From 33150f5de388e34cc776a517db099dea39aab392 Mon Sep 17 00:00:00 2001 From: Samantaz Fox Date: Sun, 21 Aug 2022 17:26:30 +0200 Subject: [PATCH] spec: Add test cases for regular videos extraction --- mocks | 2 +- .../videos/regular_videos_extract_spec.cr | 168 ++++++++++++++++++ 2 files changed, 169 insertions(+), 1 deletion(-) create mode 100644 spec/invidious/videos/regular_videos_extract_spec.cr diff --git a/mocks b/mocks index c401dd92..dfd53ea6 160000 --- a/mocks +++ b/mocks @@ -1 +1 @@ -Subproject commit c401dd9203434b561022242c24b0c200d72284c0 +Subproject commit dfd53ea6ceb3cbcbbce6004f6ce60b330ad0f9b1 diff --git a/spec/invidious/videos/regular_videos_extract_spec.cr b/spec/invidious/videos/regular_videos_extract_spec.cr new file mode 100644 index 00000000..132b37a3 --- /dev/null +++ b/spec/invidious/videos/regular_videos_extract_spec.cr @@ -0,0 +1,168 @@ +require "../../parsers_helper.cr" + +Spectator.describe "parse_video_info" do + it "parses a regular video" do + # Enable mock + _player = load_mock("video/regular_mrbeast.player") + _next = load_mock("video/regular_mrbeast.next") + + raw_data = _player.merge!(_next) + info = parse_video_info("2isYuQZMbdU", raw_data) + + # Some basic verifications + expect(typeof(info)).to eq(Hash(String, JSON::Any)) + + expect(info["videoType"].as_s).to eq("Video") + + # Basic video infos + + expect(info["title"].as_s).to eq("I Gave My 100,000,000th Subscriber An Island") + expect(info["views"].as_i).to eq(32_846_329) + expect(info["likes"].as_i).to eq(2_611_650) + + # For some reason the video length from VideoDetails and the + # one from microformat differs by 1s... + expect(info["lengthSeconds"].as_i).to be_between(930_i64, 931_i64) + + expect(info["published"].as_s).to eq("2022-08-04T00:00:00Z") + + # Extra video infos + + expect(info["allowedRegions"].as_a).to_not be_empty + expect(info["allowedRegions"].as_a.size).to eq(249) + + expect(info["allowedRegions"].as_a).to contain( + "AD", "BA", "BB", "BW", "BY", "EG", "GG", "HN", "NP", "NR", "TR", + "TT", "TV", "TW", "TZ", "VA", "VC", "VE", "VG", "VI", "VN", "VU", + "WF", "WS", "YE", "YT", "ZA", "ZM", "ZW" + ) + + expect(info["keywords"].as_a).to be_empty + + expect(info["allowRatings"].as_bool).to be_true + expect(info["isFamilyFriendly"].as_bool).to be_true + expect(info["isListed"].as_bool).to be_true + expect(info["isUpcoming"].as_bool).to be_false + + # Related videos + + expect(info["relatedVideos"].as_a.size).to eq(19) + + expect(info["relatedVideos"][0]["id"]).to eq("tVWWp1PqDus") + expect(info["relatedVideos"][0]["title"]).to eq("100 Girls Vs 100 Boys For $500,000") + expect(info["relatedVideos"][0]["author"]).to eq("MrBeast") + expect(info["relatedVideos"][0]["ucid"]).to eq("UCX6OQ3DkcsbYNE6H8uQQuVA") + expect(info["relatedVideos"][0]["view_count"]).to eq("49702799") + expect(info["relatedVideos"][0]["short_view_count"]).to eq("49M") + expect(info["relatedVideos"][0]["author_verified"]).to eq("true") + + # Description + + description = "🚀Launch a store on Shopify, I’ll buy from 100 random stores that do ▸ " + + expect(info["description"].as_s).to start_with(description) + expect(info["shortDescription"].as_s).to start_with(description) + expect(info["descriptionHtml"].as_s).to start_with(description) + + # Video