#!/usr/bin/env python """Provides support functions and classes for parsers. """ from copy import deepcopy from cogent.util.misc import iterable __author__ = "Rob Knight" __copyright__ = "Copyright 2007-2009, The Cogent Project" __credits__ = ["Rob Knight", "Peter Maxwell"] __license__ = "GPL" __version__ = "1.4.1" __maintainer__ = "Rob Knight" __email__ = "rob@spot.colorado.edu" __status__ = "Development" class FileFormatError(Exception): """Exception raised when a file can not be parsed.""" pass class RecordError(FileFormatError): """Exception raised when a record is bad.""" pass class FieldError(RecordError): """Exception raised when a field within a record is bad.""" pass class Grouper(object): """Acts as iterator that returns lists of n items at a time from seq. Note: returns a partial list if not evenly divisible by n. """ def __init__(self, NumItems=1): """Returns new Grouper object: will return n items at a time from seq""" self.NumItems = NumItems def __call__(self, seq): """Returns iterator over seq, returning n items at a time.""" try: num = int(self.NumItems) assert num >= 1 except: raise ValueError, "Grouper.NumItems must be positive int, not %s" \ % (self.NumItems) curr = [] for i, item in enumerate(seq): if (i % num == 0) and curr: yield curr curr = [item] else: curr.append(item) #return any leftover items if curr: yield curr #Example of the instances Grouper provides: ByPairs = Grouper(2) def string_and_strip(*items): """Converts items to strings and strips them.""" return [str(i).strip() for i in items] def DelimitedSplitter(delimiter=None, max_splits=1): """Returns function that returns stripped fields split by delimiter. Unlike the default behavior of split, max_splits can be negative, in which case it counts from the end instead of the start (i.e. splits at the _last_ delimiter, last two delimiters, etc. for -1, -2, etc.) However, if the delimiter is None (the default) and max_splits is negative, will not preserve internal spaces. Note: leaves empty fields in place. """ is_int = isinstance(max_splits, int) or isinstance(max_splits, long) if is_int and (max_splits > 0): def parser(line): return [i.strip() for i in line.split(delimiter, max_splits)] elif is_int and (max_splits < 0): def parser(line): to_insert = delimiter or ' ' #re-join fields w/ space if None fields = line.split(delimiter) if (fields == []) or (fields == ['']): return [] #empty string or only delimiter: return nothing #if not enough fields, count from the start, not the end if len(fields) < max_splits: first_fields = fields[0] last_fields = fields[1:] #otherwise, count off the last n fields and join the remainder else: first_fields = fields[:max_splits] last_fields = fields[max_splits:] pieces = [] #if first_fields is empty, don't make up an extra empty string if first_fields: pieces.append(to_insert.join(first_fields)) pieces.extend(last_fields) return [i.strip() for i in pieces] else: #ignore max_splits if it was 0 def parser(line): return [i.strip() for i in line.split(delimiter)] return parser #The following provide examples of the kinds of functions DelimitedSplitter #returns. semi_splitter = DelimitedSplitter(';', None) space_pairs = DelimitedSplitter(None) equal_pairs = DelimitedSplitter('=') last_colon = DelimitedSplitter(':', -1) class GenericRecord(dict): """Holds data for a generic field ->: value mapping. Override Required with {name:prototype} mapping. Each required name will get a deepcopy of its prototype. For example, use an empty list to guarantee that each instance has its own list for a particular field to which items can be appended. Raises AttributeError on attempt to delete required item, but does not raise an exception on attempt to delete absent item. This class explicitly does _not_ override __getitem__ or __setitem__ for performance reasons: if you need to transform keys on get/set or if you need to access items as attributes and vice versa, use MappedRecord instead. """ Required = {} def __init__(self, *args, **kwargs): """Reads kwargs as properties of self.""" #perform init on temp dict to preserve interface: will then translate #aliased keys when loading into self temp = {} dict.__init__(temp, *args, **kwargs) self.update(temp) for name, prototype in self.Required.iteritems(): if not name in self: self[name] = deepcopy(prototype) def __delitem__(self, item): """Deletes item or raises exception if item required. Note: Fails silently if item absent. """ if item in self.Required: raise AttributeError, "%s is a required item" % (item,) try: super(GenericRecord, self).__delitem__(item) except KeyError: pass def copy(self): """Coerces copy to correct type""" temp = self.__class__(super(GenericRecord,self).copy()) #don't forget to copy attributes! for attr, val in self.__dict__.iteritems(): temp.__dict__[attr] = deepcopy(val) return temp class MappedRecord(GenericRecord): """GenericRecord that maps names of fields onto standardized names. Override Aliases in subclass for new mapping of OldName->NewName. Each OldName can have only one NewName, but it's OK if several OldNames map to the same NewName. Note: can access fields either as items or as attributes. In addition, can access either using nonstandard names or using standard names. Implementation note: currently, just a dict with appropriate get/set overrides and ability to access items as attributes. Attribute access is about 10x slower than in GenericRecord, so make sure you need the additional capabilities if you use MappedRecord instead of GenericRecord. WARNING: MappedRecord pretends to have every attribute, so will never raise AttributeError when trying to find an unknown attribute. This feature can cause surprising interactions when a Delegator is delegating its attributes to a MappedRecord, since any attributes defined in __init__ will be set in the MappedRecord and not in the object itself. The solution is to use the self.