How can I determine which exceptions can be thrown by a given method?
Solution 1:
Following up to my previous answer, I've managed to create a basic exception finder. It utilises a reflection-based ILReader class, available here on Haibo Luo's MSDN blog. (Just add a reference to the project.)
Updates:
- Now handles local variables and the stack.
- Correctly detects exceptions returned from method calls or fields and later thrown.
- Now handles stack pushes/pops fully and appropiately.
Here is the code, in full. You simply want to use the GetAllExceptions(MethodBase) method either as an extension or static method.
using System;
using System.Collections.Generic;
using System.Collections.ObjectModel;
using System.Linq;
using System.Reflection;
using System.Reflection.Emit;
using System.Text;
using ClrTest.Reflection;
public static class ExceptionAnalyser
{
public static ReadOnlyCollection<Type> GetAllExceptions(this MethodBase method)
{
var exceptionTypes = new HashSet<Type>();
var visitedMethods = new HashSet<MethodBase>();
var localVars = new Type[ushort.MaxValue];
var stack = new Stack<Type>();
GetAllExceptions(method, exceptionTypes, visitedMethods, localVars, stack, 0);
return exceptionTypes.ToList().AsReadOnly();
}
public static void GetAllExceptions(MethodBase method, HashSet<Type> exceptionTypes,
HashSet<MethodBase> visitedMethods, Type[] localVars, Stack<Type> stack, int depth)
{
var ilReader = new ILReader(method);
var allInstructions = ilReader.ToArray();
ILInstruction instruction;
for (int i = 0; i < allInstructions.Length; i++)
{
instruction = allInstructions[i];
if (instruction is InlineMethodInstruction)
{
var methodInstruction = (InlineMethodInstruction)instruction;
if (!visitedMethods.Contains(methodInstruction.Method))
{
visitedMethods.Add(methodInstruction.Method);
GetAllExceptions(methodInstruction.Method, exceptionTypes, visitedMethods,
localVars, stack, depth + 1);
}
var curMethod = methodInstruction.Method;
if (curMethod is ConstructorInfo)
stack.Push(((ConstructorInfo)curMethod).DeclaringType);
else if (method is MethodInfo)
stack.Push(((MethodInfo)curMethod).ReturnParameter.ParameterType);
}
else if (instruction is InlineFieldInstruction)
{
var fieldInstruction = (InlineFieldInstruction)instruction;
stack.Push(fieldInstruction.Field.FieldType);
}
else if (instruction is ShortInlineBrTargetInstruction)
{
}
else if (instruction is InlineBrTargetInstruction)
{
}
else
{
switch (instruction.OpCode.Value)
{
// ld*
case 0x06:
stack.Push(localVars[0]);
break;
case 0x07:
stack.Push(localVars[1]);
break;
case 0x08:
stack.Push(localVars[2]);
break;
case 0x09:
stack.Push(localVars[3]);
break;
case 0x11:
{
var index = (ushort)allInstructions[i + 1].OpCode.Value;
stack.Push(localVars[index]);
break;
}
// st*
case 0x0A:
localVars[0] = stack.Pop();
break;
case 0x0B:
localVars[1] = stack.Pop();
break;
case 0x0C:
localVars[2] = stack.Pop();
break;
case 0x0D:
localVars[3] = stack.Pop();
break;
case 0x13:
{
var index = (ushort)allInstructions[i + 1].OpCode.Value;
localVars[index] = stack.Pop();
break;
}
// throw
case 0x7A:
if (stack.Peek() == null)
break;
if (!typeof(Exception).IsAssignableFrom(stack.Peek()))
{
//var ops = allInstructions.Select(f => f.OpCode).ToArray();
//break;
}
exceptionTypes.Add(stack.Pop());
break;
default:
switch (instruction.OpCode.StackBehaviourPop)
{
case StackBehaviour.Pop0:
break;
case StackBehaviour.Pop1:
case StackBehaviour.Popi:
case StackBehaviour.Popref:
case StackBehaviour.Varpop:
stack.Pop();
break;
case StackBehaviour.Pop1_pop1:
case StackBehaviour.Popi_pop1:
case StackBehaviour.Popi_popi:
case StackBehaviour.Popi_popi8:
case StackBehaviour.Popi_popr4:
case StackBehaviour.Popi_popr8:
case StackBehaviour.Popref_pop1:
case StackBehaviour.Popref_popi:
stack.Pop();
stack.Pop();
break;
case StackBehaviour.Popref_popi_pop1:
case StackBehaviour.Popref_popi_popi:
case StackBehaviour.Popref_popi_popi8:
case StackBehaviour.Popref_popi_popr4:
case StackBehaviour.Popref_popi_popr8:
case StackBehaviour.Popref_popi_popref:
stack.Pop();
stack.Pop();
stack.Pop();
break;
}
switch (instruction.OpCode.StackBehaviourPush)
{
case StackBehaviour.Push0:
break;
case StackBehaviour.Push1:
case StackBehaviour.Pushi:
case StackBehaviour.Pushi8:
case StackBehaviour.Pushr4:
