Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

"cannot call non-const fn kani::assert in constant functions" #1586

Closed
tedinski opened this issue Aug 25, 2022 · 14 comments · Fixed by #1590, #1858 or #1918
Closed

"cannot call non-const fn kani::assert in constant functions" #1586

tedinski opened this issue Aug 25, 2022 · 14 comments · Fixed by #1590, #1858 or #1918
Assignees
Labels
[C] Bug This is a bug. Something isn't working.

Comments

@tedinski
Copy link
Contributor

Having a poke at building top-100 crates with Kani. Found this one frequently.

Cause is typically an assert! or panic! etc in a constant function. Probably we just need to make kani::assert and friends const?

@tedinski tedinski added the [C] Bug This is a bug. Something isn't working. label Aug 25, 2022
@zhassan-aws
Copy link
Contributor

Do you have an example?

@tedinski
Copy link
Contributor Author

Good point. I believe this was the time crate?

@nano-o
Copy link

nano-o commented Aug 26, 2022

Hello, I'm trying kani and ran into the same issue. The codebase I want to analyze uses panic! in constant functions and this causes cargo kani to fail with "cannot call non-const fn kani::panic in constant functions".
Simple example:

const fn my_const_fn() {
    panic!()
}
#[cfg(kani)]
#[kani::proof]
pub fn check_something() {
    my_const_fn();
}

@zhassan-aws
Copy link
Contributor

Thanks for the example @nano-o! This is due to how Kani defines the panic function:

pub fn panic(message: &'static str) -> ! {

Adding const to the function seems to fix the issue. I'm running the regressions and will submit a PR if everything looks good.

@tedinski
Copy link
Contributor Author

Hmm. This is possibly not yet completely solved. It did fix most of the issues I saw but I ran into this:

error: `std::fmt::Arguments::<'a>::new_v1` is not yet stable as a const fn
   --> /home/ubuntu/.cargo/registry/src/git.luolix.top-1ecc6299db9ec823/time-0.3.14/src/format_description/well_known/iso8601/adt_hack.rs:148:13
    |
148 |             assert!(bytes[idx] == 0, "invalid configuration");
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    |
    = help: add `#![feature(const_fmt_arguments_new)]` to the crate attributes to enable
    = note: this error originates in the macro `format_args` which comes from the expansion of the macro `assert` (in Nightly builds, run with -Z macro-backtrace for more info)

Which slightly confuses me. I think it might be triggerable just by building with the time crate.

@tedinski tedinski reopened this Aug 26, 2022
@tedinski
Copy link
Contributor Author

Reproduction steps:

$ cargo new time-test
     Created binary (application) `time-test` package
$ cd time-test/
$ cargo add time --features formatting
[..SNIP..]
$ cargo kani
   Compiling itoa v1.0.3
   Compiling time v0.3.14
error: `std::fmt::Arguments::<'a>::new_v1` is not yet stable as a const fn
   --> /home/ubuntu/.cargo/registry/src/git.luolix.top-1ecc6299db9ec823/time-0.3.14/src/format_description/well_known/iso8601/adt_hack.rs:148:13
    |
148 |             assert!(bytes[idx] == 0, "invalid configuration");
    |             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    |
    = help: add `#![feature(const_fmt_arguments_new)]` to the crate attributes to enable
    = note: this error originates in the macro `format_args` which comes from the expansion of the macro `assert` (in Nightly builds, run with -Z macro-backtrace for more info)

Error: "Failed to compile crate."
error: could not compile `time` due to previous error; 2 warnings emitted
Error: cargo exited with status exit status: 101

@zhassan-aws
Copy link
Contributor

The non-const format_args was mentioned in an issue related to the const panic RFC: rust-lang/rust#86998. It seems they implemented a special solution for that, which is currently unstable. I'll dig into this.

