]> Raphaƫl G. Git Repositories - youtubedl/blobdiff - youtube_dl/FileDownloader.py
debian/control: Bump build-dep on debhelper to >= 9.
[youtubedl] / youtube_dl / FileDownloader.py
index b43acd19b42a1b7ad8b4c8329657864b35b365b3..49f3a871261d3816ec0b537ae60644691dc52f19 100644 (file)
@@ -4,8 +4,10 @@
 from __future__ import absolute_import
 
 import math
 from __future__ import absolute_import
 
 import math
+import io
 import os
 import re
 import os
 import re
+import shutil
 import socket
 import subprocess
 import sys
 import socket
 import subprocess
 import sys
@@ -16,6 +18,7 @@ if os.name == 'nt':
     import ctypes
 
 from .utils import *
     import ctypes
 
 from .utils import *
+from .InfoExtractors import get_info_extractor
 
 
 class FileDownloader(object):
 
 
 class FileDownloader(object):
@@ -51,6 +54,7 @@ class FileDownloader(object):
     quiet:             Do not print messages to stdout.
     forceurl:          Force printing final URL.
     forcetitle:        Force printing title.
     quiet:             Do not print messages to stdout.
     forceurl:          Force printing final URL.
     forcetitle:        Force printing title.
+    forceid:           Force printing ID.
     forcethumbnail:    Force printing thumbnail URL.
     forcedescription:  Force printing description.
     forcefilename:     Force printing final filename.
     forcethumbnail:    Force printing thumbnail URL.
     forcedescription:  Force printing description.
     forcefilename:     Force printing final filename.
@@ -77,9 +81,18 @@ class FileDownloader(object):
     updatetime:        Use the Last-modified header to set output file timestamps.
     writedescription:  Write the video description to a .description file
     writeinfojson:     Write the video description to a .info.json file
     updatetime:        Use the Last-modified header to set output file timestamps.
     writedescription:  Write the video description to a .description file
     writeinfojson:     Write the video description to a .info.json file
-    writesubtitles:    Write the video subtitles to a .srt file
+    writethumbnail:    Write the thumbnail image to a file
+    writesubtitles:    Write the video subtitles to a file
+    allsubtitles:      Downloads all the subtitles of the video
+    listsubtitles:     Lists all available subtitles for the video
+    subtitlesformat:   Subtitle format [sbv/srt] (default=srt)
     subtitleslang:     Language of the subtitles to download
     test:              Download only first bytes to test the downloader.
     subtitleslang:     Language of the subtitles to download
     test:              Download only first bytes to test the downloader.
+    keepvideo:         Keep the video file after post-processing
+    min_filesize:      Skip files smaller than this size
+    max_filesize:      Skip files larger than this size
+    daterange:         A DateRange object, download only if the upload_date is in the range.
+    skip_download:     Skip the actual download of the video file
     """
 
     params = None
     """
 
     params = None
@@ -93,13 +106,14 @@ class FileDownloader(object):
         """Create a FileDownloader object with the given options."""
         self._ies = []
         self._pps = []
         """Create a FileDownloader object with the given options."""
         self._ies = []
         self._pps = []
+        self._progress_hooks = []
         self._download_retcode = 0
         self._num_downloads = 0
         self._screen_file = [sys.stdout, sys.stderr][params.get('logtostderr', False)]
         self.params = params
 
         if '%(stitle)s' in self.params['outtmpl']:
         self._download_retcode = 0
         self._num_downloads = 0
         self._screen_file = [sys.stdout, sys.stderr][params.get('logtostderr', False)]
         self.params = params
 
         if '%(stitle)s' in self.params['outtmpl']:
-            self.to_stderr(u'WARNING: %(stitle)s is deprecated. Use the %(title)s and the --restrict-filenames flag(which also secures %(uploader)s et al) instead.')
+            self.report_warning(u'%(stitle)s is deprecated. Use the %(title)s and the --restrict-filenames flag(which also secures %(uploader)s et al) instead.')
 
     @staticmethod
     def format_bytes(bytes):
 
     @staticmethod
     def format_bytes(bytes):
@@ -111,7 +125,7 @@ class FileDownloader(object):
             exponent = 0
         else:
             exponent = int(math.log(bytes, 1024.0))
             exponent = 0
         else:
             exponent = int(math.log(bytes, 1024.0))
-        suffix = 'bkMGTPEZY'[exponent]
+        suffix = ['B','KiB','MiB','GiB','TiB','PiB','EiB','ZiB','YiB'][exponent]
         converted = float(bytes) / float(1024 ** exponent)
         return '%.2f%s' % (converted, suffix)
 
         converted = float(bytes) / float(1024 ** exponent)
         return '%.2f%s' % (converted, suffix)
 
@@ -203,27 +217,66 @@ class FileDownloader(object):
             # already of type unicode()
             ctypes.windll.kernel32.SetConsoleTitleW(ctypes.c_wchar_p(message))
         elif 'TERM' in os.environ:
             # already of type unicode()
             ctypes.windll.kernel32.SetConsoleTitleW(ctypes.c_wchar_p(message))
         elif 'TERM' in os.environ:
-            sys.stderr.write('\033]0;%s\007' % message.encode(preferredencoding()))
+            self.to_screen('\033]0;%s\007' % message, skip_eol=True)
 
     def fixed_template(self):
         """Checks if the output template is fixed."""
         return (re.search(u'(?u)%\\(.+?\\)s', self.params['outtmpl']) is None)
 
 
     def fixed_template(self):
         """Checks if the output template is fixed."""
         return (re.search(u'(?u)%\\(.+?\\)s', self.params['outtmpl']) is None)
 
-    def trouble(self, message=None):
+    def trouble(self, message=None, tb=None):
         """Determine action to take when a download problem appears.
 
