gitlab-org--gitlab-foss/spec/lib/gitlab/github_import/client_spec.rb

533 lines
15 KiB
Ruby

# frozen_string_literal: true
require 'spec_helper'
RSpec.describe Gitlab::GithubImport::Client do
describe '#parallel?' do
it 'returns true when the client is running in parallel mode' do
client = described_class.new('foo', parallel: true)
expect(client).to be_parallel
end
it 'returns false when the client is running in sequential mode' do
client = described_class.new('foo', parallel: false)
expect(client).not_to be_parallel
end
end
describe '#user' do
it 'returns the details for the given username' do
client = described_class.new('foo')
expect(client.octokit).to receive(:user).with('foo')
expect(client).to receive(:with_rate_limit).and_yield
client.user('foo')
end
end
describe '#pull_request_reviews' do
it 'returns the pull request reviews' do
client = described_class.new('foo')
expect(client.octokit).to receive(:pull_request_reviews).with('foo/bar', 999)
expect(client).to receive(:with_rate_limit).and_yield
client.pull_request_reviews('foo/bar', 999)
end
end
describe '#repository' do
it 'returns the details of a repository' do
client = described_class.new('foo')
expect(client.octokit).to receive(:repo).with('foo/bar')
expect(client).to receive(:with_rate_limit).and_yield
client.repository('foo/bar')
end
end
describe '#pull_request' do
it 'returns the details of a pull_request' do
client = described_class.new('foo')
expect(client.octokit).to receive(:pull_request).with('foo/bar', 999)
expect(client).to receive(:with_rate_limit).and_yield
client.pull_request('foo/bar', 999)
end
end
describe '#labels' do
it 'returns the labels' do
client = described_class.new('foo')
expect(client)
.to receive(:each_object)
.with(:labels, 'foo/bar')
client.labels('foo/bar')
end
end
describe '#milestones' do
it 'returns the milestones' do
client = described_class.new('foo')
expect(client)
.to receive(:each_object)
.with(:milestones, 'foo/bar')
client.milestones('foo/bar')
end
end
describe '#releases' do
it 'returns the releases' do
client = described_class.new('foo')
expect(client)
.to receive(:each_object)
.with(:releases, 'foo/bar')
client.releases('foo/bar')
end
end
describe '#each_page' do
let(:client) { described_class.new('foo') }
let(:object1) { double(:object1) }
let(:object2) { double(:object2) }
before do
allow(client)
.to receive(:with_rate_limit)
.and_yield
allow(client.octokit)
.to receive(:public_send)
.and_return([object1])
response = double(:response, data: [object2], rels: { next: nil })
next_page = double(:next_page, get: response)
allow(client.octokit)
.to receive(:last_response)
.and_return(double(:last_response, rels: { next: next_page }))
end
context 'without a block' do
it 'returns an Enumerator' do
expect(client.each_page(:foo)).to be_an_instance_of(Enumerator)
end
it 'the returned Enumerator returns Page objects' do
enum = client.each_page(:foo)
page1 = enum.next
page2 = enum.next
expect(page1).to be_an_instance_of(described_class::Page)
expect(page2).to be_an_instance_of(described_class::Page)
expect(page1.objects).to eq([object1])
expect(page1.number).to eq(1)
expect(page2.objects).to eq([object2])
expect(page2.number).to eq(2)
end
end
context 'with a block' do
it 'yields every retrieved page to the supplied block' do
pages = []
client.each_page(:foo) { |page| pages << page }
expect(pages[0]).to be_an_instance_of(described_class::Page)
expect(pages[1]).to be_an_instance_of(described_class::Page)
expect(pages[0].objects).to eq([object1])
expect(pages[0].number).to eq(1)
expect(pages[1].objects).to eq([object2])
expect(pages[1].number).to eq(2)
end
it 'starts at the given page' do
pages = []
client.each_page(:foo, page: 2) { |page| pages << page }
expect(pages[0].number).to eq(2)
expect(pages[1].number).to eq(3)
end
end
end
describe '#with_rate_limit' do
let(:client) { described_class.new('foo') }
it 'yields the supplied block when enough requests remain' do
expect(client).to receive(:requests_remaining?).and_return(true)
expect { |b| client.with_rate_limit(&b) }.to yield_control
end
it 'waits before yielding if not enough requests remain' do
expect(client).to receive(:requests_remaining?).and_return(false)
expect(client).to receive(:raise_or_wait_for_rate_limit)
expect { |b| client.with_rate_limit(&b) }.to yield_control
end
it 'waits and retries the operation if all requests were consumed in the supplied block' do
retries = 0
expect(client).to receive(:requests_remaining?).and_return(true)
expect(client).to receive(:raise_or_wait_for_rate_limit)
client.with_rate_limit do
if retries == 0
retries += 1
raise(Octokit::TooManyRequests)
end
end
expect(retries).to eq(1)
end
it 'increments the request count counter' do
expect(client.request_count_counter)
.to receive(:increment)
.and_call_original
expect(client).to receive(:requests_remaining?).and_return(true)
client.with_rate_limit { }
end
it 'ignores rate limiting when disabled' do
expect(client)
.to receive(:rate_limiting_enabled?)
