scinote-web/app/services/repository_zip_export.rb

107 lines
4.1 KiB
Ruby
Raw Normal View History

# frozen_string_literal: true
require 'csv'
module RepositoryZipExport
def self.generate_zip(params, repository, current_user)
# Fetch rows in the same order as in the currently viewed datatable
2020-07-17 21:17:09 +08:00
if params[:my_module_id]
rows = if repository.is_a?(RepositorySnapshot)
repository.repository_rows
else
repository.repository_rows
.joins(:my_module_repository_rows)
.where(my_module_repository_rows: { my_module_id: params[:my_module_id] })
end
if repository.has_stock_management?
rows = rows.left_joins(my_module_repository_rows: :repository_stock_unit_item)
.select(
'repository_rows.*',
'my_module_repository_rows.stock_consumption'
)
end
2020-07-17 21:17:09 +08:00
else
ordered_row_ids = params[:row_ids]
id_row_map = RepositoryRow.where(id: ordered_row_ids,
repository: repository)
.index_by(&:id)
rows = ordered_row_ids.collect { |id| id_row_map[id.to_i] }
end
zip = ZipExport.create(user: current_user)
zip.generate_exportable_zip(
current_user,
to_csv(rows, params[:header_ids], current_user, repository, nil, params[:my_module_id].present?),
:repositories
)
end
def self.to_csv(rows, column_ids, user, repository, handle_file_name_func = nil, in_module = false)
# Parse column names
csv_header = []
add_consumption = in_module && !repository.is_a?(RepositorySnapshot) && repository.has_stock_management?
column_ids.each do |c_id|
csv_header << case c_id.to_i
when -1, -2
next
when -3
I18n.t('repositories.table.id')
when -4
I18n.t('repositories.table.row_name')
when -5
I18n.t('repositories.table.added_by')
when -6
I18n.t('repositories.table.added_on')
when -7
I18n.t('repositories.table.archived_by')
when -8
I18n.t('repositories.table.archived_on')
else
column = repository.repository_columns.find_by(id: c_id)
column ? column.name : nil
end
end
csv_header << I18n.t('repositories.table.row_consumption') if add_consumption
CSV.generate do |csv|
csv << csv_header
rows.each do |row|
csv_row = []
column_ids.each do |c_id|
csv_row << case c_id.to_i
when -1, -2
next
when -3
repository.is_a?(RepositorySnapshot) ? row.parent_id : row.id
when -4
row.name
when -5
row.created_by.full_name
when -6
I18n.l(row.created_at, format: :full)
when -7
row.archived? && row.archived_by.present? ? row.archived_by.full_name : ''
when -8
row.archived? && row.archived_on.present? ? I18n.l(row.archived_on, format: :full) : ''
else
cell = row.repository_cells.find_by(repository_column_id: c_id)
if cell
if cell.value_type == 'RepositoryAssetValue' && handle_file_name_func
handle_file_name_func.call(cell.value.asset)
else
2018-09-23 17:14:59 +08:00
SmartAnnotations::TagToText.new(
user, repository.team, cell.value.export_formatted
2018-09-23 17:14:59 +08:00
).text
end
end
end
end
csv_row << row.row_consumption(row.stock_consumption) if add_consumption
csv << csv_row
end
end
end
end