         Depending on if the downloader has been configured to ignore
         download errors or not, this method may throw an exception or
         not when errors are found, after printing the message.
         """Determine action to take when a download problem appears.
 
         Depending on if the downloader has been configured to ignore
         download errors or not, this method may throw an exception or
         not when errors are found, after printing the message.
+
+        tb, if given, is additional traceback information.
         """
         if message is not None:
             self.to_stderr(message)
         if self.params.get('verbose'):
         """
         if message is not None:
             self.to_stderr(message)
         if self.params.get('verbose'):
-            self.to_stderr(u''.join(traceback.format_list(traceback.extract_stack())))
+            if tb is None:
+                if sys.exc_info()[0]:  # if .trouble has been called from an except block
+                    tb = u''
+                    if hasattr(sys.exc_info()[1], 'exc_info') and sys.exc_info()[1].exc_info[0]:
+                        tb += u''.join(traceback.format_exception(*sys.exc_info()[1].exc_info))
+                    tb += compat_str(traceback.format_exc())
+                else:
+                    tb_data = traceback.format_list(traceback.extract_stack())
+                    tb = u''.join(tb_data)
+            self.to_stderr(tb)
         if not self.params.get('ignoreerrors', False):
         if not self.params.get('ignoreerrors', False):
-            raise DownloadError(message)
+            if sys.exc_info()[0] and hasattr(sys.exc_info()[1], 'exc_info') and sys.exc_info()[1].exc_info[0]:
+                exc_info = sys.exc_info()[1].exc_info
+            else:
+                exc_info = sys.exc_info()
+            raise DownloadError(message, exc_info)
         self._download_retcode = 1
 
         self._download_retcode = 1
 
+    def report_warning(self, message):
+        '''
+        Print the message to stderr, it will be prefixed with 'WARNING:'
+        If stderr is a tty file the 'WARNING:' will be colored
+        '''
+        if sys.stderr.isatty() and os.name != 'nt':
+            _msg_header=u'\033[0;33mWARNING:\033[0m'
+        else:
+            _msg_header=u'WARNING:'
+        warning_message=u'%s %s' % (_msg_header,message)
+        self.to_stderr(warning_message)
+
+    def report_error(self, message, tb=None):
+        '''
+        Do the same as trouble, but prefixes the message with 'ERROR:', colored
+        in red if stderr is a tty file.
+        '''
+        if sys.stderr.isatty() and os.name != 'nt':
+            _msg_header = u'\033[0;31mERROR:\033[0m'
+        else:
+            _msg_header = u'ERROR:'
+        error_message = u'%s %s' % (_msg_header, message)
+        self.trouble(error_message, tb)
+
     def slow_down(self, start_time, byte_counter):
         """Sleep if the download speed is over the rate limit."""
         rate_limit = self.params.get('ratelimit', None)
     def slow_down(self, start_time, byte_counter):
         """Sleep if the download speed is over the rate limit."""
         rate_limit = self.params.get('ratelimit', None)
@@ -255,7 +308,7 @@ class FileDownloader(object):
                 return
             os.rename(encodeFilename(old_filename), encodeFilename(new_filename))
         except (IOError, OSError) as err:
                 return
             os.rename(encodeFilename(old_filename), encodeFilename(new_filename))
         except (IOError, OSError) as err:
-            self.trouble(u'ERROR: unable to rename file')
+            self.report_error(u'unable to rename file')
 
     def try_utime(self, filename, last_modified_hdr):
         """Try to set the last-modified time of the given file."""
 
     def try_utime(self, filename, last_modified_hdr):
         """Try to set the last-modified time of the given file."""
@@ -279,9 +332,9 @@ class FileDownloader(object):
         """ Report that the description file is being written """
         self.to_screen(u'[info] Writing video description to: ' + descfn)
 
         """ Report that the description file is being written """
         self.to_screen(u'[info] Writing video description to: ' + descfn)
 
