debian-mirror-gitlab/spec/lib/gitlab/encoding_helper_spec.rb
2022-11-25 23:54:43 +05:30

286 lines
12 KiB
Ruby
Raw Blame History

# frozen_string_literal: true
require "spec_helper"
RSpec.describe Gitlab::EncodingHelper do
using RSpec::Parameterized::TableSyntax
let(:ext_class) { Class.new { extend Gitlab::EncodingHelper } }
let(:binary_string) { File.read(Rails.root + "spec/fixtures/dk.png") }
describe '#encode!' do
[
["nil", nil, nil],
["empty string", "".encode("ASCII-8BIT"), "".encode("UTF-8")],
["invalid utf-8 encoded string", (+"my bad string\xE5").force_encoding("UTF-8"), "my bad string"],
["frozen non-ascii string", (+"é").force_encoding("ASCII-8BIT").freeze, "é".encode("UTF-8")],
[
'leaves ascii only string as is',
'ascii only string',
'ascii only string'
],
[
'leaves valid utf8 string as is',
'multibyte string №∑∉',
'multibyte string №∑∉'
],
[
'removes invalid bytes from ASCII-8bit encoded multibyte string. This can occur when a git diff match line truncates in the middle of a multibyte character. This occurs after the second word in this example. The test string is as short as we can get while still triggering the error condition when not looking at `detect[:confidence]`.',
(+"mu ns\xC3\n Lorem ipsum dolor sit amet, consectetur adipisicing ut\xC3\xA0y\xC3\xB9abcd\xC3\xB9efg kia elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non p\n {: .normal_pn}\n \n-Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in\n# *Lorem ipsum\xC3\xB9l\xC3\xB9l\xC3\xA0 dolor\xC3\xB9k\xC3\xB9 sit\xC3\xA8b\xC3\xA8 N\xC3\xA8 amet b\xC3\xA0d\xC3\xAC*\n+# *consectetur\xC3\xB9l\xC3\xB9l\xC3\xA0 adipisicing\xC3\xB9k\xC3\xB9 elit\xC3\xA8b\xC3\xA8 N\xC3\xA8 sed do\xC3\xA0d\xC3\xAC*{: .italic .smcaps}\n \n \xEF\x9B\xA1 eiusmod tempor incididunt, ut\xC3\xAAn\xC3\xB9 labore et dolore. Tw\xC4\x83nj\xC3\xAC magna aliqua. Ut enim ad minim veniam\n {: .normal}\n@@ -9,5 +9,5 @@ quis nostrud\xC3\xAAt\xC3\xB9 exercitiation ullamco laboris m\xC3\xB9s\xC3\xB9k\xC3\xB9abc\xC3\xB9 nisi ").force_encoding('ASCII-8BIT'),
"mu ns\n Lorem ipsum dolor sit amet, consectetur adipisicing ut\xC3\xA0y\xC3\xB9abcd\xC3\xB9efg kia elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non p\n {: .normal_pn}\n \n-Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in\n# *Lorem ipsum\xC3\xB9l\xC3\xB9l\xC3\xA0 dolor\xC3\xB9k\xC3\xB9 sit\xC3\xA8b\xC3\xA8 N\xC3\xA8 amet b\xC3\xA0d\xC3\xAC*\n+# *consectetur\xC3\xB9l\xC3\xB9l\xC3\xA0 adipisicing\xC3\xB9k\xC3\xB9 elit\xC3\xA8b\xC3\xA8 N\xC3\xA8 sed do\xC3\xA0d\xC3\xAC*{: .italic .smcaps}\n \n \xEF\x9B\xA1 eiusmod tempor incididunt, ut\xC3\xAAn\xC3\xB9 labore et dolore. Tw\xC4\x83nj\xC3\xAC magna aliqua. Ut enim ad minim veniam\n {: .normal}\n@@ -9,5 +9,5 @@ quis nostrud\xC3\xAAt\xC3\xB9 exercitiation ullamco laboris m\xC3\xB9s\xC3\xB9k\xC3\xB9abc\xC3\xB9 nisi "
],
[
'string with detected encoding that is not supported in Ruby',
"\xFFe,i\xFF,\xB8oi,'\xB8,\xFF,-",
"--broken encoding: IBM420_ltr"
]
].each do |description, test_string, xpect|
it description do
expect(ext_class.encode!(test_string)).to eq(xpect)
end
end
it 'leaves binary string as is' do
