debian-mirror-gitlab/spec/lib/gitlab/conflict/file_spec.rb

280 lines
9.6 KiB
Ruby
Raw Normal View History

2016-09-13 17:45:13 +05:30
require 'spec_helper'
2017-09-10 17:25:29 +05:30
describe Gitlab::Conflict::File do
2017-08-17 22:00:37 +05:30
let(:project) { create(:project, :repository) }
2016-09-13 17:45:13 +05:30
let(:repository) { project.repository }
let(:rugged) { repository.rugged }
let(:their_commit) { rugged.branches['conflict-start'].target }
let(:our_commit) { rugged.branches['conflict-resolvable'].target }
let(:merge_request) { create(:merge_request, source_branch: 'conflict-resolvable', target_branch: 'conflict-start', source_project: project) }
let(:index) { rugged.merge_commits(our_commit, their_commit) }
2018-03-17 18:26:18 +05:30
let(:rugged_conflict) { index.conflicts.last }
let(:raw_conflict_content) { index.merge_file('files/ruby/regex.rb')[:data] }
let(:raw_conflict_file) { Gitlab::Git::Conflict::File.new(repository, our_commit.oid, rugged_conflict, raw_conflict_content) }
let(:conflict_file) { described_class.new(raw_conflict_file, merge_request: merge_request) }
2016-09-13 17:45:13 +05:30
describe '#resolve_lines' do
let(:section_keys) { conflict_file.sections.map { |section| section[:id] }.compact }
context 'when resolving everything to the same side' do
let(:resolution_hash) { section_keys.map { |key| [key, 'head'] }.to_h }
let(:resolved_lines) { conflict_file.resolve_lines(resolution_hash) }
let(:expected_lines) { conflict_file.lines.reject { |line| line.type == 'old' } }
it 'has the correct number of lines' do
expect(resolved_lines.length).to eq(expected_lines.length)
end
it 'has content matching the chosen lines' do
expect(resolved_lines.map(&:text)).to eq(expected_lines.map(&:text))
end
end
context 'with mixed resolutions' do
let(:resolution_hash) do
section_keys.map.with_index { |key, i| [key, i.even? ? 'head' : 'origin'] }.to_h
end
let(:resolved_lines) { conflict_file.resolve_lines(resolution_hash) }
it 'has the correct number of lines' do
file_lines = conflict_file.lines.reject { |line| line.type == 'new' }
expect(resolved_lines.length).to eq(file_lines.length)
end
it 'returns a file containing only the chosen parts of the resolved sections' do
2017-09-10 17:25:29 +05:30
expect(resolved_lines.chunk { |line| line.type || 'both' }.map(&:first))
.to eq(%w(both new both old both new both))
2016-09-13 17:45:13 +05:30
end
end
2018-03-17 18:26:18 +05:30
it 'raises ResolutionError when passed a hash without resolutions for all sections' do
2016-09-13 17:45:13 +05:30
empty_hash = section_keys.map { |key| [key, nil] }.to_h
invalid_hash = section_keys.map { |key| [key, 'invalid'] }.to_h
2017-09-10 17:25:29 +05:30
expect { conflict_file.resolve_lines({}) }
2018-03-17 18:26:18 +05:30
.to raise_error(Gitlab::Git::Conflict::Resolver::ResolutionError)
2016-09-13 17:45:13 +05:30
2017-09-10 17:25:29 +05:30
expect { conflict_file.resolve_lines(empty_hash) }
2018-03-17 18:26:18 +05:30
.to raise_error(Gitlab::Git::Conflict::Resolver::ResolutionError)
2016-09-13 17:45:13 +05:30
2017-09-10 17:25:29 +05:30
expect { conflict_file.resolve_lines(invalid_hash) }
2018-03-17 18:26:18 +05:30
.to raise_error(Gitlab::Git::Conflict::Resolver::ResolutionError)
2016-09-13 17:45:13 +05:30
end
end
describe '#highlight_lines!' do
def html_to_text(html)
CGI.unescapeHTML(ActionView::Base.full_sanitizer.sanitize(html)).delete("\n")
end
it 'modifies the existing lines' do
expect { conflict_file.highlight_lines! }.to change { conflict_file.lines.map(&:instance_variables) }
end
it 'is called implicitly when rich_text is accessed on a line' do
expect(conflict_file).to receive(:highlight_lines!).once.and_call_original
conflict_file.lines.each(&:rich_text)
end
it 'sets the rich_text of the lines matching the text content' do
conflict_file.lines.each do |line|
expect(line.text).to eq(html_to_text(line.rich_text))
