def updatefact(self, fact, expression, **kwargs):
# AutoBlanking Feature - If there is no expression, zeros relevant fields
if expression != None and expression.strip() == u"":
for field in self.graphbasedupdater.updateablefields:
if field in fact and factproxy.isgeneratedfield(field, fact[field]):
fact[field] = u""
# TODO: Nick added this to give up after auto-blanking. He claims it removes a minor
# delay, but I'm not sure where the delay originates from, which worries me:
return
FieldUpdater.updatefact(self, fact, expression, **kwargs)
def updatefact(self, fact, expression, **kwargs):
# AutoBlanking Feature - If there is no expression, zeros relevant fields
if expression != None and expression.strip() == u"":
for field in self.graphbasedupdater.updateablefields:
if field in fact and factproxy.isgeneratedfield(field, fact[field]):
fact[field] = u""
# TODO: Nick added this to give up after auto-blanking. He claims it removes a minor
# delay, but I'm not sure where the delay originates from, which worries me:
return
FieldUpdater.updatefact(self, fact, expression, **kwargs)
def updatefact(self, fact, value, alwaysreformat=False):
assert self.field in fact
assert value is None or not(factproxy.isgeneratedfield(self.field, value))
# HACK ALERT: can't think of a nicer way to do mwfieldinfact
# NB: this is not quite right. The user might have changed whether there is a mw field independently of whether the
# contents of the field we are considering has changed. However, this bug should be very hard to hit (TODO: fix it).
delta = value is not None and { self.field : value, "mwfieldinfact" : "mw" in fact } or {}
graph = self.graphbasedupdater.filledgraph(fact, delta)
# For the same reason that fields may be missing from the graph but present in the fact,
# even the field we are updating can be missing. For example, this happens when tabbing
# away from a generated field.
#
# We should also avoid doing this if the field is considered generated, because in that
# case reformatting leads to changing the field to be spuriously marked as original.
if self.field in graph and not(factproxy.isgeneratedfield(self.field, value or fact[self.field])):
# EAGERLY reformat to produce the new value, which we plop back into the graph again so
# other computations can see it. Note that this is potentially unsafe because we might now actually
# need to recompute other thunks in the graph that depended on this one. Thankfully, right now
# we only have updaters that guarantee not to change the value of an expression which will change
# the value of anything *they* force to FIND that new value, so it works out.
fact[self.field] = self.reformatter.reformatfield(self.field, graph, alwaysreformat=alwaysreformat)
graph[self.field] = (False, utils.Thunk(lambda: fact[self.field]))
log.info("Keys in the graph after update: %r", graph.keys())
for field in fact:
# Fields can be missing from the graph if they are generated fields in the input fact that
# the graph failed to find a way to update. This really does happen!
if shouldupdatefield(self.graphbasedupdater.config)(field) and field in graph:
value = graph[field][1]()
# The value can be None if we tried to update the field but just couldn't find any information.
# This can happen e.g. with the "meaning" field if there is no definition in the dictionary.
if value is not None:
# Last-ditch attempt to convert to unicode to intercept any ASCII that slipped though the net
fact[field] = (graph[field][0] and factproxy.markgeneratedfield or (lambda x: x))(unicode(value))
def filledgraphforupdaters(all_updaters, fact, delta):
graph = {}
dirty = {}
finddirties = lambda usings: [using for using in usings if dirty[using]]
# The initial shell contains just the stuff that is non-generated or being set in this round.
# Everything else will be generated based off of these values
initiallyfilledfields = set([field for field in fact if not isgeneratedfield(field, fact[field]) and not isblankfield(fact[field])]).union(set(delta.keys()))
for field in initiallyfilledfields:
dirty[field] = field in delta
graph[field] = (False, Thunk(lambda field=field: cond(field in delta, lambda: delta[field], lambda: fact[field])))
log.info("Initially filled graph fields: %r", dirty)
# Remove useless updaters, and updaters that might confound a delta by updating a field from an
# old field. For example, if we change the reading we want to regenerate the color field -- but this is no
# good if we just use the updater that gets the reading from the expression!
all_updaters = [updater for updater in all_updaters if updater[0] not in initiallyfilledfields]
delta_updaters = maydependon(all_updaters, delta.keys())
