def toType(pat):
"""Convert a pattern to a type, if possible (wildcards can't be converted to types).
:type pat: titus.P
:param pat: pattern to convert
:rtype: titus.datatype.Type
:return: corresponding type (titus.datatype.Type rather than titus.datatype.AvroType to allow for titus.datatype.FcnType)
"""
if isinstance(pat, Null): return AvroNull()
elif isinstance(pat, Boolean): return AvroBoolean()
elif isinstance(pat, Int): return AvroInt()
elif isinstance(pat, Long): return AvroLong()
elif isinstance(pat, Float): return AvroFloat()
elif isinstance(pat, Double): return AvroDouble()
elif isinstance(pat, Bytes): return AvroBytes()
elif isinstance(pat, String): return AvroString()
elif isinstance(pat, Array): return AvroArray(toType(pat.items))
elif isinstance(pat, Map): return AvroMap(toType(pat.values))
elif isinstance(pat, Union):
return AvroUnion([toType(x) for x in pat.types])
elif isinstance(pat, Fixed) and pat.fullName is not None:
namebits = pat.fullName.split(".")
if len(namebits) == 1:
return AvroFixed(pat.size, namebits[-1], None)
else:
return AvroFixed(pat.size, namebits[-1], ".".join(namebits[:-1]))
elif isinstance(pat, Fixed):
return AvroFixed(pat.size)
elif isinstance(pat, Enum) and pat.fullName is not None:
namebits = pat.fullName.split(".")
if len(namebits) == 1:
return AvroEnum(pat.symbols, namebits[-1], None)
else:
return AvroEnum(pat.symbols, namebits[-1], ".".join(namebits[:-1]))
elif isinstance(pat, Enum):
return AvroEnum(pat.symbols)
elif isinstance(pat, Record) and pat.fullName is not None:
namebits = pat.fullName.split(".")
if len(namebits) == 1:
return AvroRecord(
[AvroField(k, toType(v)) for k, v in list(pat.fields.items())],
namebits[-1], None)
else:
return AvroRecord(
[AvroField(k, toType(v)) for k, v in list(pat.fields.items())],
namebits[-1], ".".join(namebits[:-1]))
elif isinstance(pat, Record):
return AvroRecord(
[AvroField(k, toType(v)) for k, v in list(pat.fields.items())])
elif isinstance(pat, Fcn):
return FcnType([toType(x) for x in pat.params()], toType(pat.ret()))
else:
raise Exception
def assignRet(self, pat, assignments):
"""Apply the label assignments (e.g. "A" matched to "int", "B" matched to "string" etc.) to the return pattern.
:type pat: titus.P
:param pat: original return pattern
:type assignments: dict from label letters to titus.datatype.AvroType
:param assignments: assigned types to apply
:rtype: titus.datatype.AvroType
:return: resolved type for the return value of the signature
"""
if isinstance(pat, P.Null):
return AvroNull()
elif isinstance(pat, P.Boolean):
return AvroBoolean()
elif isinstance(pat, P.Int):
return AvroInt()
elif isinstance(pat, P.Long):
return AvroLong()
elif isinstance(pat, P.Float):
return AvroFloat()
elif isinstance(pat, P.Double):
return AvroDouble()
elif isinstance(pat, P.Bytes):
return AvroBytes()
elif isinstance(pat, P.String):
return AvroString()
elif isinstance(pat, P.Array):
return AvroArray(self.assignRet(pat.items, assignments))
elif isinstance(pat, P.Map):
return AvroMap(self.assignRet(pat.values, assignments))
elif isinstance(pat, P.Union):
return AvroUnion(
[self.assignRet(x, assignments) for x in pat.types])
elif isinstance(pat, P.Fixed):
return P.toType(pat)
elif isinstance(pat, P.Enum):
return P.toType(pat)
elif isinstance(pat, P.Record):
return P.toType(pat)
elif isinstance(pat, P.Fcn):
return P.toType(pat)
elif isinstance(pat, P.Wildcard):
return assignments[pat.label]
elif isinstance(pat, P.WildRecord):
return assignments[pat.label]
elif isinstance(pat, P.WildEnum):
return assignments[pat.label]
elif isinstance(pat, P.WildFixed):
return assignments[pat.label]
elif isinstance(pat, P.EnumFields):
return assignments[pat.label]
else:
raise Exception(repr(pat))
def ast(pfas,
check=True,
name=None,
randseed=None,
doc=None,
version=None,
metadata={},
options={},
tryYaml=False,
verbose=False):
"""Create a single PFA from a chained workflow, returning the result as an abstract syntax tree.
