摄自:这里

引发和捕获异常不应该作为方法成功执行的一部分常规发生。在开发类库时，必须让客户端代码有机会在执行可能引发异常的操作之前测试错误条件。例如，System.IO.FileStream提供了一个CanRead属性，可以在调用Read方法之前进行检查，以防止引发潜在的异常，如下所示的代码片段:

Dim str As Stream = GetStream() If (str.CanRead) Then '代码来读取流 如果

The decision of whether to check the state of an object prior to invoking a particular method that may raise an exception depends on the expected state of the object. If a FileStream object is created using a file path that should exist and a constructor that should return a file in read mode, checking the CanRead property is not necessary; the inability to read the FileStream would be a violation of the expected behavior of the method calls made, and an exception should be raised. In contrast, if a method is documented as returning a FileStream reference that may or may not be readable, checking the CanRead property before attempting to read data is advisable.

To illustrate the performance impact that using a "run until exception" coding technique can cause, the performance of a cast, which throws an InvalidCastException if the cast fails, is compared to the C# as operator, which returns nulls if a cast fails. The performance of the two techniques is identical for the case where the cast is valid (see Test 8.05), but for the case where the cast is invalid, and using a cast causes an exception, using a cast is 600 times slower than using the as operator (see Test 8.06). The high-performance impact of the exception-throwing technique includes the cost of allocating, throwing, and catching the exception and the cost of subsequent garbage collection of the exception object, which means the instantaneous impact of throwing an exception is not this high. As more exceptions are thrown, frequent garbage collection becomes an issue, so the overall impact of the frequent use of an exception- throwing coding technique will be similar to Test 8.05.

finally块仍然会抛出任何引发的异常。最后要做的就是确保在抛出异常之前运行清理代码。

The try..catch with an empty catch will completely consume any exception and hide the fact that it happened. The reader will be closed, but there's no telling if the correct thing happened. What if your intent was to write i to the file? In this case, you won't make it to that part of the code and myfile.txt will be empty. Do all of the downstream methods handle this properly? When you see the empty file, will you be able to correctly guess that it's empty because an exception was thrown? Better to throw the exception and let it be known that you're doing something wrong.

另一个原因是这样做的try. catch是完全错误的。你这样做的意思是，“无论发生什么，我都能处理好。”StackOverflowException呢，你能在那之后清理吗?OutOfMemoryException呢?一般来说，您应该只处理您期望并知道如何处理的异常。

可能有很多原因，其中之一就是异常执行非常慢。如果这种情况经常发生，很容易缩短执行时间。

虽然下面两个代码块是等效的，但它们并不相等。

try
{
  int i = 1/0; 
}
catch
{
  reader.Close();
  throw;
}

try
{
  int i = 1/0;
}
finally
{
  reader.Close();
}

'finally'是意图揭示代码。您向编译器和其他程序员声明，无论如何都需要运行这段代码。 如果您有多个捕获块，并且有清理代码，那么您最终需要。如果没有最后，您将在每个catch块中重复清理代码。(干燥原理)

最后，积木是特殊的。CLR将finally块中的代码与catch块分开来识别和处理，并且CLR竭尽全力确保finally块始终执行。这不仅仅是编译器的语法糖。

最大的区别是尝试…Catch将吞下异常，隐藏发生错误的事实。Try .finally将运行您的清理代码，然后异常将继续运行，由知道如何处理它的程序来处理。

从可读性的角度来看，它更明确地告诉未来的代码读者“这里的东西很重要，无论发生什么都需要完成。”这很好。

此外，空catch语句对它们来说往往有某种“味道”。它们可能表明开发人员没有仔细考虑可能发生的各种异常以及如何处理它们。