Adding ChunkDemuxer implementation.

media/filters/webm_parser.cc

+// Copyright (c) 2011 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.
+
+#include "media/filters/webm_parser.h"
+

[scherkus (not reviewing), 2011/06/22 17:31:09]

I believe you're doing ebml parsing here, so how about a link to a wikipedia
page or something so people don't need to reverse engineer what you're up here

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+#include <iomanip>
+
+#include "base/logging.h"
+
+namespace media {
+
+static const int kMaxLevelDepth = 6;

[scherkus (not reviewing), 2011/06/22 17:31:09]

docs?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+
+WebMParserClient::~WebMParserClient() {}

[scherkus (not reviewing), 2011/06/22 17:31:09]

you can actually inline this in the .h if you want (the interface class doesn't
have any members, etc... so inlining the dtor is a no-op)

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

If I remember correctly there was this pass over all the code to remove inline
virtual functions. I was just making this code consistent with that.

+
+int webm_parse_num(const uint8* buf, int size, int max_bytes,

[scherkus (not reviewing), 2011/06/22 17:31:09]

docs?

I don't think we need to document the bit-twiddling going on below, but a solid
one-sentence should be enough

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+                   bool mask_first_byte, int64* num) {
+  DCHECK(buf);
+  DCHECK(num);
+
+  if (size <= 0)
+    return -1;
+
+  int mask = 0x80;
+  uint8 ch = buf[0];
+  int extra_bytes = -1;
+  for (int i = 0; i < max_bytes; ++i) {
+    if ((ch & mask) == mask) {
+      *num = mask_first_byte ? ch & ~mask : ch;
+      extra_bytes = i;
+      break;
+    }
+    mask >>= 1;
+  }
+
+  if ((extra_bytes == -1) || ((1 + extra_bytes) > size))
+    return -1;
+
+  int bytes_used = 1;
+
+  for (int i = 0; i < extra_bytes; ++i)
+    *num = (*num << 8) | (0xff & buf[bytes_used++]);
+
+  return bytes_used;
+}
+
+int webm_parse_element_header(const uint8* buf, int size,
+                              int* id, int64* element_size) {

[scherkus (not reviewing), 2011/06/22 17:31:09]

docs

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+  DCHECK(buf);
+  DCHECK_GE(size, 0);
+  DCHECK(id);
+  DCHECK(element_size);
+
+  if (size == 0)
+    return 0;
+
+  int64 tmp;
+  int num_id_bytes = webm_parse_num(buf, size, 4, false, &tmp);
+
+  if (num_id_bytes <= 0)
+    return num_id_bytes;
+
+  *id = (int)tmp;

[scherkus (not reviewing), 2011/06/22 17:31:09]

static_cast

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+
+  int num_sz_bytes = webm_parse_num(buf + num_id_bytes,

[scherkus (not reviewing), 2011/06/22 17:31:09]

sz?

I'm guessing it's size but I see a lot of other variables with "size" so let's
go with that

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+                                    size - num_id_bytes,
+                                    8, true, &tmp);
+
+  if (num_sz_bytes <= 0)
+    return num_sz_bytes;
+
+  *element_size = tmp;
+  return num_id_bytes + num_sz_bytes;
+}
+
+enum ElementType {
+  LIST,
+  UINT,
+  FLOAT,
+  BINARY,
+  STRING,
+  SBLOCK,
+  SKIP,
+};
+
+struct ElementIdInfo {
+  int level_;
+  ElementType type_;
+  int id_;
+};
+
+static const ElementIdInfo kClusterIds[] = {
+  {2, UINT, kWebMIdTimecode},
+  {2, SBLOCK, kWebMIdSimpleBlock},
+  {2, LIST, kWebMIdBlockGroup},
+};
+
+static const ElementIdInfo kInfoIds[] = {
+  {2, SKIP, kWebMIdSegmentUID},
+  {2, UINT, kWebMIdTimecodeScale},
+  {2, FLOAT, kWebMIdDuration},
+  {2, SKIP, kWebMIdDateUTC},
+  {2, SKIP, kWebMIdTitle},
+  {2, SKIP, kWebMIdMuxingApp},
+  {2, SKIP, kWebMIdWritingApp},
+};
+
+static const ElementIdInfo kTracksIds[] = {
+  {2, LIST, kWebMIdTrackEntry},
+};
+
+static const ElementIdInfo kTrackEntryIds[] = {
+  {3, UINT, kWebMIdTrackNumber},
+  {3, SKIP, kWebMIdTrackUID},
+  {3, UINT, kWebMIdTrackType},
+  {3, SKIP, kWebMIdFlagEnabled},
+  {3, SKIP, kWebMIdFlagDefault},
+  {3, SKIP, kWebMIdFlagForced},
+  {3, UINT, kWebMIdFlagLacing},
+  {3, UINT, kWebMIdDefaultDuration},
+  {3, SKIP, kWebMIdName},
+  {3, SKIP, kWebMIdLanguage},
+  {3, STRING, kWebMIdCodecID},
+  {3, BINARY, kWebMIdCodecPrivate},
+  {3, SKIP, kWebMIdCodecName},
+  {3, LIST, kWebMIdVideo},
+  {3, LIST, kWebMIdAudio},
+};
+
+static const ElementIdInfo kVideoIds[] = {
+  {4, SKIP, kWebMIdFlagInterlaced},
+  {4, SKIP, kWebMIdStereoMode},
+  {4, UINT, kWebMIdPixelWidth},
+  {4, UINT, kWebMIdPixelHeight},
+  {4, SKIP, kWebMIdPixelCropBottom},
+  {4, SKIP, kWebMIdPixelCropTop},
+  {4, SKIP, kWebMIdPixelCropLeft},
+  {4, SKIP, kWebMIdPixelCropRight},
+  {4, SKIP, kWebMIdDisplayWidth},
+  {4, SKIP, kWebMIdDisplayHeight},
+  {4, SKIP, kWebMIdDisplayUnit},
+  {4, SKIP, kWebMIdAspectRatioType},
+};
+
+static const ElementIdInfo kAudioIds[] = {
+  {4, SKIP, kWebMIdSamplingFrequency},
+  {4, SKIP, kWebMIdOutputSamplingFrequency},
+  {4, UINT, kWebMIdChannels},
+  {4, SKIP, kWebMIdBitDepth},
+};
+
+static const ElementIdInfo kInfoAndTracksOnly[] = {
+  {1, LIST, kWebMIdInfo},
+  {1, LIST, kWebMIdTracks},
+};
+
+static const ElementIdInfo kClustersOnly[] = {
+  {1, LIST, kWebMIdCluster},
+};
+

