|
|
"""Base classes for the notebook conversion pipeline.
|
|
|
|
|
|
This module defines Converter, from which all objects designed to implement
|
|
|
a conversion of IPython notebooks to some other format should inherit.
|
|
|
"""
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Copyright (c) 2012, the IPython Development Team.
|
|
|
#
|
|
|
# Distributed under the terms of the Modified BSD License.
|
|
|
#
|
|
|
# The full license is in the file COPYING.txt, distributed with this software.
|
|
|
#-----------------------------------------------------------------------------
|
|
|
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Imports
|
|
|
#-----------------------------------------------------------------------------
|
|
|
|
|
|
from __future__ import print_function, absolute_import
|
|
|
|
|
|
# Stdlib imports
|
|
|
import jinja2
|
|
|
import codecs
|
|
|
import io
|
|
|
import logging
|
|
|
import os
|
|
|
import pprint
|
|
|
import re
|
|
|
from types import FunctionType
|
|
|
|
|
|
from jinja2 import Environment, PackageLoader, FileSystemLoader
|
|
|
env = Environment(loader=FileSystemLoader('./templates/'))
|
|
|
|
|
|
# IPython imports
|
|
|
from IPython.nbformat import current as nbformat
|
|
|
from IPython.config.configurable import Configurable, SingletonConfigurable
|
|
|
from IPython.utils.traitlets import (List, Unicode, Type, Bool, Dict, CaselessStrEnum,
|
|
|
Any)
|
|
|
|
|
|
# Our own imports
|
|
|
from IPython.utils.text import indent
|
|
|
from .utils import remove_ansi
|
|
|
from markdown import markdown
|
|
|
from .utils import highlight
|
|
|
#-----------------------------------------------------------------------------
|
|
|
# Class declarations
|
|
|
#-----------------------------------------------------------------------------
|
|
|
def rm_fake(strng):
|
|
|
return strng.replace('/files/', '')
|
|
|
|
|
|
class ConversionException(Exception):
|
|
|
pass
|
|
|
|
|
|
|
|
|
def python_comment(string):
|
|
|
return '# '+'\n# '.join(string.split('\n'))
|
|
|
|
|
|
env.filters['pycomment'] = python_comment
|
|
|
env.filters['indent'] = indent
|
|
|
env.filters['rm_fake'] = rm_fake
|
|
|
env.filters['rm_ansi'] = remove_ansi
|
|
|
env.filters['markdown'] = markdown
|
|
|
env.filters['highlight'] = highlight
|
|
|
|
|
|
class ConverterTemplate(Configurable):
|
|
|
|
|
|
display_data_priority = ['pdf', 'svg', 'png', 'jpg', 'text']
|
|
|
#-------------------------------------------------------------------------
|
|
|
# Instance-level attributes that are set in the constructor for this
|
|
|
# class.
|
|
|
#-------------------------------------------------------------------------
|
|
|
infile = Any()
|
|
|
|
|
|
|
|
|
infile_dir = Unicode()
|
|
|
|
|
|
def __init__(self, tplfile='fullhtml', config=None, **kw):
|
|
|
self.template = env.get_template(tplfile+'.tpl')
|
|
|
super(ConverterTemplate,self).__init__(config=config)
|
|
|
|
|
|
def _get_prompt_number(self, cell):
|
|
|
return cell.prompt_number if hasattr(cell, 'prompt_number') \
|
|
|
else self.blank_symbol
|
|
|
|
|
|
|
|
|
def process(self):
|
|
|
converted_cells = []
|
|
|
for worksheet in self.nb.worksheets:
|
|
|
for cell in worksheet.cells:
|
|
|
cell.type = cell.cell_type
|
|
|
converted_cells.append(worksheet)
|
|
|
|
|
|
return converted_cells
|
|
|
|
|
|
def convert(self, cell_separator='\n'):
|
|
|
return self.template.render(worksheets=self.process())
|
|
|
|
|
|
|
|
|
def read(self, filename):
|
|
|
"read and parse notebook into NotebookNode called self.nb"
|
|
|
with open(filename) as f:
|
|
|
self.nb = nbformat.read(f, 'json')
|
|
|
|
|
|
|