Go中处理error的模式

2017年5月11日 393点热度 0人点赞 0条评论

点击上方蓝字“Golang开发者”一起玩耍

小编:今天推荐给大家一篇不错的英文文章。文内有部分链接,可以点击页面底部的【原文链接】查看。


One of the main strengths of the Go programming language is its error model. It’s not my favorite—that would be something like this—but it’s still one of the best on the market. Much has been already written about best practices for error handling in Go (Error handling and Go and Don’t just check errors, handle them gracefully are both good resources). In this post I will talk about some error handling patterns that are less known, but nevertheless important for writing reliable and simple Go code.

Always check for errors

Almost everybody agrees that you should always check for errors in Go (tools like errcheck can help you with that). Same holds for releasing resources using defer statement. Situation becomes more interesting when you combine these two—what are we supposed to do with errors returned from functions called inside defer statement? One pattern you see often is this:

resp, err := http.Get(url)if err != nil {
	return err}defer resp.Body.Close()

Even though the Close call could return an error, we can ignore it without sacrificing correctness. That doesn’t mean we can always ignore errors in deferred calls. Consider this function for writing some data to a file (let’s ignore for the moment that ioutil.WriteFile function does almost the same thing):

func WriteFile(filename string, data []byte) error {
  f, err := os.Create(filename)

  if err != nil {
    return err
  }

  defer f.Close()

  _, err = f.Write(data)
  return err}

This looks fine on the surface, but it actually has a serious flaw. Linux man page for close system call says this:

Not checking the return value of close() is a common but nevertheless serious programming error. It is quite possible that errors on a previous write(2) operation are first reported at the final close(). Not checking the return value when closing the file may lead to silent loss of data.

Now that we know we have to check for errors even when calling Close, what’s the best way to do that? Here is one of the possible correct solutions:

func WriteFile(filename string, data []byte) (err error) {
  f, err := os.Create(filename)

  if err != nil {
    return err
  }

  defer func() {
    if cerr := f.Close(); cerr != nil && err == nil {
      err = cerr
    }
  }()

  _, err = f.Write(data)
  return err}

In this version, we are still calling Close in the defer statement (it was not strictly necessary to use the defer statement here because we only have one Write after if—the code would have been simpler if we had just reordered them—but it’s much safer approach in general), but this time we are correctly checking for errors and updating function’s return value if there was an error during the Close call. To do that, we had to modify function’s signature to use a named return (see this great post if you are not familiar with it).

This approach is correct, but the code has become much uglier: we now have five lines of code just for handling situation that is not the primary task of our function, and we also have to keep track of two different errors instead of just one. If you have multiple resources you have to release in this way, it gets even uglier. I usually solve this problem by defining helper function called safeClose:

func safeClose(c io.Closer, err *error) {
  if cerr := c.Close(); cerr != nil && *err == nil {
    *err = cerr
  }}func WriteFile(filename string, data []byte) (err error) {
  f, err := os.Create(filename)

  if err != nil {
    return err
  }

  defer safeClose(f, &err)

  _, err = f.Write(data)
  return err}

The code looks almost the same as in the first example, but this time it’s actually correct. I was thinking about making a public package that exports this function, but decided against it because the function is not good enough abstraction to be made public.

If you are not convinced that all of this is necessary, take a look at the implementation of ioutil.WriteFile from the standard library:

func WriteFile(filename string, data []byte, perm os.FileMode) error {
  f, err := os.OpenFile(filename, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, perm)
  if err != nil {
    return err
  }
  n, err := f.Write(data)
  if err == nil && n < len(data) {
    err = io.ErrShortWrite
  }
  if err1 := f.Close(); err == nil {
    err = err1
  }
  return err}

It doesn’t use defer, but the approach is the same. Even if there were no write errors, that doesn’t mean your writes were successful: you still have to check the return value of the Close call.

Don’t blindly use err != nil

Hopefully, we agree that you should always check for errors. That doesn’t mean you have to place if err != nil { ... } checks everywhere: they can obscure the meaning of your code and make real bugs harder to spot. Consider this (abridged) example from the real-world code that I wrote (original structure had even more fields):

type point struct {
  Longitude     float32
  Latitude      float32
  Distance      int32
  ElevationGain int16
  ElevationLoss int16}

Here we have a structure with bunch of fields with different types. They are encoded in Big Endian format on the server and sent to a client. To decode single point from the input stream, you might initially write code looking something like this:

func parse(r io.Reader) (*point, error) {
  var p point

  if err := binary.Read(r, binary.BigEndian, &p.Longitude); err != nil {
    return nil, err
  }

  if err := binary.Read(r, binary.BigEndian, &p.Latitude); err != nil {
    return nil, err
  }

  if err := binary.Read(r, binary.BigEndian, &p.Distance); err != nil {
    return nil, err
  }

  if err := binary.Read(r, binary.BigEndian, &p.ElevationGain); err != nil {
    return nil, err
  }

  if err := binary.Read(r, binary.BigEndian, &p.ElevationLoss); err != nil {
    return nil, err
  }

  return &p, nil}

This code looks horrible—even looking at it is causing me pain. This is where monadic error handling comes to the rescue: just define a helper structure that holds the actual io.Reader and the last error encountered so far, and a read function that calls underlying Read only if previously there were no errors. You only check for errors once, when all the reads are completed. Here’s the complete example:

type reader struct {
  r   io.Reader
  err error}func (r *reader) read(data interface{}) {
  if r.err == nil {
    r.err = binary.Read(r.r, binary.BigEndian, data)
  }}func parse(input io.Reader) (*point, error) {
  var p point
  r := reader{r: input}

  r.read(&p.Longitude)
  r.read(&p.Latitude)
  r.read(&p.Distance)
  r.read(&p.ElevationGain)
  r.read(&p.ElevationLoss)

  if r.err != nil {
    return nil, r.err
  }

  return &p, nil}

This is much better: main code path is no more obfuscated and you can see at the first glance what the function is doing. The bad thing about this technique is that you can’t generalize it—you have to define a wrapper type for each function. The good thing is that it’s applicable to both io.Reader and io.Writer, and lots of other read/write functions in the standard library.

If you ever find yourself in situation where too much of your code is devoted to error handling, consider using this approach to make your functions shorter and easier to understand.

Add context to your errors

While I was working on a draft for this post, Coda Hale twitted something that is much better introduction to this section than anything I could have written myself:

“Go errors are great because you have to handle them but you should wrap them and stack traces would be nice” i too like checked exceptions

It’s tempting to just pass errors to the callers in the following way:

if err != nil {
  return err}

But you don’t want to see low-level errors in high-level functions: what can the high-level caller do with “invalid syntax” error from from strconv.Atoi call from the depths of the call stack? That’s why Go programmers are encouraged to either handle errors or wrap them with additional details:

if err != nil {
  return fmt.Errorf("something failed: %v", err)}

This is better, and fine for most situations, but I don’t like this approach. The best solution for wrapping errors in year 2017 is to use printf-style functions like animals? Fortunately, Dave Cheney wrote better errors package that helps you add context to your errors in a better way:

if err != nil {
  return errors.Wrap(err, "something failed")}

You can also unwrap errors and add stack traces to them. It completely eliminates the need for errors and fmt packages from the standard library, because it also offers New and Errorf functions.

I really like this package and it’s usually the first dependency I pull for my new projects (I’ll leave the discussion about whether it’s worth depending on another package for something as simple as this for some other time). In my opinion, this package should have been the part of the standard library in the first place.

28280Go中处理error的模式

这个人很懒,什么都没留下

文章评论