[scherkus (not reviewing), 2011/06/22 17:31:09]

nit: remove extra blank lines

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+
+
+struct ListElementInfo {
+  int id_;
+  const ElementIdInfo* id_info_;
+  int id_info_size_;
+};
+
+const ListElementInfo kListElementInfo[] = {
+  { kWebMIdCluster,    kClusterIds,    sizeof(kClusterIds) },
+  { kWebMIdInfo,       kInfoIds,       sizeof(kInfoIds) },
+  { kWebMIdTracks,     kTracksIds,     sizeof(kTracksIds) },
+  { kWebMIdTrackEntry, kTrackEntryIds, sizeof(kTrackEntryIds) },
+  { kWebMIdVideo,      kVideoIds,      sizeof(kVideoIds) },
+  { kWebMIdAudio,      kAudioIds,      sizeof(kAudioIds) },
+};
+
+const ElementIdInfo* webm_find_id_info(int id,
+                                       const ElementIdInfo* id_info,
+                                       int id_info_size) {
+  int count = id_info_size / sizeof(ElementIdInfo);

[scherkus (not reviewing), 2011/06/22 17:31:09]

nit: sizeof(*id_info)

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+  for (int i = 0; i < count; ++i) {
+    if (id == id_info[i].id_)
+      return &id_info[i];
+  }
+
+  return NULL;
+}
+
+const ListElementInfo* webm_find_list_info(int id) {
+  int count = sizeof(kListElementInfo) / sizeof(ListElementInfo);

[scherkus (not reviewing), 2011/06/22 17:31:09]

isn't this a contsant?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+  for (int i = 0; i < count; ++i) {
+    if (id == kListElementInfo[i].id_)
+      return &kListElementInfo[i];
+  }
+
+  return NULL;
+}
+
+int webm_parse_simple_block(WebMParserClient* client,
+                            const uint8* buf, int size) {
+  if (size < 4)
+    return -1;
+
+  // Return an error if the trackNum > 127. We just aren't
+  // going to support large track numbers right now.
+  if ((buf[0] & 0x80) != 0x80) {
+    VLOG(1) << "TrackNumber over 127 not supported";
+    return -1;
+  }
+
+  int track_num = buf[0] & 0x7f;
+  int timecode = buf[1] << 8 | buf[2];
+  int flags = buf[3] & 0xff;
+  int lacing = (flags >> 1) & 0x3;
+
+  if (lacing != 0) {
+    VLOG(1) << "Lacing " << lacing << " not supported yet.";
+    return -1;
+  }
+
+  const uint8* frame_data = buf + 4;
+  int frame_size = size - (frame_data - buf);
+  if (!client->OnSimpleBlock(track_num, timecode, flags,
+                             frame_data, frame_size)) {
+    return -1;
+  }
+
+  return size;
+}
+
+int webm_parse_elements(WebMParserClient* client,
+                        const ElementIdInfo* id_info,
+                        int id_info_size,
+                        const uint8* buf, int size, int level) {
+  DCHECK_GE(id_info_size, 0);
+  DCHECK_GE(size, 0);
+  DCHECK_GE(level, 0);
+
+  const uint8* pCur = buf;

[scherkus (not reviewing), 2011/06/22 17:31:09]

pCur -> cur ?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done. Old habits sneak back in sometimes. :)