-    def report_writesubtitles(self, srtfn):
+    def report_writesubtitles(self, sub_filename):
         """ Report that the subtitles file is being written """
         """ Report that the subtitles file is being written """
-        self.to_screen(u'[info] Writing video subtitles to: ' + srtfn)
+        self.to_screen(u'[info] Writing video subtitles to: ' + sub_filename)
 
     def report_writeinfojson(self, infofn):
         """ Report that the metadata file has been written """
 
     def report_writeinfojson(self, infofn):
         """ Report that the metadata file has been written """
@@ -295,8 +348,13 @@ class FileDownloader(object):
         """Report download progress."""
         if self.params.get('noprogress', False):
             return
         """Report download progress."""
         if self.params.get('noprogress', False):
             return
-        self.to_screen(u'\r[download] %s of %s at %s ETA %s' %
-                (percent_str, data_len_str, speed_str, eta_str), skip_eol=True)
+        clear_line = (u'\x1b[K' if sys.stderr.isatty() and os.name != 'nt' else u'')
+        if self.params.get('progress_with_newline', False):
+            self.to_screen(u'[download] %s of %s at %s ETA %s' %
+                (percent_str, data_len_str, speed_str, eta_str))
+        else:
+            self.to_screen(u'\r%s[download] %s of %s at %s ETA %s' %
+                (clear_line, percent_str, data_len_str, speed_str, eta_str), skip_eol=True)
         self.to_cons_title(u'youtube-dl - %s of %s at %s ETA %s' %
                 (percent_str.strip(), data_len_str.strip(), speed_str.strip(), eta_str.strip()))
 
         self.to_cons_title(u'youtube-dl - %s of %s at %s ETA %s' %
                 (percent_str.strip(), data_len_str.strip(), speed_str.strip(), eta_str.strip()))
 
@@ -336,7 +394,13 @@ class FileDownloader(object):
             template_dict = dict(info_dict)
 
             template_dict['epoch'] = int(time.time())
             template_dict = dict(info_dict)
 
             template_dict['epoch'] = int(time.time())
-            template_dict['autonumber'] = u'%05d' % self._num_downloads
+            autonumber_size = self.params.get('autonumber_size')
+            if autonumber_size is None:
+                autonumber_size = 5
+            autonumber_templ = u'%0' + str(autonumber_size) + u'd'
+            template_dict['autonumber'] = autonumber_templ % self._num_downloads
+            if template_dict['playlist_index'] is not None:
+                template_dict['playlist_index'] = u'%05d' % template_dict['playlist_index']
 
             sanitize = lambda k,v: sanitize_filename(
                 u'NA' if v is None else compat_str(v),
 
             sanitize = lambda k,v: sanitize_filename(
                 u'NA' if v is None else compat_str(v),
@@ -346,8 +410,11 @@ class FileDownloader(object):
 
             filename = self.params['outtmpl'] % template_dict
             return filename
 
             filename = self.params['outtmpl'] % template_dict
             return filename
-        except (ValueError, KeyError) as err:
-            self.trouble(u'ERROR: invalid system charset or erroneous output template')
+        except KeyError as err:
+            self.report_error(u'Erroneous output template')
+            return None
+        except ValueError as err:
+            self.report_error(u'Insufficient system charset ' + repr(preferredencoding()))
             return None
 