@zhassan-aws
Copy link
Contributor

Currently, the format_args! cannot be used in const contexts unless the const_fmt_arguments_new language feature is enabled. So, while this Rust program is accepted by rustc:

const fn const_fn() {
    assert!(1 + 1 == 2, "A message");
}

fn main() {
    const_fn();
}
$ rustc assert.rs && ./assert
$

This one isn't:

const fn const_fn() {
    format_args!("A message");
}

fn main() {
    const_fn();
}
$ rustc assert.rs && ./assert 
error: `Arguments::<'a>::new_v1` is not yet stable as a const fn
 --> assert.rs:2:5
  |
2 |     format_args!("A message");
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^
  |
  = help: add `#![feature(const_fmt_arguments_new)]` to the crate attributes to enable
  = note: this error originates in the macro `format_args` (in Nightly builds, run with -Z macro-backtrace for more info)

error: erroneous constant used
 --> assert.rs:2:18
  |
2 |     format_args!("A message");
  |                  ^^^^^^^^^^^ referenced constant has errors
  |
  = note: `#[deny(const_err)]` on by default
  = warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!
  = note: for more information, see issue #71800 <https://github.com/rust-lang/rust/issues/71800>

error: erroneous constant used
 --> assert.rs:2:5
  |
2 |     format_args!("A message");
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^ referenced constant has errors
  |
  = warning: this was previously accepted by the compiler but is being phased out; it will become a hard error in a future release!
  = note: for more information, see issue #71800 <https://github.com/rust-lang/rust/issues/71800>
  = note: this error originates in the macro `format_args` (in Nightly builds, run with -Z macro-backtrace for more info)

error: aborting due to 3 previous errors

A workaround is to add:

![feature(const_fmt_arguments_new)]

to the root of the crate in which the assert is used (which may be difficult/impossible to do for dependent crates).

This issue should be resolved if/when const_fmt_arguments_new is stabilized .

@tedinski
Copy link
Contributor Author

tedinski commented Nov 3, 2022

This got hit again: time-rs/time#521

@camshaft
Copy link
Contributor

camshaft commented Nov 4, 2022

I think we need to redefine the assert macro

macro_rules! assert {
    ($cond:expr $(,)?) => {
        kani::assert($cond, concat!("assertion failed: ", stringify!($cond)));
    };
    ($cond:expr, $fmt:literal $(,)?) => {{
        kani::assert($cond, concat!(stringify!($fmt)));
        if false {
            let _ = $fmt;
        }
    }};
    ($cond:expr, $fmt:literal, $($arg:tt)+) => {{
        kani::assert($cond, concat!(stringify!($fmt, $($arg)+)));
        if false {
            let _ = format_args!($fmt, $($arg)+);
        }
    }};
}

@zhassan-aws
Copy link
Contributor

Thanks @camshaft for the great suggestion! The only downside is that Kani would fail to report errors/warnings in the format literal, e.g.

#[cfg(kani)]
macro_rules! assert {
    ($cond:expr $(,)?) => {
        kani::assert($cond, concat!("assertion failed: ", stringify!($cond)));
    };
    ($cond:expr, $fmt:literal $(,)?) => {{
        kani::assert($cond, concat!(stringify!($fmt)));
        if false {
            let _ = $fmt;
        }
    }};
    ($cond:expr, $fmt:literal, $($arg:tt)+) => {{
        kani::assert($cond, concat!(stringify!($fmt, $($arg)+)));
        if false {
            let _ = format_args!($fmt, $($arg)+);
        }
    }};
}

#[cfg_attr(kani, kani::proof)]
fn main() {
    assert!(1 + 1 == 2, "Hello world {}");
}

rustc:

$ rustc test_const.rs 
warning: panic message contains an unused formatting placeholder
  --> test_const.rs:22:38
   |
22 |     assert!(1 + 1 == 2, "Hello world {}");
   |                                      ^^
   |
   = note: this message is not used as a format string when given without arguments, but will be in Rust 2021
   = note: `#[warn(non_fmt_panics)]` on by default
help: add the missing argument
   |
22 |     assert!(1 + 1 == 2, "Hello world {}", ...);
   |                                         +++++
help: or add a "{}" format string to use the message literally
   |
22 |     assert!(1 + 1 == 2, "{}", "Hello world {}");
   |                         +++++

warning: 1 warning emitted

rustc 2021:

