When to use []byte or string in Go?

Frequently in writing Go applications, I find myself with the choice to use []byte or string. Apart from the obvious mutability of []byte, how do I decide which one to use?

I have several use cases for examples:

1. A function returns a new []byte. Since the slice capacity is fixed, what reason is there to not return a string?
2. []byte are not printed as nicely as string by default, so I often find myself casting to string for logging purposes. Should it always have been a string?
3. When prepending []byte, a new underlying array is always created. If the data to prepend is constant, why should this not be a string?

My advice would be to use string by default when you're working with text. But use []byte instead if one of the following conditions applies:

• The mutability of a []byte will significantly reduce the number of allocations needed.

• You are dealing with an API that uses []byte, and avoiding a conversion to string will simplify your code.

I've gotten the sense that in Go, more than in any other non-ML style language, the type is used to convey meaning and intended use. So, the best way to figure out which type to use is to ask yourself what the data is.

A string represents text. Just text. The encoding is not something you have to worry about and all operations work on a character by character basis, regardless of that a 'character' actually is.

An array represents either binary data or a specific encoding of that data. []byte means that the data is either just a byte stream or a stream of single byte characters. []int16 represents an integer stream or a stream of two byte characters.

Given that fact that pretty much everything that deals with bytes also has functions to deal with strings and vice versa, I would suggest that instead of asking what you need to do with the data, you ask what that data represents. And then optimize things once you figure out bottlenecks.

EDIT: This post is where I got the rationale for using type conversion to break up the string.

1. One difference is that the returned []byte can be potentially
   reused to hold another/new data (w/o new memory allocation), while
   string cannot. Another one is that, in the gc implementation at
   least, string is a one word smaller entity than []byte. Can be
   used to save some memory when there is a lot of such items live.

2. Casting a []byte to string for logging is not necessary. Typical 'text' verbs, like %s, %q work for string and []byte expressions
   equally. In the other direction the same holds for e.g. %x or % 02x.

3. Depends on why is the concatenation performed and if the result is ever
   to be again combined w/ something/somewhere else afterwards. If that's the case then []byte may perform better.