     def _match_entry(self, info_dict):
             return None
 
     def _match_entry(self, info_dict):
@@ -356,18 +423,150 @@ class FileDownloader(object):
         title = info_dict['title']
         matchtitle = self.params.get('matchtitle', False)
         if matchtitle:
         title = info_dict['title']
         matchtitle = self.params.get('matchtitle', False)
         if matchtitle:
-            matchtitle = matchtitle.decode('utf8')
             if not re.search(matchtitle, title, re.IGNORECASE):
                 return u'[download] "' + title + '" title did not match pattern "' + matchtitle + '"'
         rejecttitle = self.params.get('rejecttitle', False)
         if rejecttitle:
             if not re.search(matchtitle, title, re.IGNORECASE):
                 return u'[download] "' + title + '" title did not match pattern "' + matchtitle + '"'
         rejecttitle = self.params.get('rejecttitle', False)
         if rejecttitle:
-            rejecttitle = rejecttitle.decode('utf8')
             if re.search(rejecttitle, title, re.IGNORECASE):
                 return u'"' + title + '" title matched reject pattern "' + rejecttitle + '"'
             if re.search(rejecttitle, title, re.IGNORECASE):
                 return u'"' + title + '" title matched reject pattern "' + rejecttitle + '"'
+        date = info_dict.get('upload_date', None)
+        if date is not None:
+            dateRange = self.params.get('daterange', DateRange())
+            if date not in dateRange:
+                return u'[download] %s upload date is not in range %s' % (date_from_str(date).isoformat(), dateRange)
         return None
         return None
+        
+    def extract_info(self, url, download=True, ie_key=None, extra_info={}):
+        '''
+        Returns a list with a dictionary for each video we find.
+        If 'download', also downloads the videos.
+        extra_info is a dict containing the extra values to add to each result
+         '''
+        
+        if ie_key:
+            ie = get_info_extractor(ie_key)()
+            ie.set_downloader(self)
+            ies = [ie]
+        else:
+            ies = self._ies
+
+        for ie in ies:
+            if not ie.suitable(url):
+                continue
+
+            if not ie.working():
+                self.report_warning(u'The program functionality for this site has been marked as broken, '
+                                    u'and will probably not work.')
+
+            try:
+                ie_result = ie.extract(url)
+                if ie_result is None: # Finished already (backwards compatibility; listformats and friends should be moved here)
+                    break
+                if isinstance(ie_result, list):
+                    # Backwards compatibility: old IE result format
+                    for result in ie_result:
+                        result.update(extra_info)
+                    ie_result = {
+                        '_type': 'compat_list',
+                        'entries': ie_result,
+                    }
+                else:
+                    ie_result.update(extra_info)
+                if 'extractor' not in ie_result:
+                    ie_result['extractor'] = ie.IE_NAME
+                return self.process_ie_result(ie_result, download=download)
+            except ExtractorError as de: # An error we somewhat expected
+                self.report_error(compat_str(de), de.format_traceback())
+                break
+            except Exception as e:
+                if self.params.get('ignoreerrors', False):
+                    self.report_error(compat_str(e), tb=compat_str(traceback.format_exc()))
+                    break
+                else:
+                    raise
+        else:
+            self.report_error(u'no suitable InfoExtractor: %s' % url)
+        
+    def process_ie_result(self, ie_result, download=True, extra_info={}):
+        """
+        Take the result of the ie(may be modified) and resolve all unresolved
+        references (URLs, playlist items).
+
+        It will also download the videos if 'download'.
+        Returns the resolved ie_result.
+        """
+
+        result_type = ie_result.get('_type', 'video') # If not given we suppose it's a video, support the default old system
+        if result_type == 'video':
+            if 'playlist' not in ie_result:
+                # It isn't part of a playlist
+                ie_result['playlist'] = None
+                ie_result['playlist_index'] = None
+            if download:
+                self.process_info(ie_result)
+            return ie_result
+        elif result_type == 'url':
+            # We have to add extra_info to the results because it may be
+            # contained in a playlist
+            return self.extract_info(ie_result['url'],
+                                     download,
+                                     ie_key=ie_result.get('ie_key'),
+                                     extra_info=extra_info)
+        elif result_type == 'playlist':
+            # We process each entry in the playlist
+            playlist = ie_result.get('title', None) or ie_result.get('id', None)
+            self.to_screen(u'[download] Downloading playlist: %s'  % playlist)
+
+            playlist_results = []
+
+            n_all_entries = len(ie_result['entries'])
+            playliststart = self.params.get('playliststart', 1) - 1
+            playlistend = self.params.get('playlistend', -1)
+
+            if playlistend == -1:
+                entries = ie_result['entries'][playliststart:]
+            else:
+                entries = ie_result['entries'][playliststart:playlistend]
+
+            n_entries = len(entries)
+
+            self.to_screen(u"[%s] playlist '%s': Collected %d video ids (downloading %d of them)" %
+                (ie_result['extractor'], playlist, n_all_entries, n_entries))
+
+            for i,entry in enumerate(entries,1):
+                self.to_screen(u'[download] Downloading video #%s of %s' %(i, n_entries))
+                extra = {
+                         'playlist': playlist, 
+                         'playlist_index': i + playliststart,
+                         }
+                entry_result = self.process_ie_result(entry,
+                                                      download=download,
+                                                      extra_info=extra)
+                playlist_results.append(entry_result)
+            ie_result['entries'] = playlist_results
+            return ie_result
+        elif result_type == 'compat_list':
+            def _fixup(r):
+                r.setdefault('extractor', ie_result['extractor'])
+                return r
+            ie_result['entries'] = [
+                self.process_ie_result(_fixup(r), download=download)
+                for r in ie_result['entries']
+            ]
+            return ie_result
+        else:
+            raise Exception('Invalid result type: %s' % result_type)
 
     def process_info(self, info_dict):
 
     def process_info(self, info_dict):
-        """Process a single dictionary returned by an InfoExtractor."""
+        """Process a single resolved IE result."""
+
+        assert info_dict.get('_type', 'video') == 'video'
+        #We increment the download the download count here to match the previous behaviour.
+        self.increment_downloads()
+
+        info_dict['fulltitle'] = info_dict['title']
+        if len(info_dict['title']) > 200:
+            info_dict['title'] = info_dict['title'][:197] + u'...'
 
         # Keep for backwards compatibility
         info_dict['stitle'] = info_dict['title']
 
         # Keep for backwards compatibility
         info_dict['stitle'] = info_dict['title']
@@ -390,6 +589,8 @@ class FileDownloader(object):
         # Forced printings
         if self.params.get('forcetitle', False):
             compat_print(info_dict['title'])
         # Forced printings
         if self.params.get('forcetitle', False):
             compat_print(info_dict['title'])
+        if self.params.get('forceid', False):
+            compat_print(info_dict['id'])
         if self.params.get('forceurl', False):
             compat_print(info_dict['url'])
         if self.params.get('forcethumbnail', False) and 'thumbnail' in info_dict:
         if self.params.get('forceurl', False):
             compat_print(info_dict['url'])
         if self.params.get('forcethumbnail', False) and 'thumbnail' in info_dict:
@@ -410,59 +611,81 @@ class FileDownloader(object):
 
         try:
             dn = os.path.dirname(encodeFilename(filename))
 