end
end
2018-03-17 18:26:18 +05:30
# This spec will break if Rouge's highlighting changes, but we need to
# ensure that the lines are actually highlighted.
it 'highlights the lines correctly' do
expect(conflict_file.lines.first.rich_text)
.to eq("<span id=\"LC1\" class=\"line\" lang=\"ruby\"><span class=\"k\">module</span> <span class=\"nn\">Gitlab</span></span>\n")
end
2016-09-13 17:45:13 +05:30
end
describe '#sections' do
it 'only inserts match lines when there is a gap between sections' do
conflict_file.sections.each_with_index do |section, i|
previous_line_number = 0
current_line_number = section[:lines].map(&:old_line).compact.min
if i > 0
previous_line_number = conflict_file.sections[i - 1][:lines].map(&:old_line).compact.last
end
if current_line_number == previous_line_number + 1
expect(section[:lines].first.type).not_to eq('match')
else
expect(section[:lines].first.type).to eq('match')
expect(section[:lines].first.text).to match(/\A@@ -#{current_line_number},\d+ \+\d+,\d+ @@ module Gitlab\Z/)
end
end
end
it 'sets conflict to false for sections with only unchanged lines' do
conflict_file.sections.reject { |section| section[:conflict] }.each do |section|
without_match = section[:lines].reject { |line| line.type == 'match' }
expect(without_match).to all(have_attributes(type: nil))
end
end
it 'only includes a maximum of CONTEXT_LINES (plus an optional match line) in context sections' do
conflict_file.sections.reject { |section| section[:conflict] }.each do |section|
without_match = section[:lines].reject { |line| line.type == 'match' }
expect(without_match.length).to be <= Gitlab::Conflict::File::CONTEXT_LINES * 2
end
end
it 'sets conflict to true for sections with only changed lines' do
conflict_file.sections.select { |section| section[:conflict] }.each do |section|
section[:lines].each do |line|
2017-08-17 22:00:37 +05:30
expect(line.type).to be_in(%w(new old))
2016-09-13 17:45:13 +05:30
end
end
end
it 'adds unique IDs to conflict sections, and not to other sections' do
section_ids = []
conflict_file.sections.each do |section|
if section[:conflict]
expect(section).to have_key(:id)
section_ids << section[:id]
else
expect(section).not_to have_key(:id)
end
end
expect(section_ids.uniq).to eq(section_ids)
end
context 'with an example file' do
2018-03-17 18:26:18 +05:30
let(:raw_conflict_content) do
2016-09-13 17:45:13 +05:30
<<FILE
# Ensure there is no match line header here
def username_regexp
default_regexp
end
<<<<<<< files/ruby/regex.rb
def project_name_regexp
/\A[a-zA-Z0-9][a-zA-Z0-9_\-\. ]*\z/
end
def name_regexp
/\A[a-zA-Z0-9_\-\. ]*\z/
=======
def project_name_regex
%r{\A[a-zA-Z0-9][a-zA-Z0-9_\-\. ]*\z}
end
def name_regex
%r{\A[a-zA-Z0-9_\-\. ]*\z}
>>>>>>> files/ruby/regex.rb
end
# Some extra lines
# To force a match line
# To be created
def path_regexp
default_regexp
end
<<<<<<< files/ruby/regex.rb
def archive_formats_regexp
/(zip|tar|7z|tar\.gz|tgz|gz|tar\.bz2|tbz|tbz2|tb2|bz2)/
=======
def archive_formats_regex
%r{(zip|tar|7z|tar\.gz|tgz|gz|tar\.bz2|tbz|tbz2|tb2|bz2)}
>>>>>>> files/ruby/regex.rb
end
def git_reference_regexp
# Valid git ref regexp, see:
# https://www.kernel.org/pub/software/scm/git/docs/git-check-ref-format.html
%r{
(?!
(?# doesn't begins with)
\/| (?# rule #6)
(?# doesn't contain)
.*(?:
[\/.]\.| (?# rule #1,3)
\/\/| (?# rule #6)
@\{| (?# rule #8)
\\ (?# rule #9)
)
)
[^\000-\040\177~^:?*\[]+ (?# rule #4-5)
(?# doesn't end with)
(?<!\.lock) (?# rule #1)
(?<![\/.]) (?# rule #6-7)
}x
end
protected
<<<<<<< files/ruby/regex.rb
def default_regexp
/\A[.?]?[a-zA-Z0-9][a-zA-Z0-9_\-\.]*(?<!\.git)\z/
=======
def default_regex
%r{\A[.?]?[a-zA-Z0-9][a-zA-Z0-9_\-\.]*(?<!\.git)\z}
>>>>>>> files/ruby/regex.rb
end
FILE
end
let(:sections) { conflict_file.sections }
it 'sets the correct match line headers' do
expect(sections[0][:lines].first).to have_attributes(type: 'match', text: '@@ -3,14 +3,14 @@')
expect(sections[3][:lines].first).to have_attributes(type: 'match', text: '@@ -19,26 +19,26 @@ def path_regexp')
expect(sections[6][:lines].first).to have_attributes(type: 'match', text: '@@ -47,52 +47,52 @@ end')
end
it 'does not add match lines where they are not needed' do
expect(sections[1][:lines].first.type).not_to eq('match')
expect(sections[2][:lines].first.type).not_to eq('match')
expect(sections[4][:lines].first.type).not_to eq('match')
expect(sections[5][:lines].first.type).not_to eq('match')
expect(sections[7][:lines].first.type).not_to eq('match')
end
it 'creates context sections of the correct length' do
expect(sections[0][:lines].reject(&:type).length).to eq(3)
expect(sections[2][:lines].reject(&:type).length).to eq(3)
expect(sections[3][:lines].reject(&:type).length).to eq(3)
expect(sections[5][:lines].reject(&:type).length).to eq(3)
expect(sections[6][:lines].reject(&:type).length).to eq(3)
expect(sections[8][:lines].reject(&:type).length).to eq(1)
end
end
end
describe '#as_json' do
it 'includes the blob path for the file' do
2017-09-10 17:25:29 +05:30
expect(conflict_file.as_json[:blob_path])
.to eq("/#{project.full_path}/blob/#{our_commit.oid}/files/ruby/regex.rb")
2016-09-13 17:45:13 +05:30
end
it 'includes the blob icon for the file' do
expect(conflict_file.as_json[:blob_icon]).to eq('file-text-o')
end
2016-11-03 12:29:30 +05:30
context 'with the full_content option passed' do
it 'includes the full content of the conflict' do
expect(conflict_file.as_json(full_content: true)).to have_key(:content)
end
it 'includes the detected language of the conflict file' do
2017-09-10 17:25:29 +05:30
expect(conflict_file.as_json(full_content: true)[:blob_ace_mode])
.to eq('ruby')
2016-11-03 12:29:30 +05:30
end
end
2016-09-13 17:45:13 +05:30
end
end