Request: ability to read a line into a passed `&mut String`

[Lokathor]

I'd like the ability to provide a &mut String value to something similar to Interface::read_line and then have it write the line message into there, instead of allocating a new String each time.

[murarth]

I understand that this is the convention applied to line-reading operations on Stdin and BufRead, but I don't think this is a good fit for linefeed.

The primary reason being that the result of Interface::read_line is an enum with three variants, only one of which contains text input. Therefore, a successful call to read_line would not necessarily fill a string buffer with input. The method signature in this case would be misleading.

[Lokathor]

Right, I'm not suggesting that you alter the existing read_line in any way. Sorry if that was unclear.

Instead, I'd like a separate method that can exist in parallel. Something like this

pub fn read_to_string(&self, &mut String) -> Result<ReadToResult> { ... }

pub enum ReadToResult {
    Eof,
    // the data is already in the String, so we just have an empty variant here
    Success,
    Signal(Signal),
}

[murarth]

Whether it's a change to read_line or a new method, the issue remains that it's misleading to users. The buffer is always available after a call to the method, but its contents do not always represent the resulting user input. One could mistakenly ignore the ReadToResult value and treat an empty string buffer as end-of-file. The similarity to Stdin and BufRead methods contributes to this possibility of confusion.

[Lokathor]

Sure, that's a general danger with methods that allow people to provide a buffer, they might ignore the output. However, since we specifically have a warning against that in Rust ("warning: unused must_use"), I think it's reasonable to allow people to use that style API if they want.

Basically, the library should offer a path where the library user allocates the memory and the library just does stuff with the memory given.

[Evrey]

@murarth I'd like to point you at the io::Read trait in Rust, which is able to do what @Lokathor wants: Fill a preëxisting buffer.

• Read::read fills a buffer slice and returns a result. On success, the buffer is mutated. On failure it may or may not be partially filled with garbage.
• Read::read_vectored fills a bunch of I/O buffers that are given mutably by the user.
• Read::read_to_end fills a preëxisting Vec<u8> that is given by the user, analogous to the wish here to fill a given String. Again, a Result is returned and the buffer may contain garbage partial data in the Err case.
• Read::read_to_string does read_to_end, but as UTF-8 text into a borrowed-from-the-user String. Hmm…
• Read::read_exact fills a slice again, but this time with an expected amount of data.

That's all reading functions. There is no function in io::Read that takes nothing and returns a secretly allocated buffer. So I'd argue that…

The primary reason being that the result of Interface::read_line is an enum with three variants, only one of which contains text input. Therefore, a successful call to read_line would not necessarily fill a string buffer with input. The method signature in this case would be misleading.

… is either not misleading at all, or that whoever might be mislead by that is also not a user of I/O in Rust and therefore wouldn't use this crate.

[murarth]

@Evrey: I am quite familiar with the io::Read trait, thank you. I acknowledged the similarity of this request to methods of standard types in my first comment on this issue.

In my most recent comment, I made the point that this similarity is itself a possible source of confusion. Allow me to demonstrate what I mean by that.

This is a valid and correct usage of Stdin::read_line:

let mut buf = String::new();

loop {
    stdin().read_line(&mut buf)?;

    if buf.is_empty() {
        break;
    }

    process_input(&buf);

    buf.clear();
}

If one were to naively substitute Interface::read_line for Stdin::read_line, they would have a program that appeared to function correctly, but would exit prematurely if the user entered an empty prompt or input a signal-inducing character sequence. Requiring the user to inspect the result of Interface::read_line prevents this kind of mistake.

[Lokathor]

For the version where you read into a buffer, you can move the Signal and EOF outputs into the Err side of the result, and then when they happen the user won't think that things are 100% fine. They'll inspect the error and then see what's up.

[murarth]

Using the Err variant to indicate signal or EOF would be inconsitent with the existing Interface::read_line method.

[Lokathor]

So, is there any way for there to be a low-allocation-chance version of reading in a line to an existing String?

(I understand that of course if the input is very long it would realloc the String as it pushes in characters).