+  int cur_size = size;
+  int used = 0;
+
+  if (level > kMaxLevelDepth)
+    return -1;
+
+  while (cur_size > 0) {
+    int id;
+    int64 element_size;
+    int res = webm_parse_element_header(pCur, cur_size,
+                                        &id, &element_size);
+
+    if (res < 0)
+      return res;
+
+    if (res == 0)
+      break;
+
+    pCur += res;
+    cur_size -= res;
+    used += res;
+
+    // Check to see if the element is larger than the remaining data.
+    if (element_size > cur_size)
+      return -1;
+
+    const ElementIdInfo* info = webm_find_id_info(id, id_info, id_info_size);
+
+    if (info == NULL) {
+      VLOG(1) << "No info for ID " << std::hex << id;
+
+      // TODO(acolwell): Change this to return -1 after the API has solidified.
+      // We don't want to allow elements we don't recognize.
+      pCur += element_size;
+      cur_size -= element_size;
+      used += element_size;
+      continue;
+    }
+
+    if (info->level_ != level) {
+      VLOG(1) << "ID " << std::hex << id << std::dec << " at level "
+              << level << " instead of " << info->level_;
+      return -1;
+    }
+
+    if (info->type_ == SBLOCK) {

[scherkus (not reviewing), 2011/06/22 17:31:09]

use switch/case + helper functions to parse each type?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+      int r2  = webm_parse_simple_block(client, pCur, element_size);
+      if (r2 <= 0)
+        return -1;
+
+      DCHECK_EQ(r2, element_size);
+    } else if (info->type_ == LIST) {
+      const ListElementInfo* list_info = webm_find_list_info(id);
+
+      if (!list_info) {
+        VLOG(1) << "Failed to find list info for ID " << std::hex << id;
+        return -1;
+      }
+
+      if (!client->OnListStart(id))
+        return -1;
+
+      int r2 = webm_parse_elements(client,
+                                   list_info->id_info_,
+                                   list_info->id_info_size_,
+                                   pCur, element_size,
+                                   level + 1);
+
+      if (r2 < 0)
+        return -1;
+
+      if (!client->OnListEnd(id))
+        return -1;
+
+      DCHECK_EQ(r2, element_size);
+    } else if (info->type_ == UINT) {
+      if ((element_size <= 0) || (element_size > 8))
+        return -1;
+
+      int64 tmp = 0;
+      for (int i = 0; i < element_size; ++i)

[scherkus (not reviewing), 2011/06/22 17:31:09]

endianess?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Added a comment indicating the integer is stored as big endian in the buffer.

+        tmp = (tmp << 8) | pCur[i];
+
+      if (!client->OnUInt(id, tmp))
+        return -1;
+    } else if (info->type_ == FLOAT) {
+

[scherkus (not reviewing), 2011/06/22 17:31:09]

nit: remove blank line

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+      if ((element_size != 4) && (element_size != 8))
+        return -1;
+
+      double value = -1;
+
+      int64 tmp = 0;
+      for (int i = 0; i < element_size; ++i)
+        tmp = (tmp << 8) | pCur[i];
+
+      if (element_size == 4) {
+        union {
+          int32 src;
+          float dst;
+        } tmp2;
+        tmp2.src = (int32)tmp;

[scherkus (not reviewing), 2011/06/22 17:31:09]

does ebml specify anything w.r.t. endianess?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Yes all fields are big endian. I've added a comment mentioning this.

+        value = tmp2.dst;
+      } else if (element_size == 8) {
+        union {
+          int64 src;
+          double dst;
+        } tmp2;

[scherkus (not reviewing), 2011/06/22 17:31:09]

ditto

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

Done.

+        tmp2.src = tmp;
+        value = tmp2.dst;
+      } else {
+        return -1;
+      }
+
+      if (!client->OnFloat(id, value))
+        return -1;
+    } else if (info->type_ == BINARY) {
+      if (!client->OnBinary(id, pCur, element_size))
+        return -1;
+    } else if (info->type_ == STRING) {
+      std::string s((const char*)pCur, element_size);

[scherkus (not reviewing), 2011/06/22 17:31:09]

reinterpret_cast<>

is the string null terminated or will the std::string ctor copy + do the proper
thing?

[acolwell GONE FROM CHROMIUM, 2011/06/23 16:51:28]

The string is not null terminated. The constructor will do the right thing.

+      if (!client->OnString(id, s))
+        return -1;
+    } else if (info->type_ != SKIP) {
+      VLOG(1) << "Unhandled id type " << info->type_;
+      return -1;
+    }
+
+    pCur += element_size;
+    cur_size -= element_size;
+    used += element_size;
+  }
+
+  return used;
+}
+
+int webm_parse_headers(WebMParserClient* client,
+                       const uint8* buf, int size) {
+  return webm_parse_elements(client,
+                             kInfoAndTracksOnly,
+                             sizeof(kInfoAndTracksOnly),
+                             buf, size, 1);
+}
+
+int webm_parse_cluster(WebMParserClient* client,
+                       const uint8* buf, int size) {
+  return webm_parse_elements(client,
+                             kClustersOnly,
+                             sizeof(kClustersOnly),
+                             buf, size, 1);
+}
+
+}  // namespace media