__dict__['AttributeName'] = foo syntax to force the attributes to be set in the object and not the MappedRecord to which it forwards. """ Aliases = {} DefaultValue = None def _copy(self, prototype): """Returns a copy of item.""" if hasattr(prototype, 'copy'): return prototype.copy() elif isinstance(prototype, list): return prototype[:] elif isinstance(prototype,str) or isinstance(prototype,int) or\ isinstance(prototype,long) or isinstance(prototype,tuple)\ or isinstance(prototype,complex) or prototype is None: return prototype #immutable type: use directly else: return deepcopy(prototype) def __init__(self, *args, **kwargs): """Reads kwargs as properties of self.""" #perform init on temp dict to preserve interface: will then translate #aliased keys when loading into self temp = {} unalias = self.unalias dict.__init__(temp, *args, **kwargs) for key, val in temp.iteritems(): self[unalias(key)] = val for name, prototype in self.Required.iteritems(): new_name = unalias(name) if not new_name in self: self[new_name] = self._copy(prototype) def unalias(self, key): """Returns dealiased name for key, or key if not in alias.""" try: return self.Aliases.get(key, key) except TypeError: return key def __getattr__(self, attr): """Returns None if field is absent, rather than raising exception.""" if attr in self: return self[attr] elif attr in self.__dict__: return self.__dict__[attr] elif attr.startswith('__'): #don't retrieve private class attrs raise AttributeError elif hasattr(self.__class__, attr): return getattr(self.__class__, attr) else: return self._copy(self.DefaultValue) def __setattr__(self, attr, value): """Sets attribute in self if absent, converting name if necessary.""" normal_attr = self.unalias(attr) #we overrode __getattr__, so have to simulate getattr(self, attr) by #calling superclass method and checking for AttributeError. #BEWARE: dict defines __getattribute__, not __getattr__! try: super(MappedRecord, self).__getattribute__(normal_attr) super(MappedRecord, self).__setattr__(normal_attr, value) except AttributeError: self[normal_attr] = value def __delattr__(self, attr): """Deletes attribute, converting name if necessary. Fails silently.""" normal_attr = self.unalias(attr) if normal_attr in self.Required: raise AttributeError, "%s is a required attribute" % (attr,) else: try: super(MappedRecord, self).__delattr__(normal_attr) except AttributeError: del self[normal_attr] def __getitem__(self, item): """Returns default if item is absent, rather than raising exception.""" normal_item = self.unalias(item) return self.get(normal_item, self._copy(self.DefaultValue)) def __setitem__(self, item, val): """Sets item, converting name if necessary.""" super(MappedRecord, self).__setitem__(self.unalias(item), val) def __delitem__(self, item): """Deletes item, converting name if necessary. Fails silently.""" normal_item = self.unalias(item) super(MappedRecord, self).__delitem__(normal_item) def __contains__(self, item): """Tests membership, converting name if necessary.""" return super(MappedRecord, self).__contains__(self.unalias(item)) def get(self, item, default): """Returns self[item] or default if not present. Silent when unhashable.""" try: return super(MappedRecord, self).get(self.unalias(item), default) except TypeError: return default def setdefault(self, key, default=None): """Returns self[key] or default (and sets self[key]=default)""" return super(MappedRecord, self).setdefault(self.unalias(key),default) def update(self, *args, **kwargs): """Updates self with items in other""" temp = {} unalias = self.unalias temp.update(*args, **kwargs) for key, val in temp.iteritems(): self[unalias(key)] = val #The following methods are useful for handling particular types of fields in #line-oriented parsers def TypeSetter(constructor=None): """Returns function that takes obj, field, val and sets obj.field = val. constructor can be any callable that returns an object. """ if constructor: def setter(obj, field, val): setattr(obj, field, constructor(val)) else: def setter(obj, field, val): setattr(obj, field, val) return setter int_setter = TypeSetter(int) str_setter = TypeSetter(str) list_setter = TypeSetter(list) tuple_setter = TypeSetter(tuple) dict_setter = TypeSetter(dict) float_setter = TypeSetter(float) complex_setter = TypeSetter(complex) bool_setter = TypeSetter(bool) identity_setter = TypeSetter() def list_adder(obj, field, val): """Adds val to list in obj.field, creating list if necessary.""" try: getattr(obj, field).append(val) except AttributeError: setattr(obj, field, [val]) def list_extender(obj, field, val): """Adds val to list in obj.field, creating list if necessary.""" try: getattr(obj, field).extend(iterable(val)) except AttributeError: setattr(obj, field, list(val)) def dict_adder(obj, field, val): """If val is a sequence, adds key/value pair in obj.field: else adds key.""" try: key, value = val except (ValueError, TypeError): key, value = val, None try: getattr(obj, field)[key] = value except AttributeError: setattr(obj, field, {key:value}) class LineOrientedConstructor(object): """Constructs a MappedRecord from a sequence of lines.""" def __init__(self, Lines=None, LabelSplitter=space_pairs, FieldMap=None, Constructor=MappedRecord, Strict=False): """Returns