2020-10-13 19:54:16 +08:00
|
|
|
|
# frozen_string_literal: true
|
|
|
|
|
|
|
|
|
|
module Reports
|
|
|
|
|
class HtmlToWordConverter
|
2020-10-14 22:09:49 +08:00
|
|
|
|
def initialize(document, options = {})
|
2020-10-13 19:54:16 +08:00
|
|
|
|
@docx = document
|
2020-10-14 22:09:49 +08:00
|
|
|
|
@scinote_url = options[:scinote_url]
|
|
|
|
|
@link_style = options[:link_style]
|
2020-10-13 19:54:16 +08:00
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def html_to_word_converter(text)
|
|
|
|
|
html = Nokogiri::HTML(text)
|
2020-10-14 22:09:49 +08:00
|
|
|
|
raw_elements = recursive_children(html.css('body').children, []).compact
|
2020-10-13 19:54:16 +08:00
|
|
|
|
|
|
|
|
|
# Combined raw text blocks in paragraphs
|
|
|
|
|
elements = combine_docx_elements(raw_elements)
|
|
|
|
|
|
|
|
|
|
# Draw elements
|
|
|
|
|
elements.each do |elem|
|
|
|
|
|
if elem[:type] == 'p'
|
|
|
|
|
Reports::DocxRenderer.render_p_element(@docx, elem, scinote_url: @scinote_url, link_style: @link_style)
|
|
|
|
|
elsif elem[:type] == 'table'
|
|
|
|
|
Reports::DocxRenderer.render_table_element(@docx, elem)
|
|
|
|
|
elsif elem[:type] == 'newline'
|
|
|
|
|
style = elem[:style] || {}
|
|
|
|
|
# print heading if its heading
|
|
|
|
|
# Mixing heading with other style setting causes problems for Word
|
2020-11-27 20:33:44 +08:00
|
|
|
|
if %w(h1 h2 h3 h4 h5 h6).include?(style[:style])
|
2020-10-13 19:54:16 +08:00
|
|
|
|
@docx.public_send(style[:style], elem[:value])
|
|
|
|
|
else
|
|
|
|
|
@docx.p elem[:value] do
|
|
|
|
|
align style[:align]
|
|
|
|
|
color style[:color]
|
|
|
|
|
bold style[:bold]
|
|
|
|
|
italic style[:italic]
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
elsif elem[:type] == 'image'
|
|
|
|
|
Reports::DocxRenderer.render_img_element(@docx, elem)
|
|
|
|
|
elsif %w(ul ol).include?(elem[:type])
|
|
|
|
|
Reports::DocxRenderer.render_list_element(@docx, elem)
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
private
|
|
|
|
|
|
|
|
|
|
def combine_docx_elements(raw_elements)
|
|
|
|
|
# Word does not support some nested elements, move some elements to root level
|
|
|
|
|
elements = []
|
|
|
|
|
temp_p = []
|
|
|
|
|
raw_elements.each do |elem|
|
2021-06-14 18:41:34 +08:00
|
|
|
|
next unless elem
|
|
|
|
|
|
2020-10-13 19:54:16 +08:00
|
|
|
|
if %w(image newline table ol ul).include? elem[:type]
|
2021-07-23 17:56:28 +08:00
|
|
|
|
unless temp_p.blank?