expect(ext_class.encode!(binary_string)).to eq(binary_string)
end
context 'with corrupted diff' do
let(:project) { create(:project, :empty_repo) }
let(:repository) { project.repository }
let(:content) { fixture_file('encoding/Japanese.md') }
let(:corrupted_diff) do
commit_a = repository.create_file(
project.creator,
'Japanese.md',
content,
branch_name: 'HEAD',
message: 'Create Japanese.md'
)
commit_b = repository.update_file(
project.creator,
'Japanese.md',
content.sub('[TODO: Link]', '[現在作業中です: Link]'),
branch_name: 'HEAD',
message: 'Update Japanese.md'
)
repository.diff(commit_a, commit_b).map(&:diff).join
end
let(:cleaned_diff) do
corrupted_diff.dup.force_encoding('UTF-8')
.encode!('UTF-8', invalid: :replace, replace: '')
end
let(:encoded_diff) do
described_class.encode!(corrupted_diff.dup)
end
it 'does not corrupt data but remove invalid characters' do
expect(encoded_diff).to eq(cleaned_diff)
end
end
end
describe '#encode_utf8_no_detect' do
where(:input, :expected) do
"abcd" | "abcd"
"DzDzDz" | "DzDzDz"
"\xC7\xB2\xC7DzDzDz" | "Dz<EFBFBD>DzDzDz"
"🐤🐤🐤🐤\xF0\x9F\x90" | "🐤🐤🐤🐤<EFBFBD>"
end
with_them do
it 'drops invalid UTF-8' do
expect(ext_class.encode_utf8_no_detect(input.dup.force_encoding(Encoding::ASCII_8BIT))).to eq(expected)
expect(ext_class.encode_utf8_no_detect(input)).to eq(expected)
end
end
end
describe '#encode_utf8_with_escaping!' do
where(:input, :expected) do
"abcd" | "abcd"
"DzDzDz" | "DzDzDz"
"\xC7\xB2\xC7DzDzDz" | "Dz%C7DzDzDz"
"🐤🐤🐤🐤\xF0\x9F\x90" | "🐤🐤🐤🐤%F0%9F%90"
"\xD0\x9F\xD1\x80 \x90" | "Пр %90"
"\x41" | "A"
end
with_them do
it 'escapes invalid UTF-8' do
expect(ext_class.encode_utf8_with_escaping!(input.dup.force_encoding(Encoding::ASCII_8BIT))).to eq(expected)
expect(ext_class.encode_utf8_with_escaping!(input)).to eq(expected)
end
end
end
describe '#encode_utf8' do
[
["nil", nil, nil],
["empty string", "".encode("ASCII-8BIT"), "".encode("UTF-8")],
["invalid utf-8 encoded string", (+"my bad string\xE5").force_encoding("UTF-8"), "my bad stringå"],
[
"encodes valid utf8 encoded string to utf8",
"λ, λ, λ".encode("UTF-8"),
"λ, λ, λ".encode("UTF-8")
],
[
"encodes valid ASCII-8BIT encoded string to utf8",
"ascii only".encode("ASCII-8BIT"),
"ascii only".encode("UTF-8")
],
[
"encodes valid ISO-8859-1 encoded string to utf8",
"Rüby ist eine Programmiersprache. Wir verlängern den text damit ICU die Sprache erkennen kann.".encode("ISO-8859-1", "UTF-8"),
"Rüby ist eine Programmiersprache. Wir verlängern den text damit ICU die Sprache erkennen kann.".encode("UTF-8")
],
[
# Test case from https://gitlab.com/gitlab-org/gitlab-foss/issues/39227
"Equifax branch name",
"refs/heads/Equifax".encode("UTF-8"),
"refs/heads/Equifax".encode("UTF-8")
]
].each do |description, test_string, xpect|
it description do
r = ext_class.encode_utf8(test_string)
expect(r).to eq(xpect)
expect(r.encoding.name).to eq('UTF-8') if xpect
end
end
it 'returns empty string on conversion errors' do
expect { ext_class.encode_utf8('') }.not_to raise_error
end
it 'replaces invalid and undefined chars with the replace argument' do
