Cypris’ lookout

.Net: The limits of using Reflection

Tuesday, June 13, 2006

Reflection is a hugely useful technology: it allows you to get inside objects and get their intimate details, modify their values and even rewrite part of their code.
Today I wanted to build a simple treeview control that would display the hierarchical structure of an arbitrary object by going through its properties and building a tree-like structure. Useful to visualise the internals of an object and see changes at runtime.

Using PropertyInfo you get enough information about a particular property to decide if you want to display it or recurse into it if it’s a complex type that contain other properties.

The only big issue I faced was when dealing with indexers.

The issue with Indexers

Indexers are defined in code like that:
using System; using System.Collections; public class DayPlanner { // We store our indexer values in here Hashtable _meetings = new System.Collections.Hashtable(); // First indexer public string this[DateTime date] { get { return _meetings[date] as string; } set { _meetings[date] = value; } } // Second indexer, overloads the first public string this[string datestr] { get { return this[DateTime.Parse(datestr)] as string; } set { this[DateTime.Parse(datestr)] = value; } } } We’ve defined an overloaded indexer: one takes a DateTime, the other just a string representation of a date.

We could use them like that for instance:
DayPlanner myDay = new DayPlanner(); myDay[DateTime.Parse("2006/02/03")] = "Lunch"; ... string whatNow = myDay["2006/06/26"]; Now, let’s say we want to print all property names and their values for an arbitrary object.
Let’s try on our myDay intance:
Type type = myDay.GetType(); PropertyInfo[] picoll; picoll = type.GetProperties( BindingFlags.Public | BindingFlags.Instance | BindingFlags.Static); foreach (PropertyInfo pi in picoll) { object val; val = pi.GetValue(myDay, null); Debug.WriteLine(pi.Name+"="+val) ); } So, that part is easy: we iterate through all properties of myDay and print their Name and they value.
Problem is, when we reach the indexer property, we get a TargetParameterCountException on line 8.

We need to detect the indexer before attempting to get its value.
To do that, we need to modify our foreach loop to:
foreach (PropertyInfo pi in picoll) { object val; ParameterInfo[] parcoll = pi.GetIndexParameters(); if (parcoll.Length == 0) { val = pi.GetValue(o, null); Debug.WriteLine(pi.Name+"="+val) ); } } Here we attempt to get the indexer parameters for every property. If we’re dealing with an indexer, the parcoll array will contain a list of ParameterInfo related to that particular indexer.

So what really is this parcoll and its ParameterInfos?

A ParameterInfo contain information about the parameter of an indexer property.
The parcoll array will contain one entry for each overloaded indexer.
In our concrete example this is what we get:

• pi.Name will be Items, this is standard for all indexer properties.
• pi.PropertyType is the type of the returned class of our indexer, in our case, a System.String.
• parcoll.Length will be equal to 2 because we have 2 overloaded indexers.
• parcoll[0].Name will be date.
• parcoll[1].Name will be datestr.
• parcoll[0].ParameterType will be System.DateTime.
• parcoll[1].ParameterType System.String.
• parcoll[0].Member and parcoll[1].Member are in fact references to pi itself

That’s a lot of nice information about our indexer, but what if we want to get a value from it? Consider the definition for GetValue:
PropertyInfo.GetValue(object o, object[]{} index). That second parameter is used to get a value from the indexer.

So, how do we use it?

If you know valid keys that you can feed to the indexer, then you just do that (assuming that pi is the PropertyInfo for the indexer).
string meeting; meeting = pi.GetValue(myDay, new string[]{"2006/02/25"}) as string; Or, if you prefer to use a DateTime: string meeting; meeting = pi.GetValue(myDay, new DateTime[]{DateTime.Now}) as string; But what if you don’t know anything about the values contained by the indexer?
Well, unless you use attributes to decorate the indexer or the underlying storage field with specific custom information, you’re basically screwed.

To understand why, just consider how the indexer is implemented:
When you pass it a DateTime or a string as a parameter, it just passes it on to the getter/setter as the value.
What is done with that is up to the implementer.
If you want to maintain a collection of some kind, then you need to use private fields to hold the keys and their values.
.Net doesn’t store the values passed to the indexer or the values returned by it anywhere.

Consider how .Net doesn’t give you any means of iterating through the keys and values of an indexer.
You can’t do a for(;;) or a foreach() loop, you can’t even know how many items are maintained through the indexer by directly questioning it.
So, it certainly is normal that you cannot get anything out of the indexer without knowing exactly how to access it.

The Observer Effect

In many fields of science, the act of measuring something, whether it’s an electronic circuit or a psychological experiment, can actually influence the observed phenomenon.
This is unsurprisingly called the Observer Effect.

In .Net, Reflecting on object can be very much the same.

Consider the use of Reflection when it comes to getting information about an object:
You use metadata describing the object to get more meta data information about its members You can use that meta data to direcly query the state of an object.

If you limit yourself to getting metada, then no harm is done: you’re not querying the object instance itself but the metadata about its class.

If you start querying an actual object instance for its values then you are actively interacting with it.
In most cases, if you’re just getting the value of Fields or Properties defined as follow, then it’s fine:
class GetMe { private string _meeting; public string Meeting { get { return _meeting; } set { _meeting = value; } } } Using reflection, we can get the values of both _meeting and Meeting without any side-effect to the object.

Now consider this:
class GetMeNot { private int _collection = null; public int Collection { get { if ( _collection == null ) { OpenDatabaseConnection(); _collection = FetchAllData(); } return _collection; } } } Now what we’ve got here is a case of lazy loading or lazy initialization.
The consequence is that by querying the Collection property we’re actually creating it, with a potential for a lot of side-effects, like the change of other properties, here maybe related to opening the database connection for instance.

So, when getting values of the properties of an object instance, we can actually completely change its state.
Whether this is a concern or not really depends on what you’re trying to achieve: you may not be able to achieve being a passive observer using Reflection unless you know about the construction of the particular class you’re dealing with.
Using Attributes, you could give hints about which Property not to query and which Field to get instead. In our case, we could add a custom attribute like [ImplementedBy("_collection")] to our Collection property for instance.

If you want to do something generic, then you’re out of luck it and have to deal with that fact.
One possible way to minimize the effect would be to not query Properties that only have a getter and no setter, but that’s not a guarantee that all properties that implement a getter/setter are going to play nice.
Then there is also the -probably rarer- case of the Property that only has a setter.
Then there is also the case of the property getter that has some in-build security and will return different things depending on who’s calling…

All this seems pretty obvious once you understand the whole picture but it’s easy to get so impressed and dazzled by the power of Reflection that you don’t realise its natural limits.

Entry Filed under  :  .Net, Programming