nsCOMPtr<nsIRequest> mRequest;
  nsCOMPtr<nsIRequestObserver> mObserver;

  /**
   * The document title to use if this turns out to be a View Source parser.
   */
  nsCString mViewSourceTitle;

  /**
   * The Unicode decoder
   */
  mozilla::UniquePtr<mozilla::Decoder> mUnicodeDecoder;

  /**
   * The buffer for sniffing the character encoding
   */
  mozilla::UniquePtr<uint8_t[]> mSniffingBuffer;

  /**
   * The number of meaningful bytes in mSniffingBuffer
   */
  uint32_t mSniffingLength;

  /**
   * BOM sniffing state
   */
  eBomState mBomState;

  /**
   * <meta> prescan implementation
   */
  nsAutoPtr<nsHtml5MetaScanner> mMetaScanner;

  // encoding-related stuff
  /**
   * The source (confidence) of the character encoding in use
   */
  int32_t mCharsetSource;

  /**
   * The character encoding in use
   */
  NotNull<const Encoding*> mEncoding;

  /**
   * Whether the Cyrillic or Japanese detector should still be fed.
   */
  bool mFeedChardet;

  /**
   * Whether reparse is forbidden
   */
  bool mReparseForbidden;

  // Portable parser objects
  /**
   * The first buffer in the pending UTF-16 buffer queue
   */
  RefPtr<nsHtml5OwningUTF16Buffer> mFirstBuffer;

  /**
   * The last buffer in the pending UTF-16 buffer queue
   */
  nsHtml5OwningUTF16Buffer*
      mLastBuffer;  // weak ref; always points to
                    // a buffer of the size
                    // NS_HTML5_STREAM_PARSER_READ_BUFFER_SIZE

  /**
   * The tree operation executor
   */
  nsHtml5TreeOpExecutor* mExecutor;

  /**
   * The same as mExecutor->mDocument->mDocGroup.
   */
  RefPtr<mozilla::dom::DocGroup> mDocGroup;

  /**
   * The HTML5 tree builder
   */
  nsAutoPtr<nsHtml5TreeBuilder> mTreeBuilder;

  /**
   * The HTML5 tokenizer
   */
  nsAutoPtr<nsHtml5Tokenizer> mTokenizer;

  /**
   * Makes sure the main thread can't mess the tokenizer state while it's
   * tokenizing. This mutex also protects the current speculation.
   */
  mozilla::Mutex mTokenizerMutex;

  /**
   * The scoped atom table
   */
  nsHtml5AtomTable mAtomTable;

  /**
   * The owner parser.
   */
  RefPtr<nsHtml5Parser> mOwner;

  /**
   * Whether the last character tokenized was a carriage return (for CRLF)
   */
  bool mLastWasCR;

  /**
   * For tracking stream life cycle
   */
  eHtml5StreamState mStreamState;

  /**
   * Whether we are speculating.
   */
  bool mSpeculating;

  /**
   * Whether the tokenizer has reached EOF. (Reset when stream rewinded.)
   */
  bool mAtEOF;

  /**
   * The speculations. The mutex protects the nsTArray itself.
   * To access the queue of current speculation, mTokenizerMutex must be
   * obtained.
   * The current speculation is the last element
   */
  nsTArray<nsAutoPtr<nsHtml5Speculation>> mSpeculations;
  mozilla::Mutex mSpeculationMutex;

  /**
   * Number of times speculation has failed for this parser.
   */
  uint32_t mSpeculationFailureCount;

  /**
   * Number of bytes already buffered into mBufferedLocalFileData.
   * Never counts above LOCAL_FILE_UTF_8_BUFFER_SIZE.
   */
  uint32_t mLocalFileBytesBuffered;

  nsTArray<mozilla::Buffer<uint8_t>> mBufferedLocalFileData;

  /**
   * True to terminate early; protected by mTerminatedMutex
   */
  bool mTerminated;
  bool mInterrupted;
  mozilla::Mutex mTerminatedMutex;

  /**
   * The thread this stream parser runs on.
   */
  nsCOMPtr<nsISerialEventTarget> mEventTarget;

  nsCOMPtr<nsIRunnable> mExecutorFlusher;

  nsCOMPtr<nsIRunnable> mLoadFlusher;

  /**
   * The Cyrillic detector if enabled.
   */
  nsCOMPtr<nsICharsetDetector> mChardet;

  /**
   * The Japanese detector.
   */
  mozilla::UniquePtr<mozilla::JapaneseDetector> mJapaneseDetector;

  /**
   * Whether the initial charset source was kCharsetFromParentFrame
   */
  bool mInitialEncodingWasFromParentFrame;

  bool mHasHadErrors;

  /**
   * If true, we are decoding a local file that lacks an encoding
   * declaration as UTF-8 and we are not tokenizing yet.
   */
  bool mDecodingLocalFileAsUTF8;

  /**
   * Timer for flushing tree ops once in a while when not speculating.
   */
  nsCOMPtr<nsITimer> mFlushTimer;

  /**
   * Mutex for protecting access to mFlushTimer (but not for the two
   * mFlushTimerFoo booleans below).
   */
  mozilla::Mutex mFlushTimerMutex;

  /**
   * Keeps track whether mFlushTimer has been armed. Unfortunately,
   * nsITimer doesn't enable querying this from the timer itself.
   */
  bool mFlushTimerArmed;

  /**
   * False initially and true after the timer has fired at least once.
   */
  bool mFlushTimerEverFired;

  /**
   * Whether the parser is doing a normal parse, view source or plain text.
   */
  eParserMode mMode;

  /**
   * Whether the parser should not sniff the content type.
   */
  bool mSkipContentSniffing;

  /**
   * If the associated docshell is being watched by the devtools, this is
   * set to the URI associated with the parse. All parse data is sent to the
   * devtools, along with this URI. This URI is cleared out after the parse has
   * been marked as completed.
   */
  nsCOMPtr<nsIURI> mURIToSendToDevtools;

  /**
   * If content is being sent to the devtools, an encoded UUID for the parser.
   */
  nsString mUUIDForDevtools;
};

#endif  // nsHtml5StreamParser_h