''' Provide special versions of list and dict, that can automatically notify
about changes when used for property values.
Mutations to these values are detected, and the properties owning the
collection is notified of the changes. Consider the following model
definition:
.. code-block:: python
class SomeModel(Model):
options = List(String)
If we have an instance of this model, ``m`` then we can set the entire
value of the ``options`` property at once:
.. code-block:: python
m.options = ["foo", "bar"]
When we do this in the context of a Bokeh server application that is being
viewed in a browser, this change is automatically noticed, and the
corresponding BokehJS property in the browser is synchronized, possibly
causing some change in the visual state of the application in the browser.
But it is also desirable that changes *inside* the ``options`` list also
be detected. That is, the following kinds of operations should also be
automatically synchronized between BokehJS and a Bokeh server:
.. code-block:: python
m.options.append("baz")
m.options[2] = "quux"
m.options.insert(0, "bar")
The classes in this module provide this functionality.
.. note::
These classes form part of the very low-level machinery that implements
the Bokeh model and property system. It is unlikely that any of these
classes or their methods will be applicable to any standard usage or to
anyone who is not directly developing on Bokeh's own infrastructure.
'''
from __future__ import absolute_import, print_function
from ...util.dependencies import import_optional
pd = import_optional('pandas')
[docs]def notify_owner(func):
''' A decorator for mutating methods of property container classes
that notifies owners of the property container about mutating changes.
Args:
func (callable) : the container method to wrap in a notification
Returns:
wrapped method
Examples:
A ``__setitem__`` could be wrapped like this:
.. code-block:: python
# x[i] = y
@notify_owner
def __setitem__(self, i, y):
return super(PropertyValueDict, self).__setitem__(i, y)
The returned wrapped method will have a docstring indicating what
original method it is wrapping.
'''
def wrapper(self, *args, **kwargs):
old = self._saved_copy()
result = func(self, *args, **kwargs)
self._notify_owners(old)
return result
wrapper.__doc__ = "Container method ``%s`` instrumented to notify property owners" % func.__name__
return wrapper
[docs]class PropertyValueContainer(object):
''' A base class for property container classes that support change
notifications on mutating operations.
This class maintains an internal list of property owners, and also
provides a private mechanism for methods wrapped with
:func:`~bokeh.core.property_containers.notify_owners` to update
those owners when mutating changes occur.
'''
def __init__(self, *args, **kwargs):
self._owners = set()
super(PropertyValueContainer, self).__init__(*args, **kwargs)
def _register_owner(self, owner, descriptor):
self._owners.add((owner, descriptor))
def _unregister_owner(self, owner, descriptor):
self._owners.discard((owner, descriptor))
def _notify_owners(self, old, hint=None):
for (owner, descriptor) in self._owners:
descriptor._notify_mutated(owner, old, hint=hint)
def _saved_copy(self):
raise RuntimeError("Subtypes must implement this to make a backup copy")
[docs]class PropertyValueList(PropertyValueContainer, list):
''' A list property value container that supports change notifications on
mutating operations.
When a Bokeh model has a ``List`` property, the ``PropertyValueLists`` are
transparently created to wrap those values. These ``PropertyValueList``
values are subject to normal property validation. If the property type
``foo = List(Str)`` then attempting to set ``x.foo[0] = 10`` will raise
an error.