.and_return(false)
expect(client)
.not_to receive(:requests_remaining?)
expect(client.with_rate_limit { 10 }).to eq(10)
end
end
describe '#requests_remaining?' do
let(:client) { described_class.new('foo') }
context 'when default requests limit is set' do
before do
allow(client).to receive(:requests_limit).and_return(5000)
end
it 'returns true if enough requests remain' do
expect(client).to receive(:remaining_requests).and_return(9000)
expect(client.requests_remaining?).to eq(true)
end
it 'returns false if not enough requests remain' do
expect(client).to receive(:remaining_requests).and_return(1)
expect(client.requests_remaining?).to eq(false)
end
end
context 'when search requests limit is set' do
before do
allow(client).to receive(:requests_limit).and_return(described_class::SEARCH_MAX_REQUESTS_PER_MINUTE)
end
it 'returns true if enough requests remain' do
expect(client).to receive(:remaining_requests).and_return(described_class::SEARCH_RATE_LIMIT_THRESHOLD + 1)
expect(client.requests_remaining?).to eq(true)
end
it 'returns false if not enough requests remain' do
expect(client).to receive(:remaining_requests).and_return(described_class::SEARCH_RATE_LIMIT_THRESHOLD - 1)
expect(client.requests_remaining?).to eq(false)
end
end
end
describe '#raise_or_wait_for_rate_limit' do
it 'raises RateLimitError when running in parallel mode' do
client = described_class.new('foo', parallel: true)
expect { client.raise_or_wait_for_rate_limit }
.to raise_error(Gitlab::GithubImport::RateLimitError)
end
it 'sleeps when running in sequential mode' do
client = described_class.new('foo', parallel: false)
expect(client).to receive(:rate_limit_resets_in).and_return(1)
expect(client).to receive(:sleep).with(1)
client.raise_or_wait_for_rate_limit
end
it 'increments the rate limit counter' do
client = described_class.new('foo', parallel: false)
expect(client)
.to receive(:rate_limit_resets_in)
.and_return(1)
expect(client)
.to receive(:sleep)
.with(1)
expect(client.rate_limit_counter)
.to receive(:increment)
.and_call_original
client.raise_or_wait_for_rate_limit
end
end
describe '#remaining_requests' do
it 'returns the number of remaining requests' do
client = described_class.new('foo')
rate_limit = double(remaining: 1)
expect(client.octokit).to receive(:rate_limit).and_return(rate_limit)
expect(client.remaining_requests).to eq(1)
end
end
describe '#requests_limit' do
it 'returns requests limit' do
client = described_class.new('foo')
rate_limit = double(limit: 1)
expect(client.octokit).to receive(:rate_limit).and_return(rate_limit)
expect(client.requests_limit).to eq(1)
end
end
describe '#rate_limit_resets_in' do
it 'returns the number of seconds after which the rate limit is reset' do
client = described_class.new('foo')
rate_limit = double(resets_in: 1)
expect(client.octokit).to receive(:rate_limit).and_return(rate_limit)
expect(client.rate_limit_resets_in).to eq(6)
end
end
describe '#api_endpoint' do
let(:client) { described_class.new('foo') }
context 'without a custom endpoint configured in Omniauth' do
it 'returns the default API endpoint' do
expect(client)
.to receive(:custom_api_endpoint)
.and_return(nil)
expect(client.api_endpoint).to eq('https://api.github.com')
end
end
context 'with a custom endpoint configured in Omniauth' do
it 'returns the custom endpoint' do
endpoint = 'https://github.kittens.com'
expect(client)
.to receive(:custom_api_endpoint)
.and_return(endpoint)
expect(client.api_endpoint).to eq(endpoint)
end
end
end
describe '#web_endpoint' do
let(:client) { described_class.new('foo') }
context 'without a custom endpoint configured in Omniauth' do
it 'returns the default web endpoint' do
expect(client)
.to receive(:custom_api_endpoint)
.and_return(nil)
expect(client.web_endpoint).to eq('https://github.com')
end
end
context 'with a custom endpoint configured in Omniauth' do
it 'returns the custom endpoint' do
endpoint = 'https://github.kittens.com'
