mirror of
https://github.com/discourse/discourse.git
synced 2024-11-30 18:35:17 +08:00
200 lines
6.5 KiB
Ruby
200 lines
6.5 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
RSpec.describe RobotsTxtController do
|
|
describe "#builder" do
|
|
it "returns json information for building a robots.txt" do
|
|
get "/robots-builder.json"
|
|
json = response.parsed_body
|
|
expect(json).to be_present
|
|
expect(json["header"]).to be_present
|
|
expect(json["agents"]).to be_present
|
|
end
|
|
|
|
it "includes overridden content if robots.txt is is overridden" do
|
|
SiteSetting.overridden_robots_txt = "something"
|
|
|
|
get "/robots-builder.json"
|
|
expect(response.status).to eq(200)
|
|
json = response.parsed_body
|
|
expect(json["header"]).to be_present
|
|
expect(json["agents"]).to be_present
|
|
expect(json["overridden"]).to eq("something")
|
|
end
|
|
end
|
|
|
|
describe "#index" do
|
|
context "when the content is overridden" do
|
|
it "is not prepended if there are no overrides" do
|
|
sign_in(Fabricate(:admin))
|
|
get "/robots.txt"
|
|
expect(response.body).not_to start_with(RobotsTxtController::OVERRIDDEN_HEADER)
|
|
end
|
|
|
|
it "is prepended if there are overrides and the user is admin" do
|
|
SiteSetting.overridden_robots_txt = "overridden_content"
|
|
sign_in(Fabricate(:admin))
|
|
get "/robots.txt"
|
|
expect(response.body).to start_with(RobotsTxtController::OVERRIDDEN_HEADER)
|
|
end
|
|
|
|
it "is not prepended if the user is not admin" do
|
|
SiteSetting.overridden_robots_txt = "overridden_content"
|
|
get "/robots.txt"
|
|
expect(response.body).not_to start_with(RobotsTxtController::OVERRIDDEN_HEADER)
|
|
end
|
|
end
|
|
|
|
context "with subfolder" do
|
|
it "prefixes the rules with the directory" do
|
|
set_subfolder "/forum"
|
|
|
|
get "/robots.txt"
|
|
expect(response.body).to include("\nDisallow: /forum/email/")
|
|
end
|
|
end
|
|
|
|
context "when allow_index_in_robots_txt is true" do
|
|
def expect_allowed_and_disallowed_sections(allow_index, disallow_index)
|
|
expect(allow_index).to be_present
|
|
expect(disallow_index).to be_present
|
|
|
|
allow_section =
|
|
(
|
|
if allow_index < disallow_index
|
|
response.body[allow_index...disallow_index]
|
|
else
|
|
response.body[allow_index..-1]
|
|
end
|
|
)
|
|
|
|
expect(allow_section).to include("Disallow: /auth/")
|
|
expect(allow_section).to_not include("Disallow: /\n")
|
|
|
|
disallowed_section =
|
|
(
|
|
if allow_index < disallow_index
|
|
response.body[disallow_index..-1]
|
|
else
|
|
response.body[disallow_index...allow_index]
|
|
end
|
|
)
|
|
expect(disallowed_section).to include("Disallow: /\n")
|
|
end
|
|
|
|
it "returns index when indexing is allowed" do
|
|
SiteSetting.allow_index_in_robots_txt = true
|
|
get "/robots.txt"
|
|
|
|
i = response.body.index("User-agent: *")
|
|
expect(i).to be_present
|
|
expect(response.body[i..-1]).to include("Disallow: /auth/")
|
|
# we have to insert Googlebot for special handling
|
|
expect(response.body[i..-1]).to include("User-agent: Googlebot")
|
|
end
|
|
|
|
it "can allowlist user agents" do
|
|
SiteSetting.allowed_crawler_user_agents = "Googlebot|Twitterbot"
|
|
get "/robots.txt"
|
|