:type pfas: list of titus.pfaast.EngineConfig, Pythonized JSON, or JSON strings
:param pfas: PFA documents for which the output of document *i* is the input to document *i + 1*
:type check: bool
:param check: test the chained PFA for validity
:type name: string or ``None``
:param name: optional name for the chained PFA
:type randseed: integer or ``None``
:param randseed: optional random number seed for the chained PFA
:type doc: string or ``None``
:param doc: optional documentation string for the chained PFA
:type version: integer or ``None``
:param version: optional version number for the chained PFA
:type metadata: dict of strings
:param metadata: metadata for the chained PFA (default is ``{}``)
:type options: dict of Pythonized JSON
:param options: implementation options for the chained PFA (default is ``{}``)
:type tryYaml: bool
:param tryYaml: if ``True``, attempt to interpret ``pfas`` as YAML (assuming they fail as JSON)
:type verbose: bool
:param verbose: if ``True``, write status messages to standard output
:rtype: titus.pfaast.EngineConfig
:return: a PFA document representing the chained workflow
"""
# normalize all input forms to ASTs
if verbose:
sys.stderr.write(time.asctime() + " Converting all inputs to ASTs\n")
asts = []
for i, src in enumerate(pfas):
if verbose:
sys.stderr.write(time.asctime() + " step {0}\n".format(i + 1))
if isinstance(src, EngineConfig):
ast = src
elif isinstance(src, dict):
ast = titus.reader.jsonToAst(src)
else:
try:
ast = titus.reader.jsonToAst(src)
except ValueError:
if tryYaml:
ast = titus.reader.yamlToAst(src)
else:
raise
asts.append(ast)
pfas = asts
# helper functions for transforming names
def split(t):
if "." in t:
return t[:t.rindex(".")], t[t.rindex(".") + 1:]
else:
return None, t
def join(ns, n):
if ns is None or ns == "":
return n
else:
return ns + "." + n
def prefixType(i, pfa, t):
ns, n = split(t)
return join(ns, "Step{0:d}_{1}_{2}".format(i + 1, pfa.name, n))
def prefixAction(i, pfa):
return "step{0:d}_{1}_action".format(i + 1, pfa.name)
def prefixFcnRef(i, pfa, x):
if x.startswith("u."):
return "u.step{0:d}_{1}_fcn_{2}".format(i + 1, pfa.name, x[2:])
else:
return x
def prefixFcnDef(i, pfa, x):
return "step{0:d}_{1}_fcn_{2}".format(i + 1, pfa.name, x)
def prefixCell(i, pfa, x):
return "step{0:d}_{1}_{2}".format(i + 1, pfa.name, x)
def prefixPool(i, pfa, x):
return "step{0:d}_{1}_{2}".format(i + 1, pfa.name, x)
# define new names for all types to avoid type name collisions
if verbose:
sys.stderr.write(time.asctime() +
" Changing type names to avoid collisions\n")
originalNameToNewName = {}
for i, pfa in enumerate(pfas):
originalNameToNewName[i] = {}
for typeName in list(pfa.inputPlaceholder.parser.names.names.keys()):
keyTypeName = typeName
if (typeName[0] == "."):
keyTypeName = keyTypeName[1:]
originalNameToNewName[i][keyTypeName] = prefixType(
i, pfa, typeName)
# but any names in the input to the first and the output from the last should not be changed
def trivialName(i, avroType, memo):
if isinstance(avroType, AvroArray):
trivialName(i, avroType.items, memo)
elif isinstance(avroType, AvroMap):
trivialName(i, avroType.values, memo)
elif isinstance(avroType, AvroUnion):
for t in avroType.types:
trivialName(i, t, memo)
elif isinstance(avroType, (AvroFixed, AvroEnum)):
t = avroType.fullName
originalNameToNewName[i][t] = t
elif isinstance(avroType, AvroRecord):
t = avroType.fullName
if t not in memo:
memo.add(t)
for f in avroType.fields:
trivialName(i, f.avroType, memo)