new LineOrientedConstructor. Fields: Lines: set of lines to construct record from (for convenience). Default is None. LabelSplitter: function that returns (label, data) tuple. Default is to split on first space and strip components. FieldMap: dict of {fieldname:handler} functions. Each function has the signature (obj, field, val) and performs an inplace action like setting field to val or appending val to field. Default is empty dict. Constructor: constructor for the resulting object. Default is MappedRecord: beware of using constructors that don't subclass MappedRecord. Strict: boolean controlling whether to raise error on unrecognized field. Default is False. """ self.Lines = Lines or [] self.LabelSplitter = LabelSplitter self.FieldMap = FieldMap or {} self.Constructor = Constructor self.Strict = Strict def __call__(self, Lines=None): """Returns the record constructed from Lines, or self.Lines""" if Lines is None: Lines = self.Lines result = self.Constructor() fieldmap = self.FieldMap aka = result.unalias splitter = self.LabelSplitter for line in Lines: #find out how many items we got, setting key and val appropiately items = list(splitter(line)) num_items = len(items) if num_items == 2: #typical case: key-value pair raw_field, val = items elif num_items > 2: raw_field = items[0] val = items[1:] elif len(items) == 1: raw_field, val = items[0], None elif not items: #presumably had line with just a delimiter? continue #figure out if we know the field under its original name or as #an alias if raw_field in fieldmap: field, mapper = raw_field, fieldmap[raw_field] else: new_field = aka(raw_field) if new_field in fieldmap: field, mapper = new_field, fieldmap[new_field] else: if self.Strict: raise FieldError, \ "Got unrecognized field %s" % (raw_field,) else: identity_setter(result, raw_field, val) continue #if we found the field in the fieldmap, apply the correct function try: mapper(result, field, val) except: #Warning: this is a catchall for _any_ exception, #and may mask what's actually going wrong. if self.Strict: raise FieldError, "Could not handle line %s" % (line,) return result def FieldWrapper(fields, splitter=None, constructor=None): """Returns dict containing field->val mapping, one level. fields should be list of fields, in order. splitter should be something like a DelimitedSplitter that converts the line into a sequence of fields. constructor is a callable applied to the dict after construction. Call result on a _single_ line, not a list of lines. Note that the constructor should take a dict and return an object of some useful type. Additionally, it is the _constructor's_ responsibility to complain if there are not enough fields, since zip will silently truncate at the shorter sequence. This is actually useful in the case where many of the later fields are optional. """ if splitter is None: splitter = DelimitedSplitter(None, None) if constructor: def parser(line): return constructor(dict(zip(fields, splitter(line)))) else: def parser(line): return dict(zip(fields, splitter(line))) return parser def StrictFieldWrapper(fields, splitter=None, constructor=None): """Returns dict containing field->val mapping, one level. fields should be list of fields, in order. splitter should be something like a DelimitedSplitter that converts the line into a sequence of fields. constructor is a callable applied to the dict after construction. Call result on a _single_ line, not a list of lines. Note that the constructor should take a dict and return an object of some useful type. Raises RecordError if the wrong number of fields are returned from the split. """ if splitter is None: splitter = DelimitedSplitter(None, None) if constructor: def parser(line): items = splitter(line) if len(items) != len(fields): raise FieldError, "Expected %s items but got %s: %s" % \ (len(fields), len(items), items) return constructor(dict(zip(fields, items))) else: def parser(line): items = splitter(line) if len(items) != len(fields): raise FieldError, "Expected %s items but got %s: %s" % \ (len(fields), len(items), items) return dict(zip(fields, items)) return parser def raise_unknown_field(field, data): """Raises a FieldError, displaying the offending field and data.""" raise FieldError, "Got unknown field %s with data %s" % (field, data) class FieldMorpher(object): """When called, applies appropriate constructors to each value of dict. Initialize using a dict of fieldname:constructor pairs. """ def __init__(self, Constructors, Default=raise_unknown_field): """Returns a new FieldMorpher, using appropriate constructors. If a field is unknown, will try to set key and value to the results of Default(key, value): in other words, the signature of Default should take a key and a value and should return a key and a value. The built-in value of Default raises a FieldError instead, but it will often be useful to do things like return the key/value pair unchanged, or to strip the key and the value and then add them. """ self.Constructors = Constructors self.Default = Default def __call__(self, data): """Returns a new dict containing information converted from data.""" result = {} default = self.Default cons = self.Constructors for key, val in data.iteritems(): if key in cons: result[key] = cons[key](val) else: new_key, new_val = default(key, val) #if we now recognize the key, use its constructor on the old val if new_key in cons: result[new_key] = cons[new_key](val) #otherwise, enter the new key and the new val else: result[new_key] = new_val return result