]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/downloader/dash.py
debian/control: Add 'libfribidi-bin | bidiv' to Suggests.
[youtubedl] / youtube_dl / downloader / dash.py
index 7491fdad857af2f36433b8991b6208fbd9cef99e..c6d674bc6c0690e00dee3c70879a97574c3911f3 100644 (file)
@@ -2,6 +2,10 @@ from __future__ import unicode_literals
 
 from .fragment import FragmentFD
 from ..compat import compat_urllib_error
 
 from .fragment import FragmentFD
 from ..compat import compat_urllib_error
+from ..utils import (
+    DownloadError,
+    urljoin,
+)
 
 
 class DashSegmentsFD(FragmentFD):
 
 
 class DashSegmentsFD(FragmentFD):
@@ -12,12 +16,13 @@ class DashSegmentsFD(FragmentFD):
     FD_NAME = 'dashsegments'
 
     def real_download(self, filename, info_dict):
     FD_NAME = 'dashsegments'
 
     def real_download(self, filename, info_dict):
-        segments = info_dict['fragments'][:1] if self.params.get(
+        fragment_base_url = info_dict.get('fragment_base_url')
+        fragments = info_dict['fragments'][:1] if self.params.get(
             'test', False) else info_dict['fragments']
 
         ctx = {
             'filename': filename,
             'test', False) else info_dict['fragments']
 
         ctx = {
             'filename': filename,
-            'total_frags': len(segments),
+            'total_frags': len(fragments),
         }
 
         self._prepare_and_start_frag_download(ctx)
         }
 
         self._prepare_and_start_frag_download(ctx)
@@ -26,7 +31,7 @@ class DashSegmentsFD(FragmentFD):
         skip_unavailable_fragments = self.params.get('skip_unavailable_fragments', True)
 
         frag_index = 0
         skip_unavailable_fragments = self.params.get('skip_unavailable_fragments', True)
 
         frag_index = 0
-        for i, segment in enumerate(segments):
+        for i, fragment in enumerate(fragments):
             frag_index += 1
             if frag_index <= ctx['fragment_index']:
                 continue
             frag_index += 1
             if frag_index <= ctx['fragment_index']:
                 continue
@@ -36,7 +41,11 @@ class DashSegmentsFD(FragmentFD):
             count = 0
             while count <= fragment_retries:
                 try:
             count = 0
             while count <= fragment_retries:
                 try:
-                    success, frag_content = self._download_fragment(ctx, segment['url'], info_dict)
+                    fragment_url = fragment.get('url')
+                    if not fragment_url:
+                        assert fragment_base_url
+                        fragment_url = urljoin(fragment_base_url, fragment['path'])
+                    success, frag_content = self._download_fragment(ctx, fragment_url, info_dict)
                     if not success:
                         return False
                     self._append_fragment(ctx, frag_content)
                     if not success:
                         return False
                     self._append_fragment(ctx, frag_content)
@@ -44,13 +53,21 @@ class DashSegmentsFD(FragmentFD):
                 except compat_urllib_error.HTTPError as err:
                     # YouTube may often return 404 HTTP error for a fragment causing the
                     # whole download to fail. However if the same fragment is immediately
                 except compat_urllib_error.HTTPError as err:
                     # YouTube may often return 404 HTTP error for a fragment causing the
                     # whole download to fail. However if the same fragment is immediately
-                    # retried with the same request data this usually succeeds (1-2 attemps
+                    # retried with the same request data this usually succeeds (1-2 attempts
                     # is usually enough) thus allowing to download the whole file successfully.
                     # To be future-proof we will retry all fragments that fail with any
                     # HTTP error.
                     count += 1
                     if count <= fragment_retries:
                         self.report_retry_fragment(err, frag_index, count, fragment_retries)
                     # is usually enough) thus allowing to download the whole file successfully.
                     # To be future-proof we will retry all fragments that fail with any
                     # HTTP error.
                     count += 1
                     if count <= fragment_retries:
                         self.report_retry_fragment(err, frag_index, count, fragment_retries)
+                except DownloadError:
+                    # Don't retry fragment if error occurred during HTTP downloading
+                    # itself since it has own retry settings
+                    if not fatal:
+                        self.report_skip_fragment(frag_index)
+                        break
+                    raise
+
             if count > fragment_retries:
                 if not fatal:
                     self.report_skip_fragment(frag_index)
             if count > fragment_retries:
                 if not fatal:
                     self.report_skip_fragment(frag_index)