76 lines
2.2 KiB
Python
76 lines
2.2 KiB
Python
"""Preprocessor for merging consecutive stream outputs for easier handling."""
|
|
|
|
# Copyright (c) IPython Development Team.
|
|
# Distributed under the terms of the Modified BSD License.
|
|
|
|
import re
|
|
import functools
|
|
from traitlets.log import get_logger
|
|
|
|
def cell_preprocessor(function):
|
|
"""
|
|
Wrap a function to be executed on all cells of a notebook
|
|
|
|
The wrapped function should have these parameters:
|
|
|
|
cell : NotebookNode cell
|
|
Notebook cell being processed
|
|
resources : dictionary
|
|
Additional resources used in the conversion process. Allows
|
|
preprocessors to pass variables into the Jinja engine.
|
|
index : int
|
|
Index of the cell being processed
|
|
"""
|
|
@functools.wraps(function)
|
|
def wrappedfunc(nb, resources):
|
|
get_logger().debug(
|
|
"Applying preprocessor: %s", function.__name__
|
|
)
|
|
for index, cell in enumerate(nb.cells):
|
|
nb.cells[index], resources = function(cell, resources, index)
|
|
return nb, resources
|
|
return wrappedfunc
|
|
|
|
cr_pat = re.compile(r'.*\r(?=[^\n])')
|
|
|
|
@cell_preprocessor
|
|
def coalesce_streams(cell, resources, index):
|
|
"""
|
|
Merge consecutive sequences of stream output into single stream
|
|
to prevent extra newlines inserted at flush calls
|
|
|
|
Parameters
|
|
----------
|
|
cell : NotebookNode cell
|
|
Notebook cell being processed
|
|
resources : dictionary
|
|
Additional resources used in the conversion process. Allows
|
|
transformers to pass variables into the Jinja engine.
|
|
index : int
|
|
Index of the cell being processed
|
|
"""
|
|
|
|
outputs = cell.get('outputs', [])
|
|
if not outputs:
|
|
return cell, resources
|
|
|
|
last = outputs[0]
|
|
new_outputs = [last]
|
|
for output in outputs[1:]:
|
|
if (output.output_type == 'stream' and
|
|
last.output_type == 'stream' and
|
|
last.name == output.name
|
|
):
|
|
last.text += output.text
|
|
|
|
else:
|
|
new_outputs.append(output)
|
|
last = output
|
|
|
|
# process \r characters
|
|
for output in new_outputs:
|
|
if output.output_type == 'stream' and '\r' in output.text:
|
|
output.text = cr_pat.sub('', output.text)
|
|
|
|
cell.outputs = new_outputs
|
|
return cell, resources
|