expect(response.body).to include("User-agent: Googlebot")
|
|
expect(response.body).to include("User-agent: Twitterbot")
|
|
|
|
allowed_index = [
|
|
response.body.index("User-agent: Googlebot"),
|
|
response.body.index("User-agent: Twitterbot"),
|
|
].min
|
|
disallow_all_index = response.body.index("User-agent: *")
|
|
|
|
expect_allowed_and_disallowed_sections(allowed_index, disallow_all_index)
|
|
end
|
|
|
|
it "can blocklist user agents" do
|
|
SiteSetting.blocked_crawler_user_agents = "Googlebot|Twitterbot"
|
|
get "/robots.txt"
|
|
expect(response.body).to include("User-agent: Googlebot")
|
|
expect(response.body).to include("User-agent: Twitterbot")
|
|
|
|
disallow_index = [
|
|
response.body.index("User-agent: Googlebot"),
|
|
response.body.index("User-agent: Twitterbot"),
|
|
].min
|
|
allow_index = response.body.index("User-agent: *")
|
|
|
|
expect_allowed_and_disallowed_sections(allow_index, disallow_index)
|
|
end
|
|
|
|
it "ignores blocklist if allowlist is set" do
|
|
SiteSetting.allowed_crawler_user_agents = "Googlebot|Twitterbot"
|
|
SiteSetting.blocked_crawler_user_agents = "Bananabot"
|
|
get "/robots.txt"
|
|
expect(response.body).to_not include("Bananabot")
|
|
expect(response.body).to include("User-agent: Googlebot")
|
|
expect(response.body).to include("User-agent: Twitterbot")
|
|
end
|
|
end
|
|
|
|
it "returns noindex when indexing is disallowed" do
|
|
SiteSetting.allow_index_in_robots_txt = false
|
|
get "/robots.txt"
|
|
|
|
expect(response.body).to_not include("Disallow: /auth/")
|
|
expect(response.body).to include("User-agent: googlebot\nAllow")
|
|
end
|
|
|
|
it "returns overridden robots.txt if the file is overridden" do
|
|
SiteSetting.overridden_robots_txt = "blah whatever"
|
|
get "/robots.txt"
|
|
expect(response.status).to eq(200)
|
|
expect(response.body).to eq(SiteSetting.overridden_robots_txt)
|
|
end
|
|
|
|
describe "sitemap" do
|
|
let(:sitemap_line) do
|
|
"Sitemap: #{Discourse.base_protocol}://#{Discourse.current_hostname}/sitemap.xml"
|
|
end
|
|
|
|
it "include sitemap location when enabled" do
|
|
SiteSetting.enable_sitemap = true
|
|
SiteSetting.login_required = false
|
|
|
|
get "/robots.txt"
|
|
|
|
expect(response.body).to include(sitemap_line)
|
|
end
|
|
|
|
it "doesn't include sitemap location when disabled" do
|
|
SiteSetting.enable_sitemap = false
|
|
SiteSetting.login_required = false
|
|
|
|
get "/robots.txt"
|
|
|
|
expect(response.body).not_to include(sitemap_line)
|
|
end
|
|
|
|
it "doesn't include sitemap location when site has login_required enabled" do
|
|
SiteSetting.enable_sitemap = true
|
|
SiteSetting.login_required = true
|
|
|
|
get "/robots.txt"
|
|
|
|
expect(response.body).not_to include(sitemap_line)
|
|
end
|
|
end
|
|
|
|
describe "plugins" do
|
|
let(:event_handler) do
|
|
Proc.new { |robots_info| robots_info[:agents] << { name: "Test", disallow: ["/test/"] } }
|
|
end
|
|
|
|
before { DiscourseEvent.on(:robots_info, &event_handler) }
|
|
|
|
after { DiscourseEvent.off(:robots_info, &event_handler) }
|
|
|
|
it "can add to robots.txt" do
|
|
get "/robots.txt"
|
|
|
|
expect(response.parsed_body).to include("User-agent: Test\nDisallow: /test/")
|
|
end
|
|
end
|
|
end
|
|
end
|