         try:
             dn = os.path.dirname(encodeFilename(filename))
-            if dn != '' and not os.path.exists(dn): # dn is already encoded
+            if dn != '' and not os.path.exists(dn):
                 os.makedirs(dn)
         except (OSError, IOError) as err:
                 os.makedirs(dn)
         except (OSError, IOError) as err:
-            self.trouble(u'ERROR: unable to create directory ' + compat_str(err))
+            self.report_error(u'unable to create directory ' + compat_str(err))
             return
 
         if self.params.get('writedescription', False):
             try:
                 descfn = filename + u'.description'
                 self.report_writedescription(descfn)
             return
 
         if self.params.get('writedescription', False):
             try:
                 descfn = filename + u'.description'
                 self.report_writedescription(descfn)
-                descfile = open(encodeFilename(descfn), 'wb')
-                try:
-                    descfile.write(info_dict['description'].encode('utf-8'))
-                finally:
-                    descfile.close()
+                with io.open(encodeFilename(descfn), 'w', encoding='utf-8') as descfile:
+                    descfile.write(info_dict['description'])
             except (OSError, IOError):
             except (OSError, IOError):
-                self.trouble(u'ERROR: Cannot write description file ' + descfn)
+                self.report_error(u'Cannot write description file ' + descfn)
                 return
 
         if self.params.get('writesubtitles', False) and 'subtitles' in info_dict and info_dict['subtitles']:
             # subtitles download errors are already managed as troubles in relevant IE
             # that way it will silently go on when used with unsupporting IE
                 return
 
         if self.params.get('writesubtitles', False) and 'subtitles' in info_dict and info_dict['subtitles']:
             # subtitles download errors are already managed as troubles in relevant IE
             # that way it will silently go on when used with unsupporting IE
-            try:
-                srtfn = filename.rsplit('.', 1)[0] + u'.srt'
-                self.report_writesubtitles(srtfn)
-                srtfile = open(encodeFilename(srtfn), 'wb')
+            subtitle = info_dict['subtitles'][0]
+            (sub_error, sub_lang, sub) = subtitle
+            sub_format = self.params.get('subtitlesformat')
+            if sub_error:
+                self.report_warning("Some error while getting the subtitles")
+            else:
                 try:
                 try:
-                    srtfile.write(info_dict['subtitles'].encode('utf-8'))
-                finally:
-                    srtfile.close()
-            except (OSError, IOError):
-                self.trouble(u'ERROR: Cannot write subtitles file ' + descfn)
-                return
+                    sub_filename = filename.rsplit('.', 1)[0] + u'.' + sub_lang + u'.' + sub_format
+                    self.report_writesubtitles(sub_filename)
+                    with io.open(encodeFilename(sub_filename), 'w', encoding='utf-8') as subfile:
+                        subfile.write(sub)
+                except (OSError, IOError):
+                    self.report_error(u'Cannot write subtitles file ' + descfn)
+                    return
+
+        if self.params.get('allsubtitles', False) and 'subtitles' in info_dict and info_dict['subtitles']:
+            subtitles = info_dict['subtitles']
+            sub_format = self.params.get('subtitlesformat')
+            for subtitle in subtitles:
+                (sub_error, sub_lang, sub) = subtitle
+                if sub_error:
+                    self.report_warning("Some error while getting the subtitles")
+                else:
+                    try:
+                        sub_filename = filename.rsplit('.', 1)[0] + u'.' + sub_lang + u'.' + sub_format
+                        self.report_writesubtitles(sub_filename)
+                        with io.open(encodeFilename(sub_filename), 'w', encoding='utf-8') as subfile:
+                                subfile.write(sub)
+                    except (OSError, IOError):
+                        self.report_error(u'Cannot write subtitles file ' + descfn)
+                        return
 
         if self.params.get('writeinfojson', False):
             infofn = filename + u'.info.json'
             self.report_writeinfojson(infofn)
             try:
 
         if self.params.get('writeinfojson', False):
             infofn = filename + u'.info.json'
             self.report_writeinfojson(infofn)
             try:
-                json.dump
-            except (NameError,AttributeError):
-                self.trouble(u'ERROR: No JSON encoder found. Update to Python 2.6+, setup a json module, or leave out --write-info-json.')
-                return
-            try:
-                infof = open(encodeFilename(infofn), 'wb')
-                try:
-                    json_info_dict = dict((k,v) for k,v in info_dict.iteritems() if not k in ('urlhandle',))
-                    json.dump(json_info_dict, infof)
-                finally:
-                    infof.close()
+                json_info_dict = dict((k, v) for k,v in info_dict.items() if not k in ['urlhandle'])
+                write_json_file(json_info_dict, encodeFilename(infofn))
             except (OSError, IOError):
             except (OSError, IOError):
-                self.trouble(u'ERROR: Cannot write metadata to JSON file ' + infofn)
+                self.report_error(u'Cannot write metadata to JSON file ' + infofn)
                 return
 