originalNameToNewName[i][t] = t
trivialName(0, pfas[0].input, set())
trivialName(len(pfas) - 1, pfas[-1].output, set())
# ensure that chained types match and will be given the same names
if verbose:
sys.stderr.write(
time.asctime() +
" Verifying that input/output schemas match along the chain\n")
def chainPair(i, first, second, memo):
if isinstance(first, AvroNull) and isinstance(second, AvroNull):
return True
elif isinstance(first, AvroBoolean) and isinstance(
second, AvroBoolean):
return True
elif isinstance(first, AvroInt) and isinstance(second, AvroInt):
return True
elif isinstance(first, AvroLong) and isinstance(second, AvroLong):
return True
elif isinstance(first, AvroFloat) and isinstance(second, AvroFloat):
return True
elif isinstance(first, AvroDouble) and isinstance(second, AvroDouble):
return True
elif isinstance(first, AvroBytes) and isinstance(second, AvroBytes):
return True
elif isinstance(first, AvroFixed) and isinstance(second, AvroFixed):
if first.size == second.size:
originalNameToNewName[i + 1][
second.fullName] = originalNameToNewName[i][first.fullName]
return True
else:
return False
elif isinstance(first, AvroString) and isinstance(second, AvroString):
return True
elif isinstance(first, AvroEnum) and isinstance(second, AvroEnum):
if first.symbols == second.symbols:
originalNameToNewName[i + 1][
second.fullName] = originalNameToNewName[i][first.fullName]
return True
else:
return False
elif isinstance(first, AvroArray) and isinstance(second, AvroArray):
return chainPair(i, first.items, second.items, memo)
elif isinstance(first, AvroMap) and isinstance(second, AvroMap):
return chainPair(i, first.values, second.values, memo)
elif isinstance(first, AvroRecord) and isinstance(second, AvroRecord):
if first.fullName not in memo:
memo.add(first.fullName)
if len(first.fields) != len(second.fields):
return False
for f1, f2 in zip(first.fields, second.fields):
if f1.name != f2.name:
return False
elif not chainPair(i, f1.avroType, f2.avroType, memo):
return False
originalNameToNewName[i + 1][
second.fullName] = originalNameToNewName[i][first.fullName]
return True
elif isinstance(first, AvroUnion) and isinstance(second, AvroUnion):
for yt in second.types:
if not any(chainPair(i, xt, yt, memo) for xt in first.types):
return False
return True
else:
return False
for i in range(len(pfas) - 1):
first = pfas[i].output
second = pfas[i + 1].input
if not chainPair(i, first, second, set()):
raise PFAChainError(
"output of engine {0}: {1} not compatible with input of engine {2}: {3}"
.format(i + 1, ts(first), i + 2, ts(second)))
def rename(i, avroType, memo):
if isinstance(avroType, AvroArray):
return {"type": "array", "items": rename(i, avroType.items, memo)}
elif isinstance(avroType, AvroMap):
return {"type": "map", "values": rename(i, avroType.values, memo)}
elif isinstance(avroType, AvroUnion):
return [rename(i, t, memo) for t in avroType.types]
elif isinstance(avroType, AvroFixed):
ns, n = split(originalNameToNewName[i][avroType.fullName])
out = {"type": "fixed", "name": n, "size": avroType.size}
if ns is not None:
out["namespace"] = ns
return out
elif isinstance(avroType, AvroEnum):
ns, n = split(originalNameToNewName[i][avroType.fullName])
out = {"type": "enum", "name": n, "symbols": avroType.symbols}
if ns is not None:
out["namespace"] = ns
return out
elif isinstance(avroType, AvroRecord):
newName = originalNameToNewName[i][avroType.fullName]
if newName in memo:
return memo[newName]
else:
