pleroma/test/pleroma/web/rich_media/parser_test.exs
Mark Felder 6b7b443ff9 Pleroma.Web.RichMedia.Parser: Remove test-specific codepaths
Also consolidate Tesla mocks into the HttpRequestMock module.

Tests were not exercising the real codepaths. The Rich Media Preview only works with https, but most of these tests were only mocking http.
2024-02-06 18:33:54 -05:00

108 lines
4.1 KiB
Elixir

# Pleroma: A lightweight social networking server
# Copyright © 2017-2022 Pleroma Authors <https://pleroma.social/>
# SPDX-License-Identifier: AGPL-3.0-only
defmodule Pleroma.Web.RichMedia.ParserTest do
use Pleroma.DataCase, async: false
alias Pleroma.Web.RichMedia.Parser
import Tesla.Mock
setup do
mock_global(fn env -> apply(HttpRequestMock, :request, [env]) end)
end
test "returns error when no metadata present" do
assert {:error, _} = Parser.parse("https://example.com/empty")
end
test "doesn't just add a title" do
assert {:error, {:invalid_metadata, _}} = Parser.parse("https://example.com/non-ogp")
end
test "parses ogp" do
assert Parser.parse("https://example.com/ogp") ==
{:ok,
%{
"image" => "http://ia.media-imdb.com/images/rock.jpg",
"title" => "The Rock",
"description" =>
"Directed by Michael Bay. With Sean Connery, Nicolas Cage, Ed Harris, John Spencer.",
"type" => "video.movie",
"url" => "https://example.com/ogp"
}}
end
test "falls back to <title> when ogp:title is missing" do
assert Parser.parse("https://example.com/ogp-missing-title") ==
{:ok,
%{
"image" => "http://ia.media-imdb.com/images/rock.jpg",
"title" => "The Rock (1996)",
"description" =>
"Directed by Michael Bay. With Sean Connery, Nicolas Cage, Ed Harris, John Spencer.",
"type" => "video.movie",
"url" => "https://example.com/ogp-missing-title"
}}
end
test "parses twitter card" do
assert Parser.parse("https://example.com/twitter-card") ==
{:ok,
%{
"card" => "summary",
"site" => "@flickr",
"image" => "https://farm6.staticflickr.com/5510/14338202952_93595258ff_z.jpg",
"title" => "Small Island Developing States Photo Submission",
"description" => "View the album on Flickr.",
"url" => "https://example.com/twitter-card"
}}
end
test "parses OEmbed and filters HTML tags" do
assert Parser.parse("https://example.com/oembed") ==
{:ok,
%{
"author_name" => "\u202E\u202D\u202Cbees\u202C",
"author_url" => "https://www.flickr.com/photos/bees/",
"cache_age" => 3600,
"flickr_type" => "photo",
"height" => "768",
"html" =>
"<a href=\"https://www.flickr.com/photos/bees/2362225867/\" title=\"Bacon Lollys by \u202E\u202D\u202Cbees\u202C, on Flickr\"><img src=\"https://farm4.staticflickr.com/3040/2362225867_4a87ab8baf_b.jpg\" width=\"1024\" height=\"768\" alt=\"Bacon Lollys\"/></a>",
"license" => "All Rights Reserved",
"license_id" => 0,
"provider_name" => "Flickr",
"provider_url" => "https://www.flickr.com/",
"thumbnail_height" => 150,
"thumbnail_url" =>
"https://farm4.staticflickr.com/3040/2362225867_4a87ab8baf_q.jpg",
"thumbnail_width" => 150,
"title" => "Bacon Lollys",
"type" => "photo",
"url" => "https://example.com/oembed",
"version" => "1.0",
"web_page" => "https://www.flickr.com/photos/bees/2362225867/",
"web_page_short_url" => "https://flic.kr/p/4AK2sc",
"width" => "1024"
}}
end
test "rejects invalid OGP data" do
assert {:error, _} = Parser.parse("https://example.com/malformed")
end
test "returns error if getting page was not successful" do
assert {:error, :overload} = Parser.parse("https://example.com/error")
end
test "does a HEAD request to check if the body is too large" do
assert {:error, :body_too_large} = Parser.parse("https://example.com/huge-page")
end
test "does a HEAD request to check if the body is html" do
assert {:error, {:content_type, _}} = Parser.parse("https://example.com/pdf-file")
end
end