debian-mirror-gitlab/vendor/gems/ipynbdiff/lib/transformer.rb
2023-01-13 15:02:22 +05:30

112 lines
3.3 KiB
Ruby

# frozen_string_literal: true
module IpynbDiff
require 'oj'
class InvalidNotebookError < StandardError
end
# Returns a markdown version of the Jupyter Notebook
class Transformer
require 'json'
require 'yaml'
require 'output_transformer'
require 'symbolized_markdown_helper'
require 'symbol_map'
require 'transformed_notebook'
include SymbolizedMarkdownHelper
@include_frontmatter = true
def initialize(include_frontmatter: true, hide_images: false)
@include_frontmatter = include_frontmatter
@hide_images = hide_images
@out_transformer = OutputTransformer.new(hide_images)
end
def validate_notebook(notebook)
notebook_json = Oj::Parser.usual.parse(notebook)
return notebook_json if notebook_json&.key?('cells')
raise InvalidNotebookError
rescue EncodingError, Oj::ParseError, JSON::ParserError
raise InvalidNotebookError
end
def transform(notebook)
return TransformedNotebook.new unless notebook
notebook_json = validate_notebook(notebook)
transformed = transform_document(notebook_json)
symbol_map = SymbolMap.parse(notebook)
TransformedNotebook.new(transformed, symbol_map)
end
def transform_document(notebook)
symbol = JsonSymbol.new('.cells')
transformed_blocks = notebook['cells'].map.with_index do |cell, idx|
decorate_cell(transform_cell(cell, notebook, symbol / idx), cell, symbol / idx)
end
transformed_blocks.prepend(transform_metadata(notebook)) if @include_frontmatter
transformed_blocks.flatten
end
def decorate_cell(rows, cell, symbol)
tags = cell['metadata']&.fetch('tags', [])
type = cell['cell_type'] || 'raw'
[
_(symbol, %(%% Cell type:#{type} id:#{cell['id']} tags:#{tags&.join(',')})),
_,
rows,
_
]
end
def transform_cell(cell, notebook, symbol)
cell['cell_type'] == 'code' ? transform_code_cell(cell, notebook, symbol) : transform_text_cell(cell, symbol)
end
def transform_code_cell(cell, notebook, symbol)
[
_(symbol / 'source', %(``` #{notebook.dig('metadata', 'kernelspec', 'language') || ''})),
symbolize_array(symbol / 'source', cell['source'], &:rstrip),
_(nil, '```'),
transform_outputs(cell['outputs'], symbol)
]
end
def transform_outputs(outputs, symbol)
transformed = outputs.map
.with_index { |output, i| @out_transformer.transform(output, symbol / ['outputs', i]) }
.compact
.map { |el| [_, el] }
[
transformed.empty? ? [] : [_, _(symbol / 'outputs', '%% Output')],
transformed
]
end
def transform_text_cell(cell, symbol)
symbolize_array(symbol / 'source', cell['source'], &:rstrip)
end
def transform_metadata(notebook_json)
as_yaml = {
'jupyter' => {
'kernelspec' => notebook_json['metadata']['kernelspec'],
'language_info' => notebook_json['metadata']['language_info'],
'nbformat' => notebook_json['nbformat'],
'nbformat_minor' => notebook_json['nbformat_minor']
}
}.to_yaml
as_yaml.split("\n").map { |l| _(nil, l) }.append(_(nil, '---'), _)
end
end
end