ns, n = split(newName)
out = {"type": "record", "name": n, "fields": []}
if ns is not None:
out["namespace"] = ns
memo[newName] = join(ns, n)
for f in avroType.fields:
newf = {
"name": f.name,
"type": rename(i, f.avroType, memo)
}
if f.default is not None:
newf["default"] = f.default
if f.order is not None:
newf["order"] = f.order
out["fields"].append(newf)
return out
else:
return jsonlib.loads(repr(avroType))
avroTypeBuilder = AvroTypeBuilder()
memo = {}
def newPlaceholder(i, oldAvroType):
newAvroType = rename(i, oldAvroType, {})
return avroTypeBuilder.makePlaceholder(jsonlib.dumps(newAvroType),
memo)
# combined name, if not explicitly set
if name is None:
name = "Chain_" + "_".join(pfa.name for pfa in pfas)
# combined method (fold not supported yet, but could be)
method = Method.MAP
for pfa in pfas:
if pfa.method == Method.EMIT:
method = Method.EMIT
elif pfa.method == Method.FOLD:
raise NotImplementedError(
"chaining of fold-type scoring engines has not been implemented yet"
)
# no zero or merge until we support fold method
zero = None
merge = None
# input/output types from first and last
inputPlaceholder = newPlaceholder(0, pfas[0].input)
outputPlaceholder = newPlaceholder(len(pfas) - 1, pfas[-1].output)
if verbose:
sys.stderr.write(
time.asctime() +
" Adding [name, instance, metadata, actionsStarted, actionsFinished, version] as model parameters\n"
)
cells = {
"name":
Cell(newPlaceholder(0, AvroString()), jsonlib.dumps(""), False, False,
CellPoolSource.EMBEDDED),
"instance":
Cell(newPlaceholder(0, AvroInt()), jsonlib.dumps(0), False, False,
CellPoolSource.EMBEDDED),
"metadata":
Cell(newPlaceholder(0, AvroMap(AvroString())), jsonlib.dumps({}),
False, False, CellPoolSource.EMBEDDED),
"actionsStarted":
Cell(newPlaceholder(0, AvroLong()), jsonlib.dumps(0), False, False,
CellPoolSource.EMBEDDED),
"actionsFinished":
Cell(newPlaceholder(0, AvroLong()), jsonlib.dumps(0), False, False,
CellPoolSource.EMBEDDED)
}
if version is not None:
cells["version"] = Cell(newPlaceholder(0, AvroInt()), 0, False, False,
CellPoolSource.EMBEDDED)
pools = {}
if verbose:
sys.stderr.write(time.asctime() +
" Converting scoring engine algorithm\n")
# all code will go into user functions, including begin/action/end
fcns = {}
begin = [
CellTo("name", [], Ref("name")),
CellTo("instance", [], Ref("instance")),
CellTo("metadata", [], Ref("metadata"))
]
if version is not None:
begin.append(CellTo("version", [], Ref("version")))
action = [
CellTo("actionsStarted", [], Ref("actionsStarted")),
CellTo("actionsFinished", [], Ref("actionsFinished"))
]
end = [
CellTo("actionsStarted", [], Ref("actionsStarted")),
CellTo("actionsFinished", [], Ref("actionsFinished"))
]
for i, pfa in enumerate(pfas):
if verbose:
sys.stderr.write(time.asctime() +
" step {0}: {1}\n".format(i + 1, pfa.name))
thisActionFcnName = prefixAction(i, pfa)
if i + 1 < len(pfas):
nextActionFcnName = prefixAction(i + 1, pfas[i + 1])
else:
nextActionFcnName = None
# this is a closure; it must be defined in the loop to pick up free variables
lazyFcnReplacer = None
def genericReplacer(expr, self):
if isinstance(expr, FcnDef):
return FcnDef(
[{
list(t.keys())[0]: newPlaceholder(
i,
list(t.values())[0])
} for t in expr.params],
newPlaceholder(i, expr.ret),
[
x.replace(lazyFcnReplacer) for x in expr.body
], # this is the one place where we should pass down fcnReplacer rather than self
expr.pos)
elif isinstance(expr, FcnRef):
return FcnRef(prefixFcnRef(i, pfa, expr.name), expr.pos)
