demarches-normaliennes/spec/services/procedure_export_service_spec.rb

536 lines
18 KiB
Ruby
Raw Normal View History

2019-06-25 15:46:10 +02:00
require 'csv'
describe ProcedureExportService do
let(:procedure) { create(:procedure, :published, :for_individual, :with_all_champs) }
let(:service) { ProcedureExportService.new(procedure, procedure.dossiers) }
2023-01-19 09:43:19 +01:00
describe 'to_xlsx' do
2019-06-25 15:46:10 +02:00
subject do
service
.to_xlsx
.open { |f| SimpleXlsxReader.open(f.path) }
2019-06-25 15:46:10 +02:00
end
2019-06-25 15:46:10 +02:00
let(:dossiers_sheet) { subject.sheets.first }
let(:etablissements_sheet) { subject.sheets.second }
let(:avis_sheet) { subject.sheets.third }
let(:repetition_sheet) { subject.sheets.fourth }
2018-12-10 15:33:34 +01:00
before do
# change one tdc place to check if the header is ordered
tdc_first = procedure.active_revision.revision_types_de_champ_public.first
tdc_last = procedure.active_revision.revision_types_de_champ_public.last
2018-12-10 15:33:34 +01:00
2020-08-27 19:57:21 +02:00
tdc_first.update(position: tdc_last.position + 1)
2018-12-20 12:00:27 +01:00
procedure.reload
2018-12-10 15:33:34 +01:00
end
describe 'sheets' do
it 'should have a sheet for each record type' do
2019-06-25 15:46:10 +02:00
expect(subject.sheets.map(&:name)).to eq(['Dossiers', 'Etablissements', 'Avis'])
end
end
describe 'Dossiers sheet' do
let!(:dossier) { create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure) }
2019-06-25 15:46:10 +02:00
let(:nominal_headers) do
[
2019-06-25 15:46:10 +02:00
"ID",
"Email",
"Civilité",
"Nom",
"Prénom",
"Archivé",
"État du dossier",
"Dernière mise à jour le",
"Déposé le",
"Passé en instruction le",
"Traité le",
"Motivation de la décision",
"Instructeurs",
"textarea",
"date",
"datetime",
"number",
"decimal_number",
"integer_number",
"checkbox",
"civilite",
"email",
"phone",
"address",
"yes_no",
"simple_drop_down_list",
"multiple_drop_down_list",
"linked_drop_down_list",
"communes",
"communes (Code insee)",
"communes (Département)",
2022-09-26 10:55:37 +02:00
"departements",
"departements (Code)",
2022-09-26 10:55:37 +02:00
"regions",
"regions (Code)",
2022-09-26 10:55:37 +02:00
"pays",
"pays (Code)",
2019-06-25 15:46:10 +02:00
"dossier_link",
"piece_justificative",
2022-09-21 15:56:01 +02:00
"rna",
2019-06-25 15:46:10 +02:00
"carte",
2020-09-17 17:09:16 +02:00
"titre_identite",
2020-09-22 16:04:57 +02:00
"iban",
2022-09-26 10:55:37 +02:00
"siret",
2021-01-14 17:29:57 +01:00
"annuaire_education",
2021-09-24 14:21:13 +02:00
"cnaf",
2021-11-24 17:30:35 +01:00
"dgfip",
"pole_emploi",
2021-12-15 15:25:02 +01:00
"mesri",
2023-01-19 09:43:19 +01:00
"text",
"epci",
"epci (Code)",
2023-05-30 14:42:36 +02:00
"epci (Département)",
"cojo"
]
end
it 'should have headers' do
expect(dossiers_sheet.headers).to match_array(nominal_headers)
2019-06-25 15:46:10 +02:00
end
it 'should have data' do
expect(dossiers_sheet.data.size).to eq(1)
expect(etablissements_sheet.data.size).to eq(1)