                 return
 
+        if self.params.get('writethumbnail', False):
+            if 'thumbnail' in info_dict:
+                thumb_format = info_dict['thumbnail'].rpartition(u'/')[2].rpartition(u'.')[2]
+                if not thumb_format:
+                    thumb_format = 'jpg'
+                thumb_filename = filename.rpartition('.')[0] + u'.' + thumb_format
+                self.to_screen(u'[%s] %s: Downloading thumbnail ...' %
+                               (info_dict['extractor'], info_dict['id']))
+                uf = compat_urllib_request.urlopen(info_dict['thumbnail'])
+                with open(thumb_filename, 'wb') as thumbf:
+                    shutil.copyfileobj(uf, thumbf)
+                self.to_screen(u'[%s] %s: Writing thumbnail to: %s' %
+                               (info_dict['extractor'], info_dict['id'], thumb_filename))
+
         if not self.params.get('skip_download', False):
             if self.params.get('nooverwrites', False) and os.path.exists(encodeFilename(filename)):
                 success = True
         if not self.params.get('skip_download', False):
             if self.params.get('nooverwrites', False) and os.path.exists(encodeFilename(filename)):
                 success = True
@@ -472,17 +695,17 @@ class FileDownloader(object):
                 except (OSError, IOError) as err:
                     raise UnavailableVideoError()
                 except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
                 except (OSError, IOError) as err:
                     raise UnavailableVideoError()
                 except (compat_urllib_error.URLError, compat_http_client.HTTPException, socket.error) as err:
-                    self.trouble(u'ERROR: unable to download video data: %s' % str(err))
+                    self.report_error(u'unable to download video data: %s' % str(err))
                     return
                 except (ContentTooShortError, ) as err:
                     return
                 except (ContentTooShortError, ) as err:
-                    self.trouble(u'ERROR: content too short (expected %s bytes and served %s)' % (err.expected, err.downloaded))
+                    self.report_error(u'content too short (expected %s bytes and served %s)' % (err.expected, err.downloaded))
                     return
 
             if success:
                 try:
                     self.post_process(filename, info_dict)
                 except (PostProcessingError) as err:
                     return
 
             if success:
                 try:
                     self.post_process(filename, info_dict)
                 except (PostProcessingError) as err:
-                    self.trouble(u'ERROR: postprocessing: %s' % str(err))
+                    self.report_error(u'postprocessing: %s' % str(err))
                     return
 
     def download(self, url_list):
                     return
 
     def download(self, url_list):
@@ -491,62 +714,64 @@ class FileDownloader(object):
             raise SameFileError(self.params['outtmpl'])
 
         for url in url_list:
             raise SameFileError(self.params['outtmpl'])
 
         for url in url_list:
-            suitable_found = False
-            for ie in self._ies:
-                # Go to next InfoExtractor if not suitable
-                if not ie.suitable(url):
-                    continue
-
-                # Warn if the _WORKING attribute is False
-                if not ie.working():
-                    self.trouble(u'WARNING: the program functionality for this site has been marked as broken, '
-                                 u'and will probably not work. If you want to go on, use the -i option.')
-
-                # Suitable InfoExtractor found
-                suitable_found = True
-
-                # Extract information from URL and process it
-                videos = ie.extract(url)
-                for video in videos or []:
-                    video['extractor'] = ie.IE_NAME
-                    try:
-                        self.increment_downloads()
-                        self.process_info(video)
-                    except UnavailableVideoError:
-                        self.trouble(u'\nERROR: unable to download video')
-
-                # Suitable InfoExtractor had been found; go to next URL
-                break
-
-            if not suitable_found:
-                self.trouble(u'ERROR: no suitable InfoExtractor: %s' % url)
+            try:
+                #It also downloads the videos
+                videos = self.extract_info(url)
+            except UnavailableVideoError:
+                self.report_error(u'unable to download video')
+            except MaxDownloadsReached:
+                self.to_screen(u'[info] Maximum number of downloaded files reached.')
+                raise
 
         return self._download_retcode
 
     def post_process(self, filename, ie_info):
 
         return self._download_retcode
 
     def post_process(self, filename, ie_info):
-        """Run the postprocessing chain on the given file."""
+        """Run all the postprocessors on the given file."""
         info = dict(ie_info)
         info['filepath'] = filename
         info = dict(ie_info)
         info['filepath'] = filename
+        keep_video = None
         for pp in self._pps:
         for pp in self._pps:
-            info = pp.run(info)
-            if info is None:
-                break
+            try:
+                keep_video_wish,new_info = pp.run(info)
+                if keep_video_wish is not None:
+                    if keep_video_wish:
+                        keep_video = keep_video_wish
+                    elif keep_video is None:
+                        # No clear decision yet, let IE decide
+                        keep_video = keep_video_wish
+            except PostProcessingError as e:
+                self.to_stderr(u'ERROR: ' + e.msg)
+        if keep_video is False and not self.params.get('keepvideo', False):
+            try:
+                self.to_screen(u'Deleting original file %s (pass -k to keep)' % filename)
+                os.remove(encodeFilename(filename))
+            except (IOError, OSError):
+                self.report_warning(u'Unable to remove downloaded video file')
 
 
-    def _download_with_rtmpdump(self, filename, url, player_url):
+    def _download_with_rtmpdump(self, filename, url, player_url, page_url, play_path, tc_url):
         self.report_destination(filename)
         tmpfilename = self.temp_name(filename)
 
