1
\$\begingroup\$

This is a solution for a workshop which gave me some problems. This solution also available via git: https://github.com/skaunov/6991_ws09/. After the code there also couple of theory questions which complimented the workshop, and then also text of workshop exercise duplicated.

A few notes/comments on the solution

I feel like there should be safety arguments where unsafe is used. The only reason of their absence is that I don't actually understand what is going on in libc and FFI, and studying it to add the arguments is to not only learn Rustonomicon (which isn't that bad), but also become C programmer (which isn't on the schedule).

Starter code mentioned in the exercise and useful to understand its scope is represented via https://github.com/skaunov/6991_ws09/commit/80fbd6b4fe71557e3bbaf32516b9b87aa01dc178.

The solution is a bit boilerplaty. I would tackle this with macros, but it was overstretch to me to focus both on FFI part and do this improvement. So this code stays a good object if I will need some macros practice.

use libc::{
    c_char, c_double, c_int, fclose, fgets, fopen, fscanf, EOF, FILE
};
use std::ffi::CString;

struct File {
    stream: *mut FILE,
}

/* At this place _starter_ code contains 
`to_c_string(string: &str) -> Vec<i8>` function, which I 
replaced in spite of `ffi::CString` methods during debugging, 
and didn't want come from this standard solution. */

impl File {
    fn open(_path: &str) -> Option<Self> {
        let path_c = CString::new(_path)
            .expect("it's exercise --- `_path` should be there");
        let mode = CString::new("r").unwrap();
        let result = unsafe { fopen(path_c.as_ptr(), mode.as_ptr()) };
        if result.is_null() {
            None
        } else {
            Some(File { stream: result })
        }
    }

    /// Returns `None` if file couldn't been read (to `String`) OR it
    /// doesn't contain EOL.
    fn read_string(&mut self) -> Option<String> {
        let mut buffer_current = [i8::default(); 512];
        /* looks like it's ok to get just long enough start of the line
        let it be 512 symbols */
        let success = unsafe {
            fgets(
                buffer_current.as_mut_ptr() as *mut c_char,
                // note two bytes for EOL symbols in `buf`
                510,
                self.stream,
            )
        };
        if success.is_null() {
            return None;
        }
        Some(
            buffer_current
                .into_iter()
                .map(|x| x.to_ne_bytes()[0] as u8)
                .skip_while(|x| x == &0)
                .map(|x| x as char)
                .collect(),
        )
    }

    fn read_i64(&mut self) -> Option<i64> {
        let mut result: c_int = Default::default();
        let mode = CString::new("%d").unwrap();
        match unsafe { 
            fscanf(self.stream, mode.as_ptr(), &mut result) 
        } {
            EOF | 0 => None,
            _ => Some(result.into()),
        }
    }

    fn read_f64(&mut self) -> Option<f64> {
        let mut result: c_double = Default::default();
        let mode = CString::new("%lf").unwrap();
        match unsafe { 
            fscanf(self.stream, mode.as_ptr(), &mut result) 
    } {
            EOF | 0 => None,
            _ => Some(result.into()),
        }
    }

    fn read_char(&mut self) -> Option<char> {
        let mut result: c_char = Default::default();
        let mode = CString::new(" %c").unwrap();
        match unsafe { 
            fscanf(self.stream, mode.as_ptr(), &mut result) 
        } {
            EOF | 0 => None,
            _ => Some(result.to_be_bytes()[0].into()),
        }
    }
}

impl Drop for File {
    fn drop(&mut self) {
        println!("Dropping file.");
        let success = unsafe { fclose(self.stream) };
        if success != 0 {
            panic!("...")
        }
    }
}

fn main() {
    let mut file = File::open("data/test_file.txt")
        .expect("Could not open file.");
    let s = file.read_string().unwrap();
    let i = file.read_i64().unwrap();
    let f = file.read_f64().unwrap();
    let c = file.read_char().unwrap();

    println!("{s} {i} {f} {c}");
}

Answers to the theoretical part

Could your type be Copy, Clone, Send or Sync?