# SimpleXlsxReader is transforming datetimes in utc... It is only used in test so we just hack around.
offset = dossier.depose_at.utc_offset
depose_at = Time.zone.at(dossiers_sheet.data[0][8] - offset.seconds)
en_instruction_at = Time.zone.at(dossiers_sheet.data[0][9] - offset.seconds)
expect(dossiers_sheet.data.first.size).to eq(nominal_headers.size)
expect(depose_at).to eq(dossier.depose_at.round)
2019-06-25 15:46:10 +02:00
expect(en_instruction_at).to eq(dossier.en_instruction_at.round)
end
context 'with a birthdate' do
before { procedure.update(ask_birthday: true) }
let(:birthdate_headers) { nominal_headers.insert(nominal_headers.index('Archivé'), 'Date de naissance') }
it { expect(dossiers_sheet.headers).to match_array(birthdate_headers) }
it { expect(dossiers_sheet.data[0][dossiers_sheet.headers.index('Date de naissance')]).to be_a(Date) }
end
context 'with a procedure routee' do
before { create(:groupe_instructeur, label: '2', procedure: procedure) }
let(:routee_headers) { nominal_headers.insert(nominal_headers.index('textarea'), 'Groupe instructeur') }
it { expect(dossiers_sheet.headers).to match_array(routee_headers) }
2019-06-25 15:46:10 +02:00
it { expect(dossiers_sheet.data[0][dossiers_sheet.headers.index('Groupe instructeur')]).to eq('défaut') }
end
context 'with a dossier having multiple pjs' do
let!(:dossier_2) { create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure) }
before do
dossier_2.champs_public
.find { _1.is_a? Champs::PieceJustificativeChamp }
.piece_justificative_file
.attach(io: StringIO.new("toto"), filename: "toto.txt", content_type: "text/plain")
end
it { expect(dossiers_sheet.data.first.size).to eq(nominal_headers.size) }
end
2019-06-25 15:46:10 +02:00
end
describe 'Etablissement sheet' do
2019-06-25 15:46:10 +02:00
let(:procedure) { create(:procedure, :published, :with_all_champs) }
let!(:dossier) { create(:dossier, :en_instruction, :with_populated_champs, :with_entreprise, procedure: procedure) }
2019-06-25 15:46:10 +02:00
let(:dossier_etablissement) { etablissements_sheet.data[1] }
let(:champ_etablissement) { etablissements_sheet.data[0] }
let(:nominal_headers) do
[
"ID",
"Email",
"Entreprise raison sociale",
"Archivé",
"État du dossier",
"Dernière mise à jour le",
"Déposé le",
"Passé en instruction le",
"Traité le",
"Motivation de la décision",
"Instructeurs",
"textarea",
"date",
"datetime",
"number",
"decimal_number",
"integer_number",
"checkbox",
"civilite",
"email",
"phone",
"address",
"yes_no",
"simple_drop_down_list",
"multiple_drop_down_list",
"linked_drop_down_list",
"communes",
"communes (Code insee)",
"communes (Département)",
2022-09-26 10:55:37 +02:00
"departements",
"departements (Code)",
2022-09-26 10:55:37 +02:00
"regions",
"regions (Code)",
2022-09-26 10:55:37 +02:00
"pays",
"pays (Code)",
2019-06-25 15:46:10 +02:00
"dossier_link",
"piece_justificative",
2022-09-21 15:56:01 +02:00
"rna",
2019-06-25 15:46:10 +02:00
"carte",
2020-09-17 17:09:16 +02:00
"titre_identite",
2020-09-22 16:04:57 +02:00
"iban",
2022-09-26 10:55:37 +02:00
"siret",
2021-01-14 17:29:57 +01:00
"annuaire_education",
2021-09-24 14:21:13 +02:00
"cnaf",
2021-11-24 17:30:35 +01:00
"dgfip",
"pole_emploi",
2021-12-15 15:25:02 +01:00
"mesri",
2023-01-19 09:43:19 +01:00
"text",
"epci",
"epci (Code)",
2023-05-30 14:42:36 +02:00
"epci (Département)",
"cojo"