         # Check for rtmpdump first
         try:
         self.report_destination(filename)
         tmpfilename = self.temp_name(filename)
 
         # Check for rtmpdump first
         try:
-            subprocess.call(['rtmpdump', '-h'], stdout=(file(os.path.devnull, 'w')), stderr=subprocess.STDOUT)
+            subprocess.call(['rtmpdump', '-h'], stdout=(open(os.path.devnull, 'w')), stderr=subprocess.STDOUT)
         except (OSError, IOError):
         except (OSError, IOError):
-            self.trouble(u'ERROR: RTMP download detected but "rtmpdump" could not be run')
+            self.report_error(u'RTMP download detected but "rtmpdump" could not be run')
             return False
 
         # Download using rtmpdump. rtmpdump returns exit code 2 when
         # the connection was interrumpted and resuming appears to be
         # possible. This is part of rtmpdump's normal usage, AFAIK.
             return False
 
         # Download using rtmpdump. rtmpdump returns exit code 2 when
         # the connection was interrumpted and resuming appears to be
         # possible. This is part of rtmpdump's normal usage, AFAIK.
-        basic_args = ['rtmpdump', '-q'] + [[], ['-W', player_url]][player_url is not None] + ['-r', url, '-o', tmpfilename]
+        basic_args = ['rtmpdump', '-q', '-r', url, '-o', tmpfilename]
+        if self.params.get('verbose', False): basic_args[1] = '-v'
+        if player_url is not None:
+            basic_args += ['-W', player_url]
+        if page_url is not None:
+            basic_args += ['--pageUrl', page_url]
+        if play_path is not None:
+            basic_args += ['-y', play_path]
+        if tc_url is not None:
+            basic_args += ['--tcUrl', url]
         args = basic_args + [[], ['-e', '-k', '1']][self.params.get('continuedl', False)]
         if self.params.get('verbose', False):
             try:
         args = basic_args + [[], ['-e', '-k', '1']][self.params.get('continuedl', False)]
         if self.params.get('verbose', False):
             try:
@@ -570,31 +795,48 @@ class FileDownloader(object):
                 retval = 0
                 break
         if retval == 0:
                 retval = 0
                 break
         if retval == 0:
-            self.to_screen(u'\r[rtmpdump] %s bytes' % os.path.getsize(encodeFilename(tmpfilename)))
+            fsize = os.path.getsize(encodeFilename(tmpfilename))
+            self.to_screen(u'\r[rtmpdump] %s bytes' % fsize)
             self.try_rename(tmpfilename, filename)
             self.try_rename(tmpfilename, filename)
+            self._hook_progress({
+                'downloaded_bytes': fsize,
+                'total_bytes': fsize,
+                'filename': filename,
+                'status': 'finished',
+            })
             return True
         else:
             return True
         else:
-            self.trouble(u'\nERROR: rtmpdump exited with code %d' % retval)
+            self.to_stderr(u"\n")
+            self.report_error(u'rtmpdump exited with code %d' % retval)
             return False
 
     def _do_download(self, filename, info_dict):
         url = info_dict['url']
             return False
 
     def _do_download(self, filename, info_dict):
         url = info_dict['url']
-        player_url = info_dict.get('player_url', None)
 
         # Check file already present
         if self.params.get('continuedl', False) and os.path.isfile(encodeFilename(filename)) and not self.params.get('nopart', False):
             self.report_file_already_downloaded(filename)
 
         # Check file already present
         if self.params.get('continuedl', False) and os.path.isfile(encodeFilename(filename)) and not self.params.get('nopart', False):
             self.report_file_already_downloaded(filename)
+            self._hook_progress({
+                'filename': filename,
+                'status': 'finished',
+            })
             return True
 
         # Attempt to download using rtmpdump
         if url.startswith('rtmp'):
             return True
 
         # Attempt to download using rtmpdump
         if url.startswith('rtmp'):
-            return self._download_with_rtmpdump(filename, url, player_url)
+            return self._download_with_rtmpdump(filename, url,
+                                                info_dict.get('player_url', None),
+                                                info_dict.get('page_url', None),
+                                                info_dict.get('play_path', None),
+                                                info_dict.get('tc_url', None))
 
         tmpfilename = self.temp_name(filename)
         stream = None
 
         # Do not include the Accept-Encoding header
         headers = {'Youtubedl-no-compression': 'True'}
 
         tmpfilename = self.temp_name(filename)
         stream = None
 
         # Do not include the Accept-Encoding header
         headers = {'Youtubedl-no-compression': 'True'}
+        if 'user_agent' in info_dict:
+            headers['Youtubedl-user-agent'] = info_dict['user_agent']
         basic_request = compat_urllib_request.Request(url, None, headers)
         request = compat_urllib_request.Request(url, None, headers)
 
         basic_request = compat_urllib_request.Request(url, None, headers)
         request = compat_urllib_request.Request(url, None, headers)
 