metadata + + expect(info["genre"].as_s).to eq("Entertainment") + expect(info["genreUcid"].as_s).to be_empty + expect(info["license"].as_s).to be_empty + + # Author infos + + expect(info["author"].as_s).to eq("MrBeast") + expect(info["ucid"].as_s).to eq("UCX6OQ3DkcsbYNE6H8uQQuVA") + + expect(info["authorThumbnail"].as_s).to eq( + "https://yt3.ggpht.com/ytc/AMLnZu84dsnlYtuUFBMC8imQs0IUcTKA9khWAmUOgQZltw=s48-c-k-c0x00ffffff-no-rj" + ) + + expect(info["authorVerified"].as_bool).to be_true + expect(info["subCountText"].as_s).to eq("101M") + end + + it "parses a regular video with no descrition/comments" do + # Enable mock + _player = load_mock("video/regular_no-description.player") + _next = load_mock("video/regular_no-description.next") + + raw_data = _player.merge!(_next) + info = parse_video_info("iuevw6218F0", raw_data) + + # Some basic verifications + expect(typeof(info)).to eq(Hash(String, JSON::Any)) + + expect(info["videoType"].as_s).to eq("Video") + + # Basic video infos + + expect(info["title"].as_s).to eq("Chris Rea - Auberge") + expect(info["views"].as_i).to eq(10_356_197) + expect(info["likes"].as_i).to eq(0) + expect(info["lengthSeconds"].as_i).to eq(283_i64) + expect(info["published"].as_s).to eq("2012-05-21T00:00:00Z") + + # Extra video infos + + expect(info["allowedRegions"].as_a).to_not be_empty + expect(info["allowedRegions"].as_a.size).to eq(249) + + expect(info["allowedRegions"].as_a).to contain( + "AD", "BA", "BB", "BW", "BY", "EG", "GG", "HN", "NP", "NR", "TR", + "TT", "TV", "TW", "TZ", "VA", "VC", "VE", "VG", "VI", "VN", "VU", + "WF", "WS", "YE", "YT", "ZA", "ZM", "ZW" + ) + + expect(info["keywords"].as_a).to_not be_empty + expect(info["keywords"].as_a.size).to eq(4) + + expect(info["keywords"].as_a).to contain_exactly( + "Chris", + "Rea", + "Auberge", + "1991" + ).in_any_order + + expect(info["allowRatings"].as_bool).to be_true + expect(info["isFamilyFriendly"].as_bool).to be_true + expect(info["isListed"].as_bool).to be_true + expect(info["isUpcoming"].as_bool).to be_false + + # Related videos + + expect(info["relatedVideos"].as_a.size).to eq(19) + + expect(info["relatedVideos"][0]["id"]).to eq("0bkrY_V0yZg") + expect(info["relatedVideos"][0]["title"]).to eq( + "Chris Rea Best Songs Collection - Chris Rea Greatest Hits Full Album 2022" + ) + expect(info["relatedVideos"][0]["author"]).to eq("Rock Ultimate") + expect(info["relatedVideos"][0]["ucid"]).to eq("UCekSc2A19di9koUIpj8gxlQ") + expect(info["relatedVideos"][0]["view_count"]).to eq("1992412") + expect(info["relatedVideos"][0]["short_view_count"]).to eq("1.9M") + expect(info["relatedVideos"][0]["author_verified"]).to eq("false") + + # Description + + expect(info["description"].as_s).to eq(" ") + expect(info["shortDescription"].as_s).to be_empty + expect(info["descriptionHtml"].as_s).to eq("

") + + # Video metadata + + expect(info["genre"].as_s).to eq("Music") + expect(info["genreUcid"].as_s).to be_empty + expect(info["license"].as_s).to be_empty + + # Author infos + + expect(info["author"].as_s).to eq("ChrisReaOfficial") + expect(info["ucid"].as_s).to eq("UC_5q6nWPbD30-y6oiWF_oNA") + + expect(info["authorThumbnail"].as_s).to be_empty + expect(info["authorVerified"].as_bool).to be_false + expect(info["subCountText"].as_s).to eq("-") + end +end