2019-06-25 15:46:10 +02:00
]
end
2019-06-25 15:46:10 +02:00
context 'as csv' do
subject do
ProcedureExportService.new(procedure, procedure.dossiers)
.to_csv
.open { |f| CSV.read(f.path) }
2019-06-25 15:46:10 +02:00
end
2019-06-25 15:46:10 +02:00
let(:nominal_headers) do
[
2019-06-25 15:46:10 +02:00
"ID",
"Email",
"Établissement SIRET",
"Établissement siège social",
"Établissement NAF",
"Établissement libellé NAF",
"Établissement Adresse",
"Établissement numero voie",
"Établissement type voie",
"Établissement nom voie",
"Établissement complément adresse",
"Établissement code postal",
"Établissement localité",
"Établissement code INSEE localité",
"Entreprise SIREN",
"Entreprise capital social",
"Entreprise numero TVA intracommunautaire",
"Entreprise forme juridique",
"Entreprise forme juridique code",
"Entreprise nom commercial",
"Entreprise raison sociale",
"Entreprise SIRET siège social",
"Entreprise code effectif entreprise",
"Entreprise date de création",
"Entreprise état administratif",
2019-06-25 15:46:10 +02:00
"Entreprise nom",
"Entreprise prénom",
"Association RNA",
"Association titre",
"Association objet",
"Association date de création",
"Association date de déclaration",
"Association date de publication",
"Archivé",
"État du dossier",
"Dernière mise à jour le",
"Déposé le",
"Passé en instruction le",
"Traité le",
"Motivation de la décision",
"Instructeurs",
"textarea",
"date",
"datetime",
"number",
"decimal_number",
"integer_number",
"checkbox",
"civilite",
"email",
"phone",
"address",
"yes_no",
"simple_drop_down_list",
"multiple_drop_down_list",
"linked_drop_down_list",
"communes",
"communes (Code insee)",
"communes (Département)",
2022-09-26 10:55:37 +02:00
"departements",
"departements (Code)",
2022-09-26 10:55:37 +02:00
"regions",
"regions (Code)",
2022-09-26 10:55:37 +02:00
"pays",
"pays (Code)",
2019-06-25 15:46:10 +02:00
"dossier_link",
"piece_justificative",
2022-09-21 15:56:01 +02:00
"rna",
2019-06-25 15:46:10 +02:00
"carte",
2020-09-17 17:09:16 +02:00
"titre_identite",
2020-09-22 16:04:57 +02:00
"iban",
2022-09-26 10:55:37 +02:00
"siret",
2021-01-14 17:29:57 +01:00
"annuaire_education",
2021-09-24 14:21:13 +02:00
"cnaf",
2021-11-24 17:30:35 +01:00
"dgfip",
"pole_emploi",
2021-12-15 15:25:02 +01:00
"mesri",
2023-01-19 09:43:19 +01:00
"text",
"epci",
"epci (Code)",
2023-05-30 14:42:36 +02:00
"epci (Département)",
"cojo"
]
2019-06-25 15:46:10 +02:00
end
2019-06-25 15:46:10 +02:00
let(:dossiers_sheet_headers) { subject.first }
2019-06-25 15:46:10 +02:00
it 'should have headers' do
expect(dossiers_sheet_headers).to match_array(nominal_headers)
2019-06-25 15:46:10 +02:00
end
end
2019-06-25 15:46:10 +02:00
it 'should have headers' do
expect(dossiers_sheet.headers).to match_array(nominal_headers)
2019-06-25 15:46:10 +02:00
expect(etablissements_sheet.headers).to eq([
"Dossier ID",
"Champ",
"Établissement SIRET",
"Etablissement enseigne",
2019-06-25 15:46:10 +02:00
"Établissement siège social",
"Établissement NAF",
"Établissement libellé NAF",
"Établissement Adresse",
"Établissement numero voie",
"Établissement type voie",
"Établissement nom voie",
"Établissement complément adresse",
"Établissement code postal",
"Établissement localité",
"Établissement code INSEE localité",
"Entreprise SIREN",
"Entreprise capital social",
"Entreprise numero TVA intracommunautaire",
