取消

.NET/C# 异常处理:写一个空的 try 块代码,而把重要代码写到 finally 中(Constrained Execution Regions)

不知你是否见过 try { } finally { } 代码中,try 块留空,而只往 finally 中写代码的情况呢?这种写法有其特殊的目的。

本文就来说说这种不一样的写法。


空的 try 块

你可以点开这个链接查看 Exception 类,在里面你可以看到一段异常处理的代码非常奇怪:

1
2
3
4
5
6
7
8
9
10
11
// 代码已经过简化。
internal void RestoreExceptionDispatchInfo(ExceptionDispatchInfo exceptionDispatchInfo)
{
    // 省略代码。
    try{}
    finally
    {
        // 省略代码。
    }
    // 省略代码。
}

神奇之处就在于,其 try 块是空的,重要代码都放在 finally 中。那为什么会这么写呢?

在代码注释中的解释为:

We do this inside a finally clause to ensure ThreadAbort cannot be injected while we have taken the lock. This is to prevent unrelated exception restorations from getting blocked due to TAE.

翻译过来是:

finally 子句中执行此操作以确保在获取锁时无法注入 ThreadAbort。这是为了防止不相关的异常恢复因 TAE 而被阻止。

也就是说,此方法是为了与 Thread.Abort 对抗,防止 Thread.Abort 中断此处代码的执行。 Thread.Abort 的执行交给 CLR 管理,finally 的执行也是交给 CLR 管理。CLR 确保 finally 块执行的时候不会被 Thread.Abort 阻止。

代码在 .NET Core 和 .NET Framework 中的实现完全一样:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
// This is invoked by ExceptionDispatchInfo.Throw to restore the exception stack trace, corresponding to the original throw of the
// exception, just before the exception is "rethrown".
[SecuritySafeCritical]
internal void RestoreExceptionDispatchInfo(System.Runtime.ExceptionServices.ExceptionDispatchInfo exceptionDispatchInfo)
{
    bool fCanProcessException = !(IsImmutableAgileException(this));
    // Restore only for non-preallocated exceptions
    if (fCanProcessException)
    {
        // Take a lock to ensure only one thread can restore the details
        // at a time against this exception object that could have
        // multiple ExceptionDispatchInfo instances associated with it.
        //
        // We do this inside a finally clause to ensure ThreadAbort cannot
        // be injected while we have taken the lock. This is to prevent
        // unrelated exception restorations from getting blocked due to TAE.
        try{}
        finally
        {
            // When restoring back the fields, we again create a copy and set reference to them
            // in the exception object. This will ensure that when this exception is thrown and these
            // fields are modified, then EDI's references remain intact.
            //
            // Since deep copying can throw on OOM, try to get the copies
            // outside the lock.
            object _stackTraceCopy = (exceptionDispatchInfo.BinaryStackTraceArray == null)?null:DeepCopyStackTrace(exceptionDispatchInfo.BinaryStackTraceArray);
            object _dynamicMethodsCopy = (exceptionDispatchInfo.DynamicMethodArray == null)?null:DeepCopyDynamicMethods(exceptionDispatchInfo.DynamicMethodArray);
            
            // Finally, restore the information. 
            //
            // Since EDI can be created at various points during exception dispatch (e.g. at various frames on the stack) for the same exception instance,
            // they can have different data to be restored. Thus, to ensure atomicity of restoration from each EDI, perform the restore under a lock.
            lock(Exception.s_EDILock)
            {
                _watsonBuckets = exceptionDispatchInfo.WatsonBuckets;
                _ipForWatsonBuckets = exceptionDispatchInfo.IPForWatsonBuckets;
                _remoteStackTraceString = exceptionDispatchInfo.RemoteStackTrace;
                SaveStackTracesFromDeepCopy(this, _stackTraceCopy, _dynamicMethodsCopy);
            }
            _stackTraceString = null;

            // Marks the TES state to indicate we have restored foreign exception
            // dispatch information.
            Exception.PrepareForForeignExceptionRaise();
        }
    }
}

你可以在 这里 查看 .NET Framework 版本,在这里 查看 .NET Core 的版本。

受约束的执行区域(Constrained Execution Regions)

这种现象在微软官方文档 可靠性最佳做法 中有介绍。

Doing so instructs the just-in-time compiler to prepare all the code in the finally block before running the try block. This guarantees that the code in the finally block is built and will run in all cases. It is not uncommon in a CER to have an empty try block. Using a CER protects against asynchronous thread aborts and out-of-memory exceptions. See ExecuteCodeWithGuaranteedCleanup for a form of a CER that additionally handles stack overflows for exceedingly deep code.

使用 try-finally 形成一个受约束的执行区域,使得 finally 中的代码被可靠地执行。


参考资料

本文会经常更新,请阅读原文: https://blog.walterlv.com/post/empty-try-block.html ,以避免陈旧错误知识的误导,同时有更好的阅读体验。

知识共享许可协议

本作品采用 知识共享署名-非商业性使用-相同方式共享 4.0 国际许可协议 进行许可。欢迎转载、使用、重新发布,但务必保留文章署名 吕毅 (包含链接: https://blog.walterlv.com ),不得用于商业目的,基于本文修改后的作品务必以相同的许可发布。如有任何疑问,请 与我联系 ([email protected])

登录 GitHub 账号进行评论