@@ -651,6 +893,10 @@ class FileDownloader(object):
                             # the one in the hard drive.
                             self.report_file_already_downloaded(filename)
                             self.try_rename(tmpfilename, filename)
                             # the one in the hard drive.
                             self.report_file_already_downloaded(filename)
                             self.try_rename(tmpfilename, filename)
+                            self._hook_progress({
+                                'filename': filename,
+                                'status': 'finished',
+                            })
                             return True
                         else:
                             # The length does not match, we start the download over
                             return True
                         else:
                             # The length does not match, we start the download over
@@ -663,12 +909,21 @@ class FileDownloader(object):
                 self.report_retry(count, retries)
 
         if count > retries:
                 self.report_retry(count, retries)
 
         if count > retries:
-            self.trouble(u'ERROR: giving up after %s retries' % retries)
+            self.report_error(u'giving up after %s retries' % retries)
             return False
 
         data_len = data.info().get('Content-length', None)
         if data_len is not None:
             data_len = int(data_len) + resume_len
             return False
 
         data_len = data.info().get('Content-length', None)
         if data_len is not None:
             data_len = int(data_len) + resume_len
+            min_data_len = self.params.get("min_filesize", None)
+            max_data_len =  self.params.get("max_filesize", None)
+            if min_data_len is not None and data_len < min_data_len:
+                self.to_screen(u'\r[download] File is smaller than min-filesize (%s bytes < %s bytes). Aborting.' % (data_len, min_data_len))
+                return False
+            if max_data_len is not None and data_len > max_data_len:
+                self.to_screen(u'\r[download] File is larger than max-filesize (%s bytes > %s bytes). Aborting.' % (data_len, max_data_len))
+                return False
+
         data_len_str = self.format_bytes(data_len)
         byte_counter = 0 + resume_len
         block_size = self.params.get('buffersize', 1024)
         data_len_str = self.format_bytes(data_len)
         byte_counter = 0 + resume_len
         block_size = self.params.get('buffersize', 1024)
@@ -690,12 +945,13 @@ class FileDownloader(object):
                     filename = self.undo_temp_name(tmpfilename)
                     self.report_destination(filename)
                 except (OSError, IOError) as err:
                     filename = self.undo_temp_name(tmpfilename)
                     self.report_destination(filename)
                 except (OSError, IOError) as err:
-                    self.trouble(u'ERROR: unable to open for writing: %s' % str(err))
+                    self.report_error(u'unable to open for writing: %s' % str(err))
                     return False
             try:
                 stream.write(data_block)
             except (IOError, OSError) as err:
                     return False
             try:
                 stream.write(data_block)
             except (IOError, OSError) as err:
-                self.trouble(u'\nERROR: unable to write data: %s' % str(err))
+                self.to_stderr(u"\n")
+                self.report_error(u'unable to write data: %s' % str(err))
                 return False
             if not self.params.get('noresizebuffer', False):
                 block_size = self.best_block_size(after - before, len(data_block))
                 return False
             if not self.params.get('noresizebuffer', False):
                 block_size = self.best_block_size(after - before, len(data_block))
@@ -709,11 +965,20 @@ class FileDownloader(object):
                 eta_str = self.calc_eta(start, time.time(), data_len - resume_len, byte_counter - resume_len)
                 self.report_progress(percent_str, data_len_str, speed_str, eta_str)
 
                 eta_str = self.calc_eta(start, time.time(), data_len - resume_len, byte_counter - resume_len)
                 self.report_progress(percent_str, data_len_str, speed_str, eta_str)
 
+            self._hook_progress({
+                'downloaded_bytes': byte_counter,
+                'total_bytes': data_len,
+                'tmpfilename': tmpfilename,
+                'filename': filename,
+                'status': 'downloading',
+            })
+
             # Apply rate limit
             self.slow_down(start, byte_counter - resume_len)
 
         if stream is None:
             # Apply rate limit
             self.slow_down(start, byte_counter - resume_len)
 
         if stream is None:
-            self.trouble(u'\nERROR: Did not get any data blocks')
+            self.to_stderr(u"\n")
+            self.report_error(u'Did not get any data blocks')
             return False
         stream.close()
         self.report_finish()
             return False
         stream.close()
         self.report_finish()
@@ -725,4 +990,31 @@ class FileDownloader(object):
         if self.params.get('updatetime', True):
             info_dict['filetime'] = self.try_utime(filename, data.info().get('last-modified', None))
 
         if self.params.get('updatetime', True):
             info_dict['filetime'] = self.try_utime(filename, data.info().get('last-modified', None))
 
+        self._hook_progress({
+            'downloaded_bytes': byte_counter,
+            'total_bytes': byte_counter,
+            'filename': filename,
+            'status': 'finished',
+        })
+
         return True
         return True
+
+    def _hook_progress(self, status):
+        for ph in self._progress_hooks:
+            ph(status)
+
+    def add_progress_hook(self, ph):
+        """ ph gets called on download progress, with a dictionary with the entries
+        * filename: The final filename
+        * status: One of "downloading" and "finished"
+
+        It can also have some of the following entries:
+
+        * downloaded_bytes: Bytes on disks
+        * total_bytes: Total bytes, None if unknown
+        * tmpfilename: The filename we're currently writing to
+
+        Hooks are guaranteed to be called at least once (with status "finished")
+        if the download is successful.
+        """
+        self._progress_hooks.append(ph)