"Entreprise forme juridique",
"Entreprise forme juridique code",
"Entreprise nom commercial",
"Entreprise raison sociale",
"Entreprise SIRET siège social",
"Entreprise code effectif entreprise",
"Entreprise date de création",
"Entreprise état administratif",
2019-06-25 15:46:10 +02:00
"Entreprise nom",
"Entreprise prénom",
"Association RNA",
"Association titre",
"Association objet",
"Association date de création",
"Association date de déclaration",
"Association date de publication"
])
end
2019-06-25 15:46:10 +02:00
it 'should have data' do
expect(etablissements_sheet.data.size).to eq(2)
expect(dossier_etablissement[1]).to eq("Dossier")
expect(champ_etablissement[1]).to eq("siret")
end
end
describe 'Avis sheet' do
let!(:dossier) { create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure) }
2019-06-25 15:46:10 +02:00
let!(:avis) { create(:avis, :with_answer, dossier: dossier) }
2019-06-25 15:46:10 +02:00
it 'should have headers' do
expect(avis_sheet.headers).to eq([
"Dossier ID",
2023-03-10 14:38:37 +01:00
"Introduction",
2019-06-25 15:46:10 +02:00
"Réponse",
2023-03-10 14:38:37 +01:00
"Question",
"Réponse oui/non",
2019-06-25 15:46:10 +02:00
"Créé le",
"Répondu le",
"Instructeur",
"Expert"
2019-06-25 15:46:10 +02:00
])
end
2019-06-25 15:46:10 +02:00
it 'should have data' do
expect(avis_sheet.data.size).to eq(1)
end
end
describe 'Repetitions sheet' do
let(:procedure) { create(:procedure, :published, :for_individual, types_de_champ_public: [{ type: :repetition, children: [{ libelle: 'Nom' }, { libelle: 'Age' }] }]) }
2019-06-25 15:46:10 +02:00
let!(:dossiers) do
[
create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure),
create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure)
2019-06-25 15:46:10 +02:00
]
end
let(:champ_repetition) { dossiers.first.champs_public.find { |champ| champ.type_champ == 'repetition' } }
2019-06-25 15:46:10 +02:00
it 'should have sheets' do
expect(subject.sheets.map(&:name)).to eq(['Dossiers', 'Etablissements', 'Avis', champ_repetition.libelle_for_export])
end
context 'with cloned procedure' do
let(:other_parent) { create(:type_de_champ_repetition, stable_id: champ_repetition.stable_id) }
before do
create(:type_de_champ, parent: create(:procedure_revision_type_de_champ, type_de_champ: other_parent, revision: create(:procedure).active_revision, position: 0))
end
it 'should have headers' do
expect(repetition_sheet.headers).to eq([
"Dossier ID",
"Ligne",
"Nom",
"Age"
])
end
end
2019-06-25 15:46:10 +02:00
it 'should have data' do
expect(repetition_sheet.data.size).to eq(4)
end
2019-06-25 15:46:10 +02:00
context 'with invalid characters' do
before do
champ_repetition.type_de_champ.update(libelle: 'A / B \ C *[]?')
2019-06-25 15:46:10 +02:00
end
it 'should have valid sheet name' do
expect(subject.sheets.map(&:name)).to eq(['Dossiers', 'Etablissements', 'Avis', "(#{champ_repetition.type_de_champ.stable_id}) A - B - C"])
end
end
context 'with long libelle composed of utf8 characteres' do
before do
procedure.active_revision.types_de_champ_public.each do |c|
c.update!(libelle: "#{c.id} - ?/[] ééé ééé ééééééé ééééééé éééééééé. ééé éé éééééééé éé ééé. ééééé éééééééé ééé ééé.")
end
champ_repetition.champs.each do |c|
c.type_de_champ.update!(libelle: "#{c.id} - Quam rem nam maiores numquam dolorem nesciunt. Cum et possimus et aut. Fugit voluptas qui qui.")