|
2020-10-13 19:54:16 +08:00
|
|
|
|
elements.push(type: 'p', children: temp_p)
|
|
|
|
|
temp_p = []
|
|
|
|
|
end
|
|
|
|
|
elements.push(elem)
|
|
|
|
|
elsif %w(br text a).include? elem[:type]
|
|
|
|
|
temp_p.push(elem)
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
elements.push(type: 'p', children: temp_p)
|
|
|
|
|
elements
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# Convert HTML structure to plain text structure
|
|
|
|
|
# rubocop:disable Metrics/BlockLength
|
|
|
|
|
def recursive_children(children, elements)
|
|
|
|
|
children.each do |elem|
|
|
|
|
|
if elem.class == Nokogiri::XML::Text
|
|
|
|
|
next if elem.text.strip == ' ' # Invisible symbol
|
|
|
|
|
|
|
|
|
|
style = paragraph_styling(elem.parent)
|
|
|
|
|
type = (style[:align] && style[:align] != :justify) || style[:style] ? 'newline' : 'text'
|
|
|
|
|
|
|
|
|
|
text = smart_annotation_check(elem)
|
|
|
|
|
|
|
|
|
|
elements.push(
|
|
|
|
|
type: type,
|
|
|
|
|
value: text.strip.delete(' '), # Invisible symbol
|
|
|
|
|
style: style
|
|
|
|
|
)
|
|
|
|
|
next
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
if elem.name == 'br'
|
|
|
|
|
elements.push(type: 'br')
|
|
|
|
|
next
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
if elem.name == 'img'
|
|
|
|
|
elements.push(img_element(elem))
|
|
|
|
|
next
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
if elem.name == 'a'
|
|
|
|
|
elements.push(link_element(elem))
|
|
|
|
|
next
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
if elem.name == 'table'
|
|
|
|
|
elements.push(tiny_mce_table_element(elem))
|
|
|
|
|
next
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
if %w(ul ol).include?(elem.name)
|
|
|
|
|
elements.push(list_element(elem))
|
|
|
|
|
next
|
|
|
|
|
end
|
|
|
|
|
elements = recursive_children(elem.children, elements) if elem.children
|
|
|
|
|
end
|
|
|
|
|
elements
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# rubocop:enable Metrics/BlockLength
|
|
|
|
|
|
|
|
|
|
def img_element(elem)
|
|
|
|
|
return unless elem.attributes['data-mce-token']
|
|
|
|
|
|
|
|
|
|
image = TinyMceAsset.find_by(id: Base62.decode(elem.attributes['data-mce-token'].value))
|
|
|
|
|
return unless image
|
|
|
|
|
|
2020-10-28 23:32:22 +08:00
|
|
|
|
image_path = Reports::Utils.image_prepare(image).service_url
|
2020-10-13 19:54:16 +08:00
|
|
|
|
dimension = FastImage.size(image_path)
|
|
|
|
|
|
|
|
|
|
return unless dimension
|
|
|
|
|
|
|
|
|
|
style = image_styling(elem, dimension)
|
|
|
|
|
|
|
|
|
|
{ type: 'image', data: image_path.split('&')[0], blob: image.blob, style: style }
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def link_element(elem)
|
|
|
|
|
text = elem.text
|
|
|
|
|
link = elem.attributes['href'].value if elem.attributes['href']
|
|
|
|
|
if elem.attributes['class']&.value == 'record-info-link'
|
|
|
|
|
link = nil
|
|
|
|
|
text = "##{text}"
|
|
|
|
|
end
|
|
|
|
|
text = "##{text}" if elem.parent.attributes['class']&.value == 'atwho-inserted'
|
|
|
|
|
text = "@#{text}" if elem.attributes['class']&.value == 'atwho-user-popover'
|
|
|
|
|
{
|
|
|
|
|
type: 'a',
|
|
|
|
|
value: text,
|
|
|
|
|
link: link
|
|
|
|
|
}
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def list_element(list_element)
|
2021-02-02 18:26:16 +08:00
|
|
|
|
allowed_elements = %w(li ul ol a img strong em h1 h2 h2 h3 h4 h5 span p)
|
2020-10-21 14:27:10 +08:00
|
|
|
|
data_array = list_element.children.select { |n| allowed_elements.include?(n.name) }.map do |li_child|
|
2020-10-13 19:54:16 +08:00
|
|
|
|
li_child.children.map do |item|
|
|
|
|
|
if item.is_a? Nokogiri::XML::Text
|
|
|
|
|
item.text.chomp
|
|
|
|
|
elsif %w(ul ol).include?(item.name)
|
|
|
|
|
list_element(item)
|
|
|
|
|
elsif %w(a).include?(item.name)
|
|
|
|
|
link_element(item)
|
|
|
|
|
elsif %w(img).include?(item.name)
|
|
|
|
|
img_element(item)&.merge(bookmark_id: SecureRandom.hex)
|
|
|
|
|
elsif %w(table).include?(item.name)
|
|
|
|
|
tiny_mce_table_element(item).merge(bookmark_id: SecureRandom.hex)
|
2021-02-02 18:26:16 +08:00
|
|
|
|
elsif %w(strong em h1 h2 h2 h3 h4 h5 span p).include?(item.name)
|
2020-10-21 14:27:10 +08:00
|
|
|
|
# Pass styles and extend renderer for li with style, some limitations on li items
|
|
|
|
|
# { type: 'text', value: item[:value], style: paragraph_styling(item) }
|
|
|
|
|
item.children.text
|
2020-10-13 19:54:16 +08:00
|
|
|
|
end
|
|
|
|
|
end.reject(&:blank?)