Instances of ``PropertyValueList`` can be explicitly created by passing
any object that the standard list initializer accepts, for example:
.. code-block:: python
>>> PropertyValueList([10, 20])
[10, 20]
>>> PropertyValueList((10, 20))
[10, 20]
The following mutating operations on lists automatically trigger
notifications:
.. code-block:: python
del x[y]
del x[i:j]
x += y
x *= y
x[i] = y
x[i:j] = y
x.append
x.extend
x.insert
x.pop
x.remove
x.reverse
x.sort
'''
def __init__(self, *args, **kwargs):
return super(PropertyValueList, self).__init__(*args, **kwargs)
def _saved_copy(self):
return list(self)
# delete x[y]
@notify_owner
[docs] def __delitem__(self, y):
return super(PropertyValueList, self).__delitem__(y)
# delete x[i:j]
@notify_owner
[docs] def __delslice__(self, i, j):
# Note: this is different py2 vs py3, py3 calls __delitem__ with a
# slice index, and does not have this method at all
return super(PropertyValueList, self).__delslice__(i, j)
# x += y
@notify_owner
[docs] def __iadd__(self, y):
return super(PropertyValueList, self).__iadd__(y)
# x *= y
@notify_owner
[docs] def __imul__(self, y):
return super(PropertyValueList, self).__imul__(y)
# x[i] = y
@notify_owner
[docs] def __setitem__(self, i, y):
return super(PropertyValueList, self).__setitem__(i, y)
# x[i:j] = y
@notify_owner
[docs] def __setslice__(self, i, j, y):
# Note: this is different py2 vs py3, py3 calls __setitem__ with a
# slice index, and does not have this method at all
return super(PropertyValueList, self).__setslice__(i, j, y)
@notify_owner
[docs] def append(self, obj):
return super(PropertyValueList, self).append(obj)
@notify_owner
[docs] def extend(self, iterable):
return super(PropertyValueList, self).extend(iterable)
@notify_owner
[docs] def insert(self, index, obj):
return super(PropertyValueList, self).insert(index, obj)
@notify_owner
[docs] def pop(self, index=-1):
return super(PropertyValueList, self).pop(index)
@notify_owner
[docs] def remove(self, obj):
return super(PropertyValueList, self).remove(obj)
@notify_owner
[docs] def reverse(self):
return super(PropertyValueList, self).reverse()
@notify_owner
[docs] def sort(self, **kwargs):
return super(PropertyValueList, self).sort(**kwargs)
[docs]class PropertyValueDict(PropertyValueContainer, dict):
''' A dict property value container that supports change notifications on
mutating operations.
When a Bokeh model has a ``List`` property, the ``PropertyValueLists`` are
transparently created to wrap those values. These ``PropertyValueList``
values are subject to normal property validation. If the property type
``foo = Dict(Str, Str)`` then attempting to set ``x.foo['bar'] = 10`` will
raise an error.
Instances of ``PropertyValueDict`` can be eplicitly created by passing
any object that the standard dict initializer accepts, for example:
.. code-block:: python
>>> PropertyValueDict(dict(a=10, b=20))
{'a': 10, 'b': 20}
>>> PropertyValueDict(a=10, b=20)
{'a': 10, 'b': 20}
>>> PropertyValueDict([('a', 10), ['b', 20]])
{'a': 10, 'b': 20}
The following mutating operations on dicts automatically trigger
notifications:
.. code-block:: python
del x[y]
x[i] = y
x.clear
x.pop
x.popitem
x.setdefault
x.update
'''
def __init__(self, *args, **kwargs):
return super(PropertyValueDict, self).__init__(*args, **kwargs)
def _saved_copy(self):
return dict(self)
# delete x[y]
@notify_owner
[docs] def __delitem__(self, y):
return super(PropertyValueDict, self).__delitem__(y)
# x[i] = y
@notify_owner
[docs] def __setitem__(self, i, y):
return super(PropertyValueDict, self).__setitem__(i, y)
@notify_owner
[docs] def clear(self):
return super(PropertyValueDict, self).clear()
@notify_owner
[docs] def pop(self, *args):
return super(PropertyValueDict, self).pop(*args)
@notify_owner
[docs] def popitem(self):
return super(PropertyValueDict, self).popitem()
@notify_owner
[docs] def setdefault(self, *args):
return super(PropertyValueDict, self).setdefault(*args)
@notify_owner
[docs] def update(self, *args, **kwargs):
return super(PropertyValueDict, self).update(*args, **kwargs)
[docs]class PropertyValueColumnData(PropertyValueDict):
''' A property value container for ColumnData that supports change
notifications on mutating operations.