case StackBehaviour.Pushr8:
case StackBehaviour.Pushref:
case StackBehaviour.Varpush:
stack.Push(null);
break;
case StackBehaviour.Push1_push1:
stack.Push(null);
stack.Push(null);
break;
}
break;
}
}
}
}
}
To summarise, this algorithm recursively enumerates (depth-first) any methods called within the specified one, by reading the CIL instructions (as well as keeping track of methods already visited). It maintains a single list of collections that can be thrown using a HashSet<T> object, which is returned at the end. It additionally maintains an array of local variables and a stack, in order to keep track of exceptions that aren't thrown immediately after they are created.
Of course, this code isn't infallible in it's current state. There are a few improvements that I need to make for it to be robust, namely:
Detect exceptions that aren't thrown directly using an exception constructor. (i.e. The exception is retrieved from a local variable or a method call.)Support exceptions popped off the stack then later pushed back on.- Add flow-control detection. Try-catch blocks that handle any thrown exception should remove the appropiate exception from the list, unless a
rethrowinstruction is detected.
Apart from that, I believe the code is reasonably complete. It may take a bit more investigation before I figure out exactly how to do the flow-control detection (though I believe I can see how it operates at the IL-level now).
These functions could probably be turned into an entire library if one was to create a full-featured "exception analyser", but hopefully this will at least provide a sound starting point for such a tool, if not already good enough in its current state.
Anyway, hope that helps!
Solution 2:
This should not be extremely hard. You can get list of exceptions created by a method like this:
IEnumerable<TypeReference> GetCreatedExceptions(MethodDefinition method)
{
return method.GetInstructions()
.Where(i => i.OpCode == OpCodes.Newobj)
.Select(i => ((MemberReference) i.Operand).DeclaringType)
.Where(tr => tr.Name.EndsWith("Exception"))
.Distinct();
}
Snippets use Lokad.Quality.dll from the Open Source Lokad Shared Libraries (which uses Mono.Cecil to do the heavy-lifting around code reflection). I actually put this code into one of the test cases in trunk.
Say, we have a class like this:
class ExceptionClass
{
public void Run()
{
InnerCall();
throw new NotSupportedException();
}
void InnerCall()
{
throw new NotImplementedException();
}
}
then in order to get all exceptions from just the Run method:
var codebase = new Codebase("Lokad.Quality.Test.dll");
var type = codebase.Find<ExceptionClass>();
var method = type.GetMethods().First(md => md.Name == "Run");
var exceptions = GetCreatedExceptions(method)
.ToArray();
Assert.AreEqual(1, exceptions.Length);
Assert.AreEqual("NotSupportedException", exceptions[0].Name);
Now all that remains is to walk down the method call stack down to a certain depth. You can get a list of methods referenced by a method like this:
var references = method.GetReferencedMethods();
Now before being able to call GetCreatedExceptions upon any method down the stack we just need to actually look up into the codebase and resolve all MethodReference instances to MethodDefinition instances actually containing the byte code (with some caching to avoid scanning existing branches). That's the most time-consuming part of the code (since Codebase object does not implement any method lookups on top of Cecil), but that should be doable.
Solution 3:
This answer was posted in the other question you reference and I know I have recommended it before in another similar question. You should give Exception Hunter a try. It lists out every single exception that can possibly be thrown. When I ran it for the first time on my code I was quite surprised by the size of this list for even simple functions. There is a 30 day trial for free so there is no reason not to give it a try.
Solution 4:
My methodology for this type of situation is to handle all the exceptions that I want to and then override the UnhandledException event of the application to log any other's that I don't know about. Then if I come up against any that I think I could possibly resolve then I would just update accordingly.
Hope that helps!