$ rustc --edition 2021 test_const.rs 
error: 1 positional argument in format string, but no arguments were given
  --> test_const.rs:22:38
   |
22 |     assert!(1 + 1 == 2, "Hello world {}");
   |                                      ^^

error: aborting due to previous error

Kani

$ kani test_const.rs 
Checking harness main...
CBMC 5.67.0 (cbmc-5.67.0)
CBMC version 5.67.0 (cbmc-5.67.0) 64-bit x86_64 linux
Reading GOTO program from file test_const.out.for-main
Generating GOTO Program
Adding CPROVER library (x86_64)
Removal of function pointers and virtual functions
Generic Property Instrumentation
Running with 16 object bits, 48 offset bits (user-specified)
Starting Bounded Model Checking
Runtime Symex: 0.000678171s
size of program expression: 45 steps
slicing removed 27 assignments
Generated 2 VCC(s), 1 remaining after simplification
Runtime Postprocess Equation: 1.6311e-05s
Passing problem to propositional reduction
converting SSA
Runtime Convert SSA: 2.8521e-05s
Running propositional reduction
Post-processing
Runtime Post-process: 1.072e-05s
Solving with MiniSAT 2.2.1 with simplifier
0 variables, 0 clauses
SAT checker: instance is SATISFIABLE
Runtime Solver: 3.8802e-05s
Runtime decision procedure: 0.000100895s

RESULTS:
Check 1: main.assertion.1
         - Status: SUCCESS
         - Description: ""Hello world {a}""
         - Location: test_const.rs:22:5 in function main


SUMMARY:
 ** 0 of 1 failed

VERIFICATION:- SUCCESSFUL

Verification Time: 0.015771724s

but perhaps this is OK until format_args is allowed in const contexts?

@camshaft
Copy link
Contributor

camshaft commented Nov 4, 2022

Yeah that's a good point. Could you forward the string on to panic instead?

if false {
    ::std::panic!($($arg)+);
}

@zhassan-aws
Copy link
Contributor

zhassan-aws commented Nov 4, 2022

That's a great idea! This works if I use core::panic (since Kani overrides std::panic). Both core::panic and ::std::panic work. Will submit a PR shortly.

@zhassan-aws
Copy link
Contributor

D'oh, this still doesn't work with cf017e8 for const panics, e.g.:

const fn my_const_fn(msg: &str) -> ! {
    panic!("{}", msg)
}

#[cfg_attr(kani, kani::proof)]
fn main() {
    my_const_fn("failed");
}
$ rustc const_panic.rs 
$ ./const_panic 
thread 'main' panicked at 'failed', const_panic.rs:2:5
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
$ kani const_panic.rs 
error[E0015]: cannot call non-const formatting macro in constant functions
 --> const_panic.rs:2:18
  |
2 |     panic!("{}", msg)
  |                  ^^^
  |
  = note: calls in constant functions are limited to constant functions, tuple structs and tuple variants
  = note: this error originates in the macro `format_args` which comes from the expansion of the macro `panic` (in Nightly builds, run with -Z macro-backtrace for more info)

error: `std::fmt::Arguments::<'a>::new_v1` is not yet stable as a const fn
 --> const_panic.rs:2:5
  |
2 |     panic!("{}", msg)
  |     ^^^^^^^^^^^^^^^^^
  |
  = help: add `#![feature(const_fmt_arguments_new)]` to the crate attributes to enable
  = note: this error originates in the macro `format_args` which comes from the expansion of the macro `panic` (in Nightly builds, run with -Z macro-backtrace for more info)

error[E0080]: erroneous constant used
 --> const_panic.rs:2:12
  |
2 |     panic!("{}", msg)
  |            ^^^^ referenced constant has errors

error: aborting due to 3 previous errors

Some errors have detailed explanations: E0015, E0080.
For more information about an error, try `rustc --explain E0015`.
Error: "Failed to compile crate."
Error: /home/ubuntu