This property value container affords specialized code paths for
updating the .data dictionary for ColumnDataSource. When possible,
more efficient ColumnDataChangedEvent hints are generated to perform
the updates:
.. code-block:: python
x[i] = y
x.update
'''
# x[i] = y
# don't wrap with notify_owner --- notifies owners explicitly
def __setitem__(self, i, y):
return self.update([(i, y)])
# don't wrap with notify_owner --- notifies owners explicitly
def update(self, *args, **kwargs):
old = self._saved_copy()
result = super(PropertyValueDict, self).update(*args, **kwargs)
from ...document.events import ColumnDataChangedEvent
# Grab keys to update according to Python docstring for update([E, ]**F)
#
# If E is present and has a .keys() method, then does: for k in E: D[k] = E[k]
# If E is present and lacks a .keys() method, then does: for k, v in E: D[k] = v
# In either case, this is followed by: for k in F: D[k] = F[k]
cols = set(kwargs.keys())
if len(args) == 1:
E = args[0]
if hasattr(E, 'keys'):
cols |= set(E.keys())
else:
cols |= { x[0] for x in E }
# we must loop ourselves here instead of calling _notify_owners
# because the hint is customized for each owner separately
for (owner, descriptor) in self._owners:
hint = ColumnDataChangedEvent(owner.document, owner, cols=list(cols))
descriptor._notify_mutated(owner, old, hint=hint)
return result
# don't wrap with notify_owner --- notifies owners explicitly
def _stream(self, doc, source, new_data, rollover=None, setter=None):
''' Internal implementation to handle special-casing stream events
on ``ColumnDataSource`` columns.
Normally any changes to the ``.data`` dict attribute on a
``ColumnDataSource`` triggers a notification, causing all of the data
to be synchronized between server and clients.
The ``.stream`` method on column data sources exists to provide a
more efficient way to perform streaming (i.e. append-only) updates
to a data source, without having to perform a full synchronization,
which would needlessly re-send all the data.
To accomplish this, this function bypasses the wrapped methods on
``PropertyValueDict`` and uses the unwrapped versions on the dict
superclass directly. It then explicitly makes a notification, adding
a special ``ColumnsStreamedEvent`` hint to the message containing
only the small streamed data that BokehJS needs in order to
efficiently synchronize.
.. warning::
This function assumes the integrity of ``new_data`` has already
been verified.
'''
old = self._saved_copy()
import numpy as np
# pandas/issues/13918
if pd and isinstance(new_data, pd.DataFrame):
new_items = new_data.iteritems()
else:
new_items = new_data.items()
# TODO (bev) Currently this reports old differently for array vs list
# For arrays is reports the actual old value. For lists, the old value
# is actually the already updated value. This is because the method
# self._saved_copy() makes a shallow copy.
for k, v in new_items:
if isinstance(self[k], np.ndarray):
data = np.append(self[k], new_data[k])
if rollover and len(data) > rollover:
data = data[-rollover:]
super(PropertyValueDict, self).__setitem__(k, data)
else:
L = self[k]
L.extend(new_data[k])
if rollover is not None:
del L[:-rollover]
from ...document.events import ColumnsStreamedEvent
self._notify_owners(old,
hint=ColumnsStreamedEvent(doc, source, new_data, rollover, setter))
# don't wrap with notify_owner --- notifies owners explicitly
def _patch(self, doc, source, patches, setter=None):
''' Internal implementation to handle special-casing patch events
on ``ColumnDataSource`` columns.
Normally any changes to the ``.data`` dict attribute on a
``ColumnDataSource`` triggers a notification, causing all of the data
to be synchronized between server and clients.
The ``.patch`` method on column data sources exists to provide a
more efficient way to perform patching (i.e. random access) updates
to a data source, without having to perform a full synchronization,
which would needlessly re-send all the data.
To accomplish this, this function bypasses the wrapped methods on
``PropertyValueDict`` and uses the unwrapped versions on the dict
superclass directly. It then explicitly makes a notification, adding
a special ``ColumnsPatchedEvent`` hint to the message containing
only the small patched data that BokehJS needs in order to efficiently
synchronize.
.. warning::
This function assumes the integrity of ``patches`` has already
been verified.
'''
import numpy as np
old = self._saved_copy()
for name, patch in patches.items():
for ind, value in patch:
if isinstance(ind, (int, slice)):
self[name][ind] = value
else:
shape = self[name][ind[0]][ind[1:]].shape
self[name][ind[0]][ind[1:]] = np.array(value, copy=False).reshape(shape)
from ...document.events import ColumnsPatchedEvent
self._notify_owners(old,
hint=ColumnsPatchedEvent(doc, source, patches, setter))