elif isinstance(expr, FcnRefFill):
return FcnRefFill(
prefixFcnRef(i, pfa, expr.name),
dict((k, v.replace(self))
for k, v in list(expr.fill.items())), expr.pos)
elif isinstance(
expr, CallUserFcn
): # TODO: need to change the symbols of the corresponding enum
return CallUserFcn(expr.name.replace(self),
[x.replace(self) for x in expr.args],
expr.pos)
elif isinstance(expr, Call):
if pfa.method == Method.EMIT and i + 1 < len(
pfas) and expr.name == "emit":
return Call("u." + nextActionFcnName,
[x.replace(self) for x in expr.args], expr.pos)
else:
return Call(prefixFcnRef(i, pfa, expr.name),
[x.replace(self) for x in expr.args], expr.pos)
elif isinstance(expr, Literal):
return Literal(newPlaceholder(i, expr.avroType), expr.value,
expr.pos)
elif isinstance(expr, NewObject):
return NewObject(
dict((k, v.replace(self))
for k, v in list(expr.fields.items())),
newPlaceholder(i, expr.avroType), expr.pos)
elif isinstance(expr, NewArray):
return NewArray([x.replace(self) for x in expr.items],
newPlaceholder(i, expr.avroType), expr.pos)
elif isinstance(expr, CellGet):
return CellGet(prefixCell(i, pfa, expr.cell),
[x.replace(self) for x in expr.path], expr.pos)
elif isinstance(expr, CellTo):
return CellTo(prefixCell(i, pfa, expr.cell),
[x.replace(self) for x in expr.path],
expr.to.replace(self), expr.pos)
elif isinstance(expr, PoolGet):
return PoolGet(prefixPool(i, pfa, expr.pool),
[x.replace(self) for x in expr.path], expr.pos)
elif isinstance(expr, PoolTo):
return PoolTo(prefixPool(i, pfa, expr.pool),
[x.replace(self) for x in expr.path],
expr.to.replace(self), expr.init.replace(self),
expr.pos)
elif isinstance(expr, CastCase):
return CastCase(newPlaceholder(i, expr.avroType), expr.named,
[x.replace(self) for x in expr.body], expr.pos)
elif isinstance(expr, Upcast):
return Upcast(expr.expr.replace(self),
newPlaceholder(i, expr.avroType), expr.pos)
genericReplacer.isDefinedAt = lambda x: isinstance(
x, (FcnDef, FcnRef, FcnRefFill, CallUserFcn, Call, Literal,
NewObject, CellGet, CellTo, PoolGet, PoolTo, CastCase, Upcast))
def fcnReplacer(expr):
return genericReplacer(expr, fcnReplacer)
fcnReplacer.isDefinedAt = genericReplacer.isDefinedAt
lazyFcnReplacer = fcnReplacer
# add statements to begin
def beginReplacer(expr):
if isinstance(expr, Ref):
if expr.name in ("name", "instance",
"metadata") or (version is not None
and expr.name == "version"):
return CellGet(expr.name, [], expr.pos)
else:
return expr
else:
return genericReplacer(expr, beginReplacer)
beginReplacer.isDefinedAt = lambda x: isinstance(
x, Ref) or genericReplacer.isDefinedAt(x)
begin.extend([x.replace(beginReplacer) for x in pfa.begin])
# add statements to end
def endReplacer(expr):
if isinstance(expr, Ref):
if expr.name in ("name", "instance", "metadata",
"actionsStarted", "actionsFinished") or (
version is not None
and expr.name == "version"):
return CellGet(expr.name, [], expr.pos)
else:
return expr
else:
return genericReplacer(expr, endReplacer)
endReplacer.isDefinedAt = lambda x: isinstance(
x, Ref) or genericReplacer.isDefinedAt(x)
end.extend([x.replace(endReplacer) for x in pfa.end])
# convert the action into a user function
def actionReplacer(expr):
if isinstance(expr, Ref):
if expr.name in ("name", "instance", "metadata",
"actionsStarted", "actionsFinished") or (
version is not None
and expr.name == "version"):
return CellGet(expr.name, [], expr.pos)
else:
return expr
else:
return genericReplacer(expr, actionReplacer)
actionReplacer.isDefinedAt = lambda x: isinstance(