# A complication is the presence of non-generated blank fields. We want to try and update these even if the
# delta is empty (for example). To this end we need to make sure that the cut doesn't exclude updaters that
# might potentially be depended on by an updater for a blank field.
#
# We also exclude updaters for any field that is going to get filled out by any delta_updater, because those
# updaters should take priority.
all_updaters = [updater for updater in all_updaters if all([updater[0] not in delta_field for delta_field, _, _ in delta_updaters])]
blank_updaters = dependedonby(all_updaters, [field for field in fact if isblankfield(fact[field])])
# Yes, this really works! This is because Python has reference comparison semantics on functions.
# However, this is the reason we need to use tuples of fields we depend on, rather than lists
updaters = set(delta_updaters + blank_updaters)
def shell(alreadyfilled):
# Gather all the fields we are newly able to fill given the most recent changes
# to the list of already filled things
cannowfill = FactoryDict(lambda _: [])
for updatewhat, updatefunction, updateusings in updaters:
if updatewhat not in alreadyfilled and all([updateusing in alreadyfilled for updateusing in updateusings]):
# We have to delay the computation of whether something is dirty or not until we are inside
# the actual thunk for this particular field, hence all the thunking and lambdas here
inputs = Thunk(lambda updateusings=updateusings: [graph[updateusing][1]() for updateusing in updateusings])
cannowfill[updatewhat].append((lambda inputs=inputs, updatefunction=updatefunction: updatefunction(*(inputs())),
lambda inputs=inputs, updateusings=updateusings: seq(inputs(), lambda: finddirties(updateusings))))
# Check for quiescence
if len(cannowfill) == 0:
# NB: could do something with the unfilled set (self.updateablefields.difference(alreadyfilled)) here
return
# Set up each fillable graph field with a thunk computing the value
for field, possiblefillers in cannowfill.items():
def fillme(field=field, possiblefillers=possiblefillers):
# For preference, use a filler that will certainly return clean information (i.e. sort by the number of dirty inputs and prefer the first - i.e. the one with fewer dirty inputs)
for fillerfunction, dirtyinputs, anyinputsdirty in sorted([(f, dirties(), len(dirties()) > 0) for f, dirties in possiblefillers], using(lambda x: x[2])):
if field not in fact or isblankfield(fact[field]) or anyinputsdirty:
# Don't know what the last value was or it may have changed: recompute.
#
# We also recompute if the incoming field is blank: this can happen if we have
# deleted the contents of a field, and then come back to update the fact by tabbing
# away from a filled field. Now we have enough information to fill it, and musn't
# just reuse the old version, which will be blank.
#
# Note that if the field was *generated* as blank one then it will have a marker
# in it, and so we won't pointlessly recompute its blankness.
log.info("Attempting to fill %s field -- dirty inputs are %s", field, dirtyinputs)
result = fillerfunction()
if result is None:
log.info("Filling %s failed -- falling back on another method", field)
continue
dirty[field] = cond(field in fact, lambda: result != unmarkgeneratedfield(fact[field]), lambda: anyinputsdirty)
return result
else:
# Last value must not have changed: retain it
assert (field in fact and not anyinputsdirty)
log.info("Retaining old field value for %s (blank: %s)", field, isblankfield(fact[field]))
dirty[field] = False
return unmarkgeneratedfield(fact[field])
# What if all of the possible updaters failed? Ideally we would not be in the graph at all, but it's too late for that.
# All we can do is return None, and deal with this possibility later on.
log.info("All possible updaters failed for the field %s", field)
dirty[field] = False
return None
graph[field] = (True, Thunk(fillme))
shell(alreadyfilled.union(set(cannowfill.keys())))
shell(initiallyfilledfields)
return graph
def filledgraphforupdaters(all_updaters, fact, delta):
graph = {}
dirty = {}
finddirties = lambda usings: [using for using in usings if dirty[using]]