|
|
|
|
|
end
|
|
|
|
|
{ type: list_element.name, data: data_array }
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def smart_annotation_check(elem)
|
|
|
|
|
return "[#{elem.text}]" if elem.parent.attributes['class']&.value == 'sa-type'
|
|
|
|
|
|
|
|
|
|
elem.text
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# Prepare style for text
|
|
|
|
|
def paragraph_styling(elem)
|
|
|
|
|
style = elem.attributes['style']
|
|
|
|
|
result = {}
|
|
|
|
|
result[:style] = elem.name if elem.name.include? 'h'
|
|
|
|
|
result[:bold] = true if elem.name == 'strong'
|
|
|
|
|
result[:italic] = true if elem.name == 'em'
|
|
|
|
|
style_keys = %w(text-align color)
|
|
|
|
|
|
|
|
|
|
if style
|
|
|
|
|
style_keys.each do |key|
|
|
|
|
|
style_el = style.value.split(';').select { |i| (i.include? key) }[0]
|
|
|
|
|
next unless style_el
|
|
|
|
|
|
|
|
|
|
value = style_el.split(':')[1].strip if style_el
|
|
|
|
|
if key == 'text-align'
|
|
|
|
|
result[:align] = value.to_sym
|
|
|
|
|
elsif key == 'color' && Reports::Utils.calculate_color_hsp(value) < 190
|
|
|
|
|
result[:color] = value.delete('#')
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
result
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
# Prepare style for images
|
|
|
|
|
def image_styling(elem, dimension)
|
|
|
|
|
dimension[0] = elem.attributes['width'].value.to_i if elem.attributes['width']
|
|
|
|
|
dimension[1] = elem.attributes['height'].value.to_i if elem.attributes['height']
|
|
|
|
|
|
|
|
|
|
if elem.attributes['style']
|
|
|
|
|
align = if elem.attributes['style'].value.include? 'margin-right'
|
|
|
|
|
:center
|
|
|
|
|
elsif elem.attributes['style'].value.include? 'float: right'
|
|
|
|
|
:right
|
|
|
|
|
else
|
|
|
|
|
:left
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
margins = Constants::REPORT_DOCX_MARGIN_LEFT + Constants::REPORT_DOCX_MARGIN_RIGHT
|
|
|
|
|
max_width = (Constants::REPORT_DOCX_WIDTH - margins) / 20
|
|
|
|
|
|
|
|
|
|
if dimension[0] > max_width
|
|
|
|
|
x = max_width
|
|
|
|
|
y = dimension[1] * max_width / dimension[0]
|
|
|
|
|
else
|
|
|
|
|
x = dimension[0]
|
|
|
|
|
y = dimension[1]
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
{
|
|
|
|
|
width: x,
|
|
|
|
|
height: y,
|
|
|
|
|
align: align,
|
|
|
|
|
max_width: max_width
|
|
|
|
|
}
|
|
|
|
|
end
|
|
|
|
|
|
|
|
|
|
def tiny_mce_table_element(table_element)
|
|
|
|
|
# array of elements
|
|
|
|
|
rows = table_element.css('tbody').first.children.map do |row|
|
|
|
|
|
next unless row.name == 'tr'
|
|
|
|
|
|
|
|
|
|
cells = row.children.map do |cell|
|
|
|
|
|
next unless cell.name == 'td'
|
|
|
|
|
|
|
|
|
|
# Parse cell content
|
|
|
|
|
formated_cell = recursive_children(cell.children, [])
|
|
|
|
|
|
|
|
|
|
# Combine text elements to single paragraph
|
|
|
|
|
formated_cell = combine_docx_elements(formated_cell)
|
|
|
|
|
formated_cell
|
|
|
|
|
end.reject(&:blank?)
|
|
|
|
|
{ type: 'tr', data: cells }
|
|
|
|
|
end.reject(&:blank?)
|
|
|
|
|
{ type: 'table', data: rows }
|
|
|
|
|
end
|
|
|
|
|
end
|
|
|
|
|
end
|