x, Ref) or genericReplacer.isDefinedAt(x)
body = [x.replace(actionReplacer) for x in pfa.action]
if method == Method.MAP:
# if the overall method is MAP, then we know that all of the individual engines are MAP
# the overall action calls a nested chain of engines-as-functions and each engine-as-a-function just does its job and returns (body is unmodified)
fcns[thisActionFcnName] = FcnDef(
[{
"input": newPlaceholder(i, pfa.input)
}], newPlaceholder(i, pfa.output), body)
if i == 0:
action.append(Call("u." + thisActionFcnName, [Ref("input")]))
else:
action[-1] = Call("u." + thisActionFcnName, [action[-1]])
elif method == Method.EMIT:
# if the overall method is EMIT, then some individual engines might be MAP or might be EMIT
# the overall action calls the first engine-as-a-function and the engines-as-functions call each other (body is modified)
if pfa.method == Method.MAP and i + 1 < len(pfas):
body = [
Call("u." + nextActionFcnName, [Do(body)]),
LiteralNull()
]
elif pfa.method == Method.MAP:
body = [Call("emit", [Do(body)])]
elif pfa.method == Method.EMIT:
body.append(LiteralNull())
fcns[thisActionFcnName] = FcnDef(
[{
"input": newPlaceholder(i, pfa.input)
}], newPlaceholder(i, AvroNull()), body)
if i == 0:
action.append(Call("u." + thisActionFcnName, [Ref("input")]))
# convert all of the user functions into user functions
for fcnName, fcnDef in list(pfa.fcns.items()):
# note: some of these user-defined functions may call emit; if so, they'll call the right emit
fcns[prefixFcnDef(i, pfa, fcnName)] = FcnDef(
[{
list(t.keys())[0]: newPlaceholder(i,
list(t.values())[0])
} for t in fcnDef.paramsPlaceholder],
newPlaceholder(i, fcnDef.ret),
[x.replace(fcnReplacer) for x in fcnDef.body], fcnDef.pos)
if verbose:
sys.stderr.write(time.asctime() +
" Create types for model parameters\n")
for i, pfa in enumerate(pfas):
if verbose and len(pfa.cells) > 0:
sys.stderr.write(time.asctime() + " step {0}:\n".format(i + 1))
for cellName, cell in list(pfa.cells.items()):
if verbose:
sys.stderr.write(time.asctime() +
" cell {0}\n".format(cellName))
newCell = Cell(newPlaceholder(i, cell.avroType), cell.init,
cell.shared, cell.rollback, cell.source, cell.pos)
cells[prefixCell(i, pfa, cellName)] = newCell
if cell.source == "embedded":
def converter(avroType):
original = jsonDecoder(cell.avroType,
jsonlib.loads(cell.init))
return jsonlib.dumps(jsonEncoder(avroType, original))
newCell.converter = converter
for i, pfa in enumerate(pfas):
if verbose and len(pfa.pools) > 0:
sys.stderr.write(time.asctime() + " step {0}:\n".format(i + 1))
for poolName, pool in list(pfa.pools.items()):
if verbose:
sys.stderr.write(time.asctime() +
" pool {0}\n".format(poolName))
newPool = Pool(newPlaceholder(i, pool.avroType), pool.init,
pool.shared, pool.rollback, pool.source, pool.pos)
pools[prefixPool(i, pfa, poolName)] = newPool
if pool.source == "embedded":
def converter(avroType):
original = jsonDecoder(pool.avroType,
jsonlib.loads(pool.init))
return jsonlib.dumps(jsonEncoder(avroType, original))
newPool.converter = converter
# make sure all the types work together
if verbose: sys.stderr.write(time.asctime() + " Resolving all types\n")
avroTypeBuilder.resolveTypes()
if verbose:
sys.stderr.write(time.asctime() +
" Converting the model parameters themselves\n")
for i, pfa in enumerate(pfas):
if verbose and len(pfa.cells) > 0:
sys.stderr.write(time.asctime() + " step {0}:\n".format(i + 1))
for cellName, cell in list(pfa.cells.items()):
if verbose:
