Referring to class names through strings?

问题:

I need to parse some text file, create objects for various entities encountered in the text, and put them in some data structure (e.g., a list) for further processing. Example of the text:

laptop
 17" dell, weight: 12 lb
desktop
 24" hp

I know in advance which entities may exist in the text, and what attributes they are supposed to have. In this example, I would already have classes laptop and desktop defined (probably subclasses of class computer). The parser would just need to create objects laptop('dell', 17, 12), and dekstop('hp', 24).

If I follow this route, I would need to retrieve class names from strings, and create objects of those classes. Is it the Pythonic way of doing things? If so, what's the best approach (using Python 3.1)? If not, what should I do instead?

Thanks!

What

回答1:

If the classes are defined in computers.py, say, you can do

import computers
getattr( computers, "Laptop" )( <params> )

to instantiate a computers.Laptop. If they are defined in the same file that you are running the code in (so that they are global variables), you can do

globals()[ "Laptop" ]

but this is less elegant; it would be nicer to put them in a separate scope.

Alternatively, if you want a more powerful mapping (say you want "Nettop", "Lapbook", and "Laptop" all to instantiate Laptop), you could maintain a mapping of strings to their corresponding constructor and use that:

mapping = { "Laptop": Laptop, "Nettop": Laptop, ... }
mapping[ "Laptop" ]()

回答2:

you can create instances of a class by it's name using the following code

obj = globals()[classname]()

where classname is a string

回答3:

Technically, what you're asking for (or at least the way everyone is interpreting it) just isn't very good practice, especially if you might be taking input from an untrusted source (remember, any source other than yourself should generally be considered untrusted!). You should whitelist these things explicitly, because someone might trigger the execution of a function or creation of an object you didn't intend with properties you really don't want...

What you can do instead is something like this (this is of course wildly incomplete, but should give you the general idea):

class Laptop(object):
    pass
class Desktop(object):
    pass

possible_classes = {
    "laptop": Laptop,
    "desktop": Desktop,
}

new_object = possible_classes[identifier_string](propA, propB, propC, ...)

Then just add the mapping for each new kind of object to the possible_classes dict.

回答4:

I think a inspection-based method would potentially be quite fragile and resistant to change. What if you want to use classes from other modules?

Why not a object factory? It could be a simple function or a class.

Example:

class ComputerFactory:

   def __init__(self):
      self._classes = {}

   def register(moniker, creator):
      """Moniker is a name for the class.
      Creator is a callable that creates the object 
      for the moniker.
      """
      self._classes[moniker] = creator

   def create(moniker, *args, **kwargs):
      return self._classes[moniker](*args, **kwargs)

# Example usage
fac = ComputerFactory()
# Register constructor
fac.register("laptop", Laptop) # Laptop class is also a callable (the constructor)
# Register construction proxy
def sony_laptop_builder(make):
   return Laptop("Sony")
fac.register("laptop", sony_laptop_builder)

回答5:

Not sure about the syntax of python code but is this what you want?

for item in parsedString:
    if item == "laptop":
        laptops.add(laptop()) #laptops is a list of laptops you 
                              #have encountered so far in your list
    # add the next elements to your laptop instance

    if item == "desktop":
    # code for desktop...