Handle early pause, cancel for parallel requests.

content/browser/download/parallel_download_job.h

 // Copyright 2017 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef CONTENT_BROWSER_DOWNLOAD_PARALLEL_DOWNLOAD_JOB_H_
 #define CONTENT_BROWSER_DOWNLOAD_PARALLEL_DOWNLOAD_JOB_H_
 
 #include <memory>
 #include <unordered_map>
 #include <vector>
@@ skipping 21 matching lines @@
   // DownloadJobImpl implementation.
   void Start() override;
   void Cancel(bool user_cancel) override;
   void Pause() override;
   void Resume(bool resume_request) override;
 
  protected:
   // Virtual for testing.
   virtual int GetParallelRequestCount() const;
 
+  using WorkerMap =
+      std::unordered_map<int64_t, std::unique_ptr<DownloadWorker>>;
+
+  // Map from the offset position of the slice to the worker that downloads the
+  // slice.
+  WorkerMap workers_;
+
  private:
   friend class ParallelDownloadJobTest;
 
-  using WorkerMap =
-      std::unordered_map<int64_t, std::unique_ptr<DownloadWorker>>;
-
   // DownloadWorker::Delegate implementation.
   void OnByteStreamReady(
       DownloadWorker* worker,
       std::unique_ptr<ByteStreamReader> stream_reader) override;
   void OnServerResponseError(DownloadWorker* worker,
                              DownloadInterruptReason reason) override;
+  bool IsPaused() const override;

[qinmin, 2017/03/22 05:50:29]

nit: worker can remember whether pause() is called or not, no need to ask the
delegate

[xingliu, 2017/03/22 23:54:06]

We may pause/cancel before the worker is constructed.

Alternatively we need to pass the current state to worker's constructor, which
looks more weird. I think duplicate states can easily introduce bugs.

This is caught in the unit test, when I move the is_pause into the worker.

[qinmin, 2017/03/24 19:52:58]

wait, if we pause/cancel before worker is constructed, why do we create the
workers.

+  bool IsCanceled() const override;
 
   // Build parallel requests after a delay, to effectively measure the single
   // stream bandwidth.
   void BuildParallelRequestAfterDelay();
 
   // Build parallel requests to download. This function is the entry point for
   // all parallel downloads.
   void BuildParallelRequests();
 
   // Build one http request for each slice from the second slice.
   // The first slice represents the original request.
   void ForkSubRequests(const DownloadItem::ReceivedSlices& slices_to_download);
 
   // Create one range request, virtual for testing.
   virtual void CreateRequest(int64_t offset, int64_t length);
 
   // Information about the initial request when download is started.
   int64_t initial_request_offset_;
 
   // The length of the response body of the original request. May be less than
   // the size of the target file if the request starts from non-zero offset.
   int64_t content_length_;
 
-  // Map from the offset position of the slice to the worker that downloads the
-  // slice.
-  WorkerMap workers_;
-
   // Used to send parallel requests after a delay based on Finch config.
   base::OneShotTimer timer_;
 
   // If we have sent parallel requests.
   bool requests_sent_;
 
   DISALLOW_COPY_AND_ASSIGN(ParallelDownloadJob);
 };
 
 }  //  namespace content
 
 #endif  // CONTENT_BROWSER_DOWNLOAD_PARALLEL_DOWNLOAD_JOB_H_