expect(client)
.to receive(:custom_api_endpoint)
.and_return(endpoint)
expect(client.web_endpoint).to eq(endpoint)
end
end
end
describe '#custom_api_endpoint' do
let(:client) { described_class.new('foo') }
context 'without a custom endpoint' do
it 'returns nil' do
expect(client)
.to receive(:github_omniauth_provider)
.and_return({})
expect(client.custom_api_endpoint).to be_nil
end
end
context 'with a custom endpoint' do
it 'returns the API endpoint' do
endpoint = 'https://github.kittens.com'
expect(client)
.to receive(:github_omniauth_provider)
.and_return({ 'args' => { 'client_options' => { 'site' => endpoint } } })
expect(client.custom_api_endpoint).to eq(endpoint)
end
end
end
describe '#default_api_endpoint' do
it 'returns the default API endpoint' do
client = described_class.new('foo')
expect(client.default_api_endpoint).to eq('https://api.github.com')
end
end
describe '#verify_ssl' do
let(:client) { described_class.new('foo') }
context 'without a custom configuration' do
it 'returns true' do
expect(client)
.to receive(:github_omniauth_provider)
.and_return({})
expect(client.verify_ssl).to eq(true)
end
end
context 'with a custom configuration' do
it 'returns the configured value' do
expect(client.verify_ssl).to eq(false)
end
end
end
describe '#github_omniauth_provider' do
let(:client) { described_class.new('foo') }
context 'without a configured provider' do
it 'returns an empty Hash' do
expect(Gitlab.config.omniauth)
.to receive(:providers)
.and_return([])
expect(client.github_omniauth_provider).to eq({})
end
end
context 'with a configured provider' do
it 'returns the provider details as a Hash' do
hash = client.github_omniauth_provider
expect(hash['name']).to eq('github')
expect(hash['url']).to eq('https://github.com/')
end
end
end
describe '#rate_limiting_enabled?' do
let(:client) { described_class.new('foo') }
it 'returns true when using GitHub.com' do
expect(client.rate_limiting_enabled?).to eq(true)
end
it 'returns false for GitHub enterprise installations' do
expect(client)
.to receive(:api_endpoint)
.and_return('https://github.kittens.com/')
expect(client.rate_limiting_enabled?).to eq(false)
end
end
describe 'search' do
let(:client) { described_class.new('foo') }
let(:user) { double(:user, login: 'user') }
let(:org1) { double(:org, login: 'org1') }
let(:org2) { double(:org, login: 'org2') }
let(:repo1) { double(:repo, full_name: 'repo1') }
let(:repo2) { double(:repo, full_name: 'repo2') }
before do
allow(client)
.to receive(:each_object)
.with(:repos, nil, { affiliation: 'collaborator' })
.and_return([repo1, repo2].to_enum)
allow(client)
.to receive(:each_object)
.with(:organizations)
.and_return([org1, org2].to_enum)
allow(client.octokit).to receive(:user).and_return(user)
end
describe '#search_repos_by_name' do
it 'searches for repositories based on name' do
expected_search_query = 'test in:name is:public,private user:user repo:repo1 repo:repo2 org:org1 org:org2'
expect(client.octokit).to receive(:search_repositories).with(expected_search_query, {})
client.search_repos_by_name('test')
end
end
describe '#search_query' do
it 'returns base search query' do
result = client.search_query(str: 'test', type: :test, include_collaborations: false, include_orgs: false)
expect(result).to eq('test in:test is:public,private user:user')
end
context 'when include_collaborations is true' do
it 'returns search query including users' do
result = client.search_query(str: 'test', type: :test, include_collaborations: true, include_orgs: false)
expect(result).to eq('test in:test is:public,private user:user repo:repo1 repo:repo2')
end
end
context 'when include_orgs is true' do
it 'returns search query including users' do
result = client.search_query(str: 'test', type: :test, include_collaborations: false, include_orgs: true)
expect(result).to eq('test in:test is:public,private user:user org:org1 org:org2')
end
end
end
end
end