sys.stderr.write(time.asctime() +
" cell {0}\n".format(cellName))
if cell.source == "embedded":
newCell = cells[prefixCell(i, pfa, cellName)]
newCell.init = newCell.converter(newCell.avroType)
for i, pfa in enumerate(pfas):
if verbose and len(pfa.pools) > 0:
sys.stderr.write(time.asctime() + " step {0}:\n".format(i + 1))
for poolName, pool in list(pfa.pools.items()):
if verbose:
sys.stderr.write(time.asctime() +
" pool {0}\n".format(poolName))
if pool.source == "embedded":
newPool = pools[prefixPool(i, pfa, poolName)]
newPool.init = newPool.converter(newPool.avroType)
# randseed, doc, version, metadata, and options need to be explicitly set
# return a (possibly checked) AST
out = EngineConfig(name, method, inputPlaceholder, outputPlaceholder,
begin, action, end, fcns, zero, merge, cells, pools,
randseed, doc, version, metadata, options)
if check:
if verbose:
sys.stderr.write(time.asctime() + " Verifying PFA validity\n")
PFAEngine.fromAst(out)
if verbose: sys.stderr.write(time.asctime() + " Done\n")
return out
def assign(self, pat, arg, assignments):
"""Apply the label assignments (e.g. "A" matched to "int", "B" matched to "string", etc.) to each parameter of the signature.
:type pat: titus.P
:param pat: original parameter pattern
:type arg: titus.datatype.AvroType
:param arg: supplied argument type
:type assignments: dict from label letters to titus.datatype.AvroType
:param assignments: assigned types to apply
:rtype: titus.datatype.AvroType
:return: resolved type for one parameter of the signature
"""
if isinstance(pat, P.Null) and isinstance(arg, AvroNull):
return arg
elif isinstance(pat, P.Boolean) and isinstance(arg, AvroBoolean):
return arg
elif isinstance(pat, P.Int) and isinstance(arg, AvroInt):
return AvroInt()
elif isinstance(pat, P.Long) and (isinstance(arg, AvroInt)
or isinstance(arg, AvroLong)):
return AvroLong()
elif isinstance(pat, P.Float) and (isinstance(arg, AvroInt)
or isinstance(arg, AvroLong)
or isinstance(arg, AvroFloat)):
return AvroFloat()
elif isinstance(pat, P.Double) and (isinstance(arg, AvroInt)
or isinstance(arg, AvroLong)
or isinstance(arg, AvroFloat)
or isinstance(arg, AvroDouble)):
return AvroDouble()
elif isinstance(pat, P.Bytes) and isinstance(arg, AvroBytes):
return arg
elif isinstance(pat, P.String) and isinstance(arg, AvroString):
return arg
elif isinstance(pat, P.Array) and isinstance(arg, AvroArray):
return AvroArray(
P.mustBeAvro(self.assign(pat.items, arg.items, assignments)))
elif isinstance(pat, P.Map) and isinstance(arg, AvroMap):
return AvroMap(
P.mustBeAvro(self.assign(pat.values, arg.values, assignments)))
elif isinstance(pat, P.Union) and isinstance(arg, AvroUnion):
return arg
elif isinstance(pat, P.Union) and isinstance(arg, AvroType):
return arg
elif isinstance(pat, P.Fixed) and isinstance(arg, AvroFixed):
return arg
elif isinstance(pat, P.Enum) and isinstance(arg, AvroEnum):
return arg
elif isinstance(pat, P.Record) and isinstance(arg, AvroRecord):
return arg
elif isinstance(pat, P.Fcn) and isinstance(arg, FcnType):
return arg
elif isinstance(pat, P.Wildcard):
return assignments[pat.label]
elif isinstance(pat, P.WildRecord):
return assignments[pat.label]
elif isinstance(pat, P.WildEnum):
return assignments[pat.label]
elif isinstance(pat, P.WildFixed):
return assignments[pat.label]
elif isinstance(pat, P.EnumFields):
return assignments[pat.label]
else:
raise Exception
def broadestType(candidates):
"""Compute the narrowest possible supertype of a set of types.