str = 'hællo'.encode(Encoding::UTF_16LE).force_encoding(Encoding::ASCII_8BIT)
expect(ext_class.encode_utf8(str, replace: "\u{FFFD}")).to eq("h<EFBFBD>llo")
end
context 'with strings that can be forcefully encoded into utf8' do
let(:test_string) do
"refs/heads/FixSymbolsTitleDropdown".encode("ASCII-8BIT")
end
let(:expected_string) do
"refs/heads/FixSymbolsTitleDropdown".encode("UTF-8")
end
subject { ext_class.encode_utf8(test_string) }
it "doesn't use CharlockHolmes if the encoding can be forced into utf_8" do
expect(CharlockHolmes::EncodingDetector).not_to receive(:detect)
expect(subject).to eq(expected_string)
expect(subject.encoding.name).to eq('UTF-8')
end
end
end
describe '#clean' do
[
[
'leaves ascii only string as is',
'ascii only string',
'ascii only string'
],
[
'leaves valid utf8 string as is',
'multibyte string №∑∉',
'multibyte string №∑∉'
],
[
'removes invalid bytes from ASCII-8bit encoded multibyte string.',
(+"Lorem ipsum\xC3\n dolor sit amet, xy\xC3\xA0y\xC3\xB9abcd\xC3\xB9efg").force_encoding('ASCII-8BIT'),
"Lorem ipsum\n dolor sit amet, xyàyùabcdùefg"
],
[
'handles UTF-16BE encoded strings',
(+"\xFE\xFF\x00\x41").force_encoding('ASCII-8BIT'), # An "A" prepended with UTF-16 BOM
"\xEF\xBB\xBFA" # An "A" prepended with UTF-8 BOM
]
].each do |description, test_string, xpect|
it description do
expect(ext_class.encode!(test_string)).to eq(xpect)
end
end
end
describe 'encode_binary' do
[
[nil, ""],
["", ""],
[" ", " "],
%w(a1 a1),
["编码", "\xE7\xBC\x96\xE7\xA0\x81".b]
].each do |input, result|
it "encodes #{input.inspect} to #{result.inspect}" do
expect(ext_class.encode_binary(input)).to eq(result)
end
end
end
describe '#binary_io' do
it 'does not mutate the original string encoding' do
test = 'my-test'
io_stream = ext_class.binary_io(test)
expect(io_stream.external_encoding.name).to eq('ASCII-8BIT')
expect(test.encoding.name).to eq('UTF-8')
end
it 'returns a copy of the IO with the correct encoding' do
test = fixture_file_upload('spec/fixtures/doc_sample.txt').to_io
io_stream = ext_class.binary_io(test)
expect(io_stream.external_encoding.name).to eq('ASCII-8BIT')
expect(test).not_to eq(io_stream)
end
end
describe '#detect_encoding' do
subject { ext_class.detect_encoding(data, **kwargs) }
let(:data) { binary_string }
let(:kwargs) { {} }
context 'detects encoding' do
it { is_expected.to be_a(Hash) }
it 'correctly detects the binary' do
expect(subject[:type]).to eq(:binary)
end
context 'data is nil' do
let(:data) { nil }
it { is_expected.to be_nil }
end
context 'limit is provided' do
let(:kwargs) do
{ limit: 10 }
end
it 'correctly detects the binary' do
expect(subject[:type]).to eq(:binary)
end
end
end
end
describe '#unquote_path' do
it do
expect(described_class.unquote_path('unquoted')).to eq('unquoted')
expect(described_class.unquote_path('"quoted"')).to eq('quoted')
expect(described_class.unquote_path('"\\311\\240\\304\\253\\305\\247\\305\\200\\310\\247\\306\\200"')).to eq('ɠīŧŀȧƀ')
expect(described_class.unquote_path('"\\\\303\\\\251"')).to eq('\303\251')
expect(described_class.unquote_path('"\a\b\e\f\n\r\t\v\""')).to eq("\a\b\e\f\n\r\t\v\"")
end
end
end