# The initial shell contains just the stuff that is non-generated or being set in this round.
# Everything else will be generated based off of these values
initiallyfilledfields = set([field for field in fact if not isgeneratedfield(field, fact[field]) and not isblankfield(fact[field])]).union(set(delta.keys()))
for field in initiallyfilledfields:
dirty[field] = field in delta
graph[field] = (False, Thunk(lambda field=field: cond(field in delta, lambda: delta[field], lambda: fact[field])))
log.info("Initially filled graph fields: %r", dirty)
# Remove useless updaters, and updaters that might confound a delta by updating a field from an
# old field. For example, if we change the reading we want to regenerate the color field -- but this is no
# good if we just use the updater that gets the reading from the expression!
all_updaters = [updater for updater in all_updaters if updater[0] not in initiallyfilledfields]
delta_updaters = maydependon(all_updaters, delta.keys())
# A complication is the presence of non-generated blank fields. We want to try and update these even if the
# delta is empty (for example). To this end we need to make sure that the cut doesn't exclude updaters that
# might potentially be depended on by an updater for a blank field.
#
# We also exclude updaters for any field that is going to get filled out by any delta_updater, because those
# updaters should take priority.
all_updaters = [updater for updater in all_updaters if all([updater[0] not in delta_field for delta_field, _, _ in delta_updaters])]
blank_updaters = dependedonby(all_updaters, [field for field in fact if isblankfield(fact[field])])
# Yes, this really works! This is because Python has reference comparison semantics on functions.
# However, this is the reason we need to use tuples of fields we depend on, rather than lists
updaters = set(delta_updaters + blank_updaters)
def shell(alreadyfilled):
# Gather all the fields we are newly able to fill given the most recent changes
# to the list of already filled things
cannowfill = FactoryDict(lambda _: [])
for updatewhat, updatefunction, updateusings in updaters:
if updatewhat not in alreadyfilled and all([updateusing in alreadyfilled for updateusing in updateusings]):
# We have to delay the computation of whether something is dirty or not until we are inside
# the actual thunk for this particular field, hence all the thunking and lambdas here
inputs = Thunk(lambda updateusings=updateusings: [graph[updateusing][1]() for updateusing in updateusings])
cannowfill[updatewhat].append((lambda inputs=inputs, updatefunction=updatefunction: updatefunction(*(inputs())),
lambda inputs=inputs, updateusings=updateusings: seq(inputs(), lambda: finddirties(updateusings))))
# Check for quiescence
if len(cannowfill) == 0:
# NB: could do something with the unfilled set (self.updateablefields.difference(alreadyfilled)) here
return
# Set up each fillable graph field with a thunk computing the value
for field, possiblefillers in cannowfill.items():
def fillme(field=field, possiblefillers=possiblefillers):
# For preference, use a filler that will certainly return clean information (i.e. sort by the number of dirty inputs and prefer the first - i.e. the one with fewer dirty inputs)
for fillerfunction, dirtyinputs, anyinputsdirty in sorted([(f, dirties(), len(dirties()) > 0) for f, dirties in possiblefillers], using(lambda x: x[2])):
if field not in fact or isblankfield(fact[field]) or anyinputsdirty:
# Don't know what the last value was or it may have changed: recompute.
#
# We also recompute if the incoming field is blank: this can happen if we have
# deleted the contents of a field, and then come back to update the fact by tabbing
# away from a filled field. Now we have enough information to fill it, and musn't
# just reuse the old version, which will be blank.
#
# Note that if the field was *generated* as blank one then it will have a marker
# in it, and so we won't pointlessly recompute its blankness.
log.info("Attempting to fill %s field -- dirty inputs are %s", field, dirtyinputs)
result = fillerfunction()
if result is None:
log.info("Filling %s failed -- falling back on another method", field)
continue
dirty[field] = cond(field in fact, lambda: result != unmarkgeneratedfield(fact[field]), lambda: anyinputsdirty)
return result
else:
# Last value must not have changed: retain it
assert (field in fact and not anyinputsdirty)
log.info("Retaining old field value for %s (blank: %s)", field, isblankfield(fact[field]))
dirty[field] = False
return unmarkgeneratedfield(fact[field])
# What if all of the possible updaters failed? Ideally we would not be in the graph at all, but it's too late for that.
# All we can do is return None, and deal with this possibility later on.
log.info("All possible updaters failed for the field %s", field)
dirty[field] = False
return None
graph[field] = (True, Thunk(fillme))
shell(alreadyfilled.union(set(cannowfill.keys())))
shell(initiallyfilledfields)
return graph