end
end
it 'should have valid sheet name' do
expect { subject }.not_to raise_error
end
end
2019-06-25 15:46:10 +02:00
context 'with non unique labels' do
let(:dossier) { create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure) }
let(:champ_repetition) { dossier.champs_public.find { |champ| champ.type_champ == 'repetition' } }
let(:type_de_champ_repetition) { create(:type_de_champ_repetition, :with_types_de_champ, procedure: procedure, libelle: champ_repetition.libelle) }
2019-06-25 15:46:10 +02:00
let!(:another_champ_repetition) { create(:champ_repetition, type_de_champ: type_de_champ_repetition, dossier: dossier) }
2019-06-25 15:46:10 +02:00
it 'should have sheets' do
2022-05-13 14:59:11 +02:00
expect(subject.sheets.map(&:name)).to eq(['Dossiers', 'Etablissements', 'Avis', another_champ_repetition.libelle_for_export, champ_repetition.libelle_for_export])
end
end
context 'with empty repetition' do
before do
dossiers.flat_map { |dossier| dossier.champs_public.filter(&:repetition?) }.each do |champ|
champ.champs.destroy_all
end
end
it 'should not have data' do
expect(repetition_sheet).to be_nil
end
end
end
end
describe 'to_zip' do
subject { service.to_zip }
context 'without files' do
it 'does not raises in_batches' do
expect { subject }.not_to raise_error(NoMethodError)
end
it 'returns an empty blob' do
expect(subject).to be_an_instance_of(ActiveStorage::Blob)
end
end
context 'generate_dossier_export' do
it 'include_infos_administration (so it includes avis, champs privés)' do
expect(ActiveStorage::DownloadableFile).to receive(:create_list_from_dossiers).with(anything, with_champs_private: true, include_infos_administration: true).and_return([])
subject
end
end
context 'with files (and http calls)' do
let!(:dossier) { create(:dossier, :accepte, :with_populated_champs, :with_individual, procedure: procedure) }
2022-04-07 15:23:18 +02:00
let(:dossier_exports) { PiecesJustificativesService.generate_dossier_export(Dossier.where(id: dossier)) }
before do
allow_any_instance_of(ActiveStorage::Attachment).to receive(:url).and_return("https://opengraph.githubassets.com/d0e7862b24d8026a3c03516d865b28151eb3859029c6c6c2e86605891fbdcd7a/socketry/async-io")
end
it 'returns a blob with valid files' do
VCR.use_cassette('archive/new_file_to_get_200') do
subject
File.write('tmp.zip', subject.download, mode: 'wb')
File.open('tmp.zip') do |fd|
files = ZipTricks::FileReader.read_zip_structure(io: fd)
structure = [
"#{service.send(:base_filename)}/",
"#{service.send(:base_filename)}/dossier-#{dossier.id}/",
"#{service.send(:base_filename)}/dossier-#{dossier.id}/pieces_justificatives/",
"#{service.send(:base_filename)}/dossier-#{dossier.id}/#{ActiveStorage::DownloadableFile.timestamped_filename(ActiveStorage::Attachment.where(record_type: "Champ").first)}",
2022-04-07 15:23:18 +02:00
"#{service.send(:base_filename)}/dossier-#{dossier.id}/#{ActiveStorage::DownloadableFile.timestamped_filename(dossier_exports.first.first)}"
]
expect(files.size).to eq(structure.size)
expect(files.map(&:filename)).to match_array(structure)
end
FileUtils.remove_entry_secure('tmp.zip')
end
end
end
end
2022-11-16 11:50:19 +01:00
describe 'to_geo_json' do
subject do
service
.to_geo_json
.open { |f| JSON.parse(f.read) }
end
let(:dossier) { create(:dossier, :en_instruction, :with_populated_champs, :with_individual, procedure: procedure) }
let(:champ_carte) { dossier.champs_public.find(&:carte?) }
let(:properties) { subject['features'].first['properties'] }
before do
create(:geo_area, :polygon, champ: champ_carte)
end
it 'should have features' do
2022-11-16 11:50:19 +01:00
expect(subject['features'].size).to eq(1)
expect(properties['dossier_id']).to eq(dossier.id)
expect(properties['champ_id']).to eq(champ_carte.stable_id)
expect(properties['champ_label']).to eq(champ_carte.libelle)
end
end
end