Practically after adding all three of possible traits the output of the toy/exercise app doesn't change (including Miri, which for this exercise on my system just complaining that particular C function shouldn't be run on OS "linux"). Ofc it's way too shallow and not indicative, but at least it doesn't immediately contradict theoretical rationale/reasoning which is put below.

(Technically for Send or Sync it's enough to (unsafely) mark the struct with respective trait name.)

Copy

My arguments for Copy and Clone were same, but experiment shows that there's a formal obstacle to be Copy.

Never thought about it, but that makes sense.
"the trait Copy cannot be implemented for this type; the type has a destructor" \

"Copy not allowed on types with destructors"

Clone

This struct/type can be Clone, but must not.

"Can" since pointer at the end of the day is just a number(s), which are even Copy in their Rust essence. It's easy just to derive the trait.

It's really crazy to manage (and de-allocate) this FILE stream pointer while cloning it around.

Send

Seems like the only sane trait here, since it transfer ownership. I guess it should be noticed people generally should watch if anything is still using the File before Sending it, but since everything is unsafe here, this is the way to deal with things.

Sync

It seems as an awful idea to me: like Clone, but on an order of magnitude.

Did you have to think about memory safety at all? Could you modify your code to cause a dangling pointer?

Yes & yes; it's enough to first try to acquire CString from_raw and then read its usage limitation documentation.

How easy was inter-operating with C functions?

Near to a nightmare --- it's the first time I touched C since university, and I did avoid to delve it back then already. It's to personal, so better ask a person who was exposed to C previously.

'''

The exercise

Here's copy of relevant text from the workshop (to archive it and for convenience).

(Note that https://github.com/skaunov/6991_ws09/blob/master/data/test_file.txt for reading is provided as part of the downloaded starter code.)

Exercise:

Workshop 9 - Safety Shmafety ...

This week, we'll be looking at some unsafe code. Before the workshop, we'll have a quick refresher on the following topics:

  • Safe abstractions over unsafe code
  • The unsafe keyword
  • *const T / *mut T

...

Week 9 Code

In groups, your task for this week is to build a File struct, using primitives from the libc crate. These use functions which are common in the C language, but which Rust does not use.

You should implement the following on the starter code:

  1. Opening a file to read with the read function.
  2. Reading a string a file with read.
  3. Reading a type from a file with read_i64.
  4. Reading a type from a file with read_f64.
  5. Reading a type from a file with read_char.
  6. When your file goes out of scope, close it automatically.

To do this, you will need to know about the following functions/enums/traits, mostly from libc:

  • The FILE enum represents an open file.
  • fopen opens a file (called filename). To open in "read" mode, you should use "r" as the mode. This returns a (possibly null) file-pointer, which represents an open file.
  • fgets reads a whole line from a file. buf is a pointer to memory, which must be at least n bytes big. stream is a file-pointer. Note that fgets returns null if opening the file failed.
  • fscanf reads a different type from the file stream, depending on format: * If the string is "%d", the third argument to fscanf should be a &mut libc::c_int. * If the string is " %c" (note the leading space), the third argument to fscanf should be a &mut libc::c_char. * If the string is "%lf", the third argument to fscanf should be a &mut libc::c_double.
  • fclose should close the file pointer given to it. The Drop trait may be useful here.
  • There are a variety of types (c_int, c_char, c_double) which correspond to types from C. You will need these.

You should assume this code will only be run on Unix-like systems, and that all paths will be ASCII.

At the end of the activity, if there's time, look into the errno crate. This allows you to give the user more information about why an operation failed.

Afterwards, you might like to discuss the following points:

  • Could your type be Copy, Clone, Send or Sync?
  • Did you have to think about memory safety at all? Could you modify your code to cause a dangling pointer?
  • How easy was interoperating with C functions?

PS AFAIU the only thing that shouldn't be publicly shared are solutions to the graded exercises/activities, which I keep access restricted. If I got anything wrong and this code shouldn't be public as well, pls approach me by any mean you like, and I'll remove it without hesitations. Tom received couple of my contacts due to distant interactions (macrokata, https://discord.com/channels/1075940806004838470/1075940806004838475/1101303841531646092).

\$\endgroup\$

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.