add more details
parent
641ec13ef8
commit
075f6681c8
89
analysis.py
89
analysis.py
|
@ -1,9 +1,47 @@
|
|||
import os
|
||||
import json
|
||||
|
||||
from numpy import array
|
||||
|
||||
def get_min_max_quality_label(formats):
|
||||
formats = [f for f in formats if f.get("qualityLabel") is not None]
|
||||
formats.sort(key=lambda x: x.get("width"))
|
||||
min_format = formats[0]
|
||||
max_format = formats[-1]
|
||||
|
||||
return min_format.get("qualityLabel") + " - " + max_format.get("qualityLabel")
|
||||
|
||||
def get_unique_mime_str(formats):
|
||||
mimes = []
|
||||
for f in formats:
|
||||
if f.get("mimeType") is not None:
|
||||
mime_short = f.get("mimeType").split(";")[0]
|
||||
|
||||
if mime_short not in mimes:
|
||||
mimes.append(mime_short)
|
||||
|
||||
mimes_str = ""
|
||||
for m in mimes:
|
||||
mimes_str += "`" + m + "` "
|
||||
|
||||
return mimes_str
|
||||
|
||||
def get_structure_tree(data, depth = 0):
|
||||
result = ""
|
||||
|
||||
for attribute in data:
|
||||
if isinstance(data[attribute], dict):
|
||||
result += ("│ " * depth) + "├─`" + attribute + "`<br>"
|
||||
result += get_structure_tree(data[attribute], depth + 1)
|
||||
if isinstance(data[attribute], list):
|
||||
result += ("│ " * depth) + "├─`[" + attribute + "]` (" + str(len(data[attribute])) + ")<br>"
|
||||
|
||||
return result
|
||||
|
||||
|
||||
markdown = ""
|
||||
markdown += "| Client Name | Client Version | Quality Formats | Features |\n"
|
||||
markdown += "|-------------|----------------|-----------------|----------|\n"
|
||||
markdown += "|ID| Client Name | Client Version | Quality Formats | Features/Attributes |\n"
|
||||
markdown += "|--|-------------|----------------|-----------------|----------|\n"
|
||||
|
||||
txt_output = ""
|
||||
|
||||
|
@ -41,25 +79,36 @@ for client_id in client_ids:
|
|||
|
||||
txt_output += client_name + ";" + client_version + "\n"
|
||||
|
||||
formatsStr = "<details><summary>Formats</summary>"
|
||||
formats_combined = []
|
||||
|
||||
formatsStr = ""
|
||||
|
||||
if response_data.get('streamingData'):
|
||||
formats = response_data.get('streamingData').get('formats')
|
||||
if formats is not None and len(formats) > 0:
|
||||
if formats is not None:
|
||||
formatsStr = "<details><summary>Formats (" + str(len(formats)) + ")</summary>"
|
||||
formats_combined += formats
|
||||
|
||||
for format in formats:
|
||||
formatsStr += str(format.get("itag")) + " - " + str(format.get("qualityLabel")) + " - " + str(format.get("fps")) + " FPS - " + str(format.get("mimeType")).split(';')[0] + "<br>"
|
||||
formatsStr += str(format.get("itag")) + " - " + str(format.get("qualityLabel")) + " - " + str(format.get("mimeType")).split(';')[0] + "<br>"
|
||||
|
||||
formatsStr += "</details>"
|
||||
formatsStr += "</details>"
|
||||
|
||||
adaptiveFormatsStr= "<details><summary>Adaptive Formats</summary>"
|
||||
adaptiveFormatsStr = ""
|
||||
|
||||
if response_data.get('streamingData'):
|
||||
formats = response_data.get('streamingData').get('adaptiveFormats')
|
||||
if formats is not None and len(formats) > 0:
|
||||
for format in formats:
|
||||
adaptiveFormatsStr += str(format.get("itag")) + " - " + str(format.get("qualityLabel")) + " - " + str(format.get("fps")) + " FPS - " + str(format.get("mimeType")).split(';')[0] + "<br>"
|
||||
if formats is not None:
|
||||
adaptiveFormatsStr = "<details><summary>Adaptive Formats (" + str(len(formats)) + ")</summary>"
|
||||
formats_combined += formats
|
||||
|
||||
adaptiveFormatsStr += "</details>"
|
||||
for format in formats:
|
||||
adaptiveFormatsStr += str(format.get("itag")) + " - " + str(format.get("qualityLabel")) + " - " + str(format.get("mimeType")).split(';')[0] + "<br>"
|
||||
|
||||
adaptiveFormatsStr += "</details>"
|
||||
|
||||
formats_summary = "<b>" + get_min_max_quality_label(formats_combined) + "</b><br>"
|
||||
formats_summary += get_unique_mime_str(formats_combined) + "<br><br>"
|
||||
|
||||
extraInfo = ""
|
||||
if has_hls_format:
|
||||
|
@ -68,12 +117,26 @@ for client_id in client_ids:
|
|||
if has_mpeg_dash:
|
||||
extraInfo += "• MPEG-DASH Support"
|
||||
|
||||
ignore_attributes = ["videoDetails", "playerConfig", "responseContext", "playabilityStatus", "streamingData", "playbackTracking", "trackingParams", "adPlacements", "playerAds", "adParams", "adBreakParams"]
|
||||
|
||||
markdown += client_name + "|" + client_version + "|" + formatsStr + adaptiveFormatsStr + "|" + extraInfo + "|\n"
|
||||
if extraInfo != "":
|
||||
extraInfo += "<br><br>"
|
||||
|
||||
for attribute in response_data:
|
||||
if attribute not in ignore_attributes:
|
||||
extraInfo += "• `" + attribute + "`<br>"
|
||||
|
||||
if extraInfo != "":
|
||||
extraInfo += "<br>"
|
||||
|
||||
extraInfo += "<details><summary>Response Structure</summary>" + get_structure_tree(response_data) +"</details>"
|
||||
|
||||
|
||||
markdown += str(client_id) + "|" + client_name + "|" + client_version + "|" + formats_summary + formatsStr + adaptiveFormatsStr + "|" + extraInfo + "|\n"
|
||||
|
||||
break
|
||||
|
||||
f = open("results/working_clients.md", "w")
|
||||
f = open("results/working_clients.md", "w", encoding="utf-8")
|
||||
f.write(markdown)
|
||||
f.close()
|
||||
|
||||
|
|
|
@ -25,7 +25,7 @@
|
|||
"consistencyTokenJars": []
|
||||
}
|
||||
},
|
||||
"videoId": "xcJtL7QggTI",
|
||||
"videoId": "Zv11L-ZfrSg",
|
||||
"playbackContext": {
|
||||
"contentPlaybackContext": {
|
||||
"currentUrl": "/watch?v=7t0SqerlBA0",
|
||||
|
|
File diff suppressed because one or more lines are too long
Loading…
Reference in New Issue