debian-mirror-gitlab/spec/lib/csv_builder_spec.rb

122 lines
3.7 KiB
Ruby
Raw Normal View History

2020-04-22 19:07:51 +05:30
# frozen_string_literal: true
require 'spec_helper'
2020-07-28 23:09:34 +05:30
RSpec.describe CsvBuilder do
2020-04-22 19:07:51 +05:30
let(:object) { double(question: :answer) }
let(:fake_relation) { FakeRelation.new([object]) }
let(:subject) { described_class.new(fake_relation, 'Q & A' => :question, 'Reversed' => -> (o) { o.question.to_s.reverse }) }
let(:csv_data) { subject.render }
before do
stub_const('FakeRelation', Array)
FakeRelation.class_eval do
def find_each(&block)
each(&block)
end
end
end
it 'generates a csv' do
expect(csv_data.scan(/(,|\n)/).join).to include ",\n,"
end
it 'uses a temporary file to reduce memory allocation' do
expect(CSV).to receive(:new).with(instance_of(Tempfile)).and_call_original
subject.render
end
it 'counts the number of rows' do
subject.render
expect(subject.rows_written).to eq 1
end
describe 'rows_expected' do
it 'uses rows_written if CSV rendered successfully' do
subject.render
expect(fake_relation).not_to receive(:count)
expect(subject.rows_expected).to eq 1
end
it 'falls back to calling .count before rendering begins' do
expect(subject.rows_expected).to eq 1
end
end
describe 'truncation' do
let(:big_object) { double(question: 'Long' * 1024) }
let(:row_size) { big_object.question.length * 2 }
let(:fake_relation) { FakeRelation.new([big_object, big_object, big_object]) }
it 'occurs after given number of bytes' do
expect(subject.render(row_size * 2).length).to be_between(row_size * 2, row_size * 3)
expect(subject).to be_truncated
expect(subject.rows_written).to eq 2
end
it 'is ignored by default' do
expect(subject.render.length).to be > row_size * 3
expect(subject.rows_written).to eq 3
end
it 'causes rows_expected to fall back to .count' do
subject.render(0)
expect(fake_relation).to receive(:count).and_call_original
expect(subject.rows_expected).to eq 3
end
end
it 'avoids loading all data in a single query' do
expect(fake_relation).to receive(:find_each)
subject.render
end
it 'uses hash keys as headers' do
expect(csv_data).to start_with 'Q & A'
end
it 'gets data by calling method provided as hash value' do
expect(csv_data).to include 'answer'
end
it 'allows lamdas to look up more complicated data' do
expect(csv_data).to include 'rewsna'
end
describe 'excel sanitization' do
let(:dangerous_title) { double(title: "=cmd|' /C calc'!A0 title", description: "*safe_desc") }
let(:dangerous_desc) { double(title: "*safe_title", description: "=cmd|' /C calc'!A0 desc") }
let(:fake_relation) { FakeRelation.new([dangerous_title, dangerous_desc]) }
let(:subject) { described_class.new(fake_relation, 'Title' => 'title', 'Description' => 'description') }
let(:csv_data) { subject.render }
it 'sanitizes dangerous characters at the beginning of a column' do
expect(csv_data).to include "'=cmd|' /C calc'!A0 title"
expect(csv_data).to include "'=cmd|' /C calc'!A0 desc"
end
it 'does not sanitize safe symbols at the beginning of a column' do
expect(csv_data).not_to include "'*safe_desc"
expect(csv_data).not_to include "'*safe_title"
end
2021-09-04 01:27:46 +05:30
context 'when dangerous characters are after a line break' do
it 'does not append single quote to description' do
fake_object = double(title: "Safe title", description: "With task list\n-[x] todo 1")
fake_relation = FakeRelation.new([fake_object])
builder = described_class.new(fake_relation, 'Title' => 'title', 'Description' => 'description')
csv_data = builder.render
expect(csv_data).to eq("Title,Description\nSafe title,\"With task list\n-[x] todo 1\"\n")
end
end
2020-04-22 19:07:51 +05:30
end
end