##// END OF EJS Templates
Merge pull request #3802 from ivanov/doc-to-rst...
Merge pull request #3802 from ivanov/doc-to-rst This renames our documentation files to .rst so that it renders nicely on GitHub The docs build just fine with these changes, it'd be good to get this in before any more changes are made to the docs

File last commit:

r5745:943a2321
r11764:79d36b57 merge
Show More
nbpy.py
150 lines | 5.1 KiB | text/x-python | PythonLexer
"""Read and write notebooks as regular .py files.
Authors:
* Brian Granger
"""
#-----------------------------------------------------------------------------
# Copyright (C) 2008-2011 The IPython Development Team
#
# Distributed under the terms of the BSD License. The full license is in
# the file COPYING, distributed as part of this software.
#-----------------------------------------------------------------------------
#-----------------------------------------------------------------------------
# Imports
#-----------------------------------------------------------------------------
import re
from .rwbase import NotebookReader, NotebookWriter
from .nbbase import new_code_cell, new_text_cell, new_worksheet, new_notebook
#-----------------------------------------------------------------------------
# Code
#-----------------------------------------------------------------------------
_encoding_declaration_re = re.compile(r"^#.*coding[:=]\s*([-\w.]+)")
class PyReaderError(Exception):
pass
class PyReader(NotebookReader):
def reads(self, s, **kwargs):
return self.to_notebook(s,**kwargs)
def to_notebook(self, s, **kwargs):
lines = s.splitlines()
cells = []
cell_lines = []
state = u'codecell'
for line in lines:
if line.startswith(u'# <nbformat>') or _encoding_declaration_re.match(line):
pass
elif line.startswith(u'# <codecell>'):
cell = self.new_cell(state, cell_lines)
if cell is not None:
cells.append(cell)
state = u'codecell'
cell_lines = []
elif line.startswith(u'# <htmlcell>'):
cell = self.new_cell(state, cell_lines)
if cell is not None:
cells.append(cell)
state = u'htmlcell'
cell_lines = []
elif line.startswith(u'# <markdowncell>'):
cell = self.new_cell(state, cell_lines)
if cell is not None:
cells.append(cell)
state = u'markdowncell'
cell_lines = []
else:
cell_lines.append(line)
if cell_lines and state == u'codecell':
cell = self.new_cell(state, cell_lines)
if cell is not None:
cells.append(cell)
ws = new_worksheet(cells=cells)
nb = new_notebook(worksheets=[ws])
return nb
def new_cell(self, state, lines):
if state == u'codecell':
input = u'\n'.join(lines)
input = input.strip(u'\n')
if input:
return new_code_cell(input=input)
elif state == u'htmlcell':
text = self._remove_comments(lines)
if text:
return new_text_cell(u'html',source=text)
elif state == u'markdowncell':
text = self._remove_comments(lines)
if text:
return new_text_cell(u'markdown',source=text)
def _remove_comments(self, lines):
new_lines = []
for line in lines:
if line.startswith(u'#'):
new_lines.append(line[2:])
else:
new_lines.append(line)
text = u'\n'.join(new_lines)
text = text.strip(u'\n')
return text
def split_lines_into_blocks(self, lines):
if len(lines) == 1:
yield lines[0]
raise StopIteration()
import ast
source = '\n'.join(lines)
code = ast.parse(source)
starts = [x.lineno-1 for x in code.body]
for i in range(len(starts)-1):
yield '\n'.join(lines[starts[i]:starts[i+1]]).strip('\n')
yield '\n'.join(lines[starts[-1]:]).strip('\n')
class PyWriter(NotebookWriter):
def writes(self, nb, **kwargs):
lines = [u'# -*- coding: utf-8 -*-']
lines.extend([u'# <nbformat>2</nbformat>',''])
for ws in nb.worksheets:
for cell in ws.cells:
if cell.cell_type == u'code':
input = cell.get(u'input')
if input is not None:
lines.extend([u'# <codecell>',u''])
lines.extend(input.splitlines())
lines.append(u'')
elif cell.cell_type == u'html':
input = cell.get(u'source')
if input is not None:
lines.extend([u'# <htmlcell>',u''])
lines.extend([u'# ' + line for line in input.splitlines()])
lines.append(u'')
elif cell.cell_type == u'markdown':
input = cell.get(u'source')
if input is not None:
lines.extend([u'# <markdowncell>',u''])
lines.extend([u'# ' + line for line in input.splitlines()])
lines.append(u'')
lines.append('')
return unicode('\n'.join(lines))
_reader = PyReader()
_writer = PyWriter()
reads = _reader.reads
read = _reader.read
to_notebook = _reader.to_notebook
write = _writer.write
writes = _writer.writes