:type candidates: list of titus.datatype.AvroType
:param candidates: set of types for which to find the narrowest possible supertype
:rtype: titus.datatype.AvroType
:return: narrowest possible supertype, usually a union of the candidates
"""
realCandidates = [
x for x in candidates if not isinstance(x, ExceptionType)
]
if len(candidates) == 0:
return ValueError("empty list of types")
elif len(realCandidates) == 0:
return ValueError("list of types consists only of exception type")
elif all(isinstance(x, AvroNull) for x in realCandidates):
return realCandidates[0]
elif all(isinstance(x, AvroBoolean) for x in realCandidates):
return realCandidates[0]
elif all(isinstance(x, AvroInt) for x in realCandidates):
return AvroInt()
elif all(
isinstance(x, AvroInt) or isinstance(x, AvroLong)
for x in realCandidates):
return AvroLong()
elif all(
isinstance(x, AvroInt) or isinstance(x, AvroLong)
or isinstance(x, AvroFloat) for x in realCandidates):
return AvroFloat()
elif all(
isinstance(x, AvroInt) or isinstance(x, AvroLong)
or isinstance(x, AvroFloat) or isinstance(x, AvroDouble)
for x in realCandidates):
return AvroDouble()
elif all(isinstance(x, AvroBytes) for x in realCandidates):
return realCandidates[0]
elif all(isinstance(x, AvroString) for x in realCandidates):
return realCandidates[0]
elif all(isinstance(x, AvroArray) for x in realCandidates):
return AvroArray(
P.mustBeAvro(
LabelData.broadestType([x.items for x in realCandidates])))
elif all(isinstance(x, AvroMap) for x in realCandidates):
return AvroMap(
P.mustBeAvro(
LabelData.broadestType([x.values
for x in realCandidates])))
elif all(isinstance(x, AvroFixed) for x in realCandidates):
fullName = realCandidates[0].fullName
if all(x.fullName == fullName for x in realCandidates[1:]):
return realCandidates[0]
else:
raise IncompatibleTypes("incompatible fixed types: " +
" ".join(map(repr, realCandidates)))
elif all(isinstance(x, AvroEnum) for x in realCandidates):
fullName = realCandidates[0].fullName
if all(x.fullName == fullName for x in realCandidates[1:]):
return realCandidates[0]
else:
raise IncompatibleTypes("incompatible enum types: " +
" ".join(map(repr, realCandidates)))
elif all(isinstance(x, AvroRecord) for x in realCandidates):
fullName = realCandidates[0].fullName
if all(x.fullName == fullName for x in realCandidates[1:]):
return realCandidates[0]
else:
raise IncompatibleTypes("incompatible record types: " +
" ".join(map(repr, realCandidates)))
elif all(isinstance(x, FcnType) for x in realCandidates):
params = realCandidates[0].params
ret = realCandidates[0].ret
if all(x.params == params and x.ret == ret
for x in realCandidates[1:]):
return realCandidates[0]
else:
raise IncompatibleTypes("incompatible function types: " +
" ".join(map(repr, realCandidates)))
elif not any(isinstance(x, FcnType) for x in realCandidates):
types = LabelData.distinctTypes(realCandidates)
types = [P.mustBeAvro(x) for x in types]
countFixed = 0
countEnum = 0
for t in types:
if isinstance(t, AvroFixed):
countFixed += 1
if isinstance(t, AvroEnum):
countEnum += 1
if countFixed > 1:
raise IncompatibleTypes("incompatible fixed types")
if countEnum > 1:
raise IncompatibleTypes("incompatible enum types")
return AvroUnion(types)
else:
raise IncompatibleTypes(
"incompatible function/non-function types: " +
" ".join(map(repr, realCandidates)))