2019년 5월 17일 금요일

Type Conversion in Rust

Type conversion is not special in Rust. It's just a function that takes ownership of the value and returns the other type. So you can name convert functions anything. However, it's a convention to use as_, to_, and into_ prefixed name or to use from_ prefixed constructor.

From

You can create any function for type conversion. However, if you want to provide generic interfaces, you'd better implement the From trait. For instance, you should implement From<X> for Y when you want the interface that converts the X type value to the Y type value.
The From trait have an associated function named from. You can call this function like From::from(x). You also can call it like Y::from(x) if the compiler cannot infer the type of the destination type.

Into

From have an associated function, it makes you be able to specify the destination type. It's why From has an associated function instead of a method, but on the other hands, you cannot use it as a method, like a.from().
You should implement the Into trait to use a method for type conversion. This trait allows a variable to be converted to another type with the method. You can use into as X::into(x) or Into::into, but no one needs to do it. It's merely a verbose code. Use x.into() as long as the compiler can infer the destination type. Otherwise, use From.

TryFrom

From and Into are traits to provide a conversion function that would not fail. However, some conversions can fail. For instance, converting from i128 to i32 can fail because some values of i128 are not in the range of i32. To do these conversions, you should use TryFrom and TryInto. For instance, Rust uses TryFrom that returns TryFromIntError on the failure for the above example.

TryInto

The relation between TryFrom and TryInto is the same as From and Into. TryFrom has an associated function named try_from and TryInto has a method named try_into.

From Implies Into

from and into should have the same behavior. The different behavior confuses the users; in fact, you cannot implement them differently. If you simultaneously implement From<Y> for X and Into<X> for Y, you would see the below error message.

error[E0119]: conflicting implementations of trait `std::convert::Into<X>` for type `Y`:

It's because From implies Into. It's called a blanket implementation.
Because of this blanket implementation, you'd better implement From instead of Into when you need a type converting method.
As far as I know, there is only one exception that you should implement Into instead of From. If the output type of the conversion function is a generic type that is not declared in the current crate, you cannot implement From. It's the only case(at least as far as I know). The same rule is applied to TryFrom and TryInto.

2019년 4월 14일 일요일

Do not use garbage collection to catch memory leak

Garbage collection is a technique that automatically releases unnecessary memory. It's very famous because many programming languages adopted garbage collection after John McCarthy implemented it in Lisp. However, there are a few people who misunderstand what garbage collection does. If you think garbage collection prevents a memory leak, unfortunately, you are one of them.

Garbage collection cannot prevent a memory leak. There is no way to avoid all memory leaks if you are using Turing-complete language. To understand it you should know what a memory leak is. Wikipedia describes a memory leak as the following:

a type of resource leak that occurs when a computer program incorrectly manages memory allocations in such a way that memory which is no longer needed is not released.

Briefly, a memory leak is a bug that doesn't release a memory that you don't use. So it is first to find the memory which will not be used in order to detect memory leaks. Unfortunately, it is impossible. I'll explain the reason with the code below. When should x be freed?


Generally, x should be released after use(x). However, what if some_function does not end? If some_function never returns because there is an infinite loop, use(x) will never be called. In this case, x has no future access. Thus keeping this memory while running some_function is a memory leak. If you want to make this function have no memory leak, you need to determine when to release x before executing some_function. It's impossible. It's the halting problem. There is no way to static analyze whether some_function runs forever or not.

It's the reason that there is no way to find all memory leaks. So all automatic memory management schemes, including garbage collection, don't guarantee to catch all memory leaks. They try to release memory which applications will not use. They release only memory that can not be accessed apparently, rather than freeing all unnecessary memory. Formally speaking, all automatic memory management schemes use a sound algorithm in the question "Is it safe to free this memory?" even though the algorithm is not complete.

Releasing only memory which is safe to free is the goal of using garbage collection. Your program doesn't have a dangling pointer if you use it instead of managing memory manually. It means you are free from use-after-free bugs or double-free bugs. In conclusion, garbage collection is not for memory efficiency, but memory safety.

If you are not familiar with memory management, you can reduce the number of memory leaks by using garbage collection. You should not choose garbage-collected languages to catch memory leaks. You should not think your program doesn't have a memory leak because you used garbage-collected languages. As I said before, there is no method to catch all memory leaks. You should diagnose the source code manually to find them. You should use automatic memory management schemes, including garbage collection, as a tool to enhance safety.


This article is a translation of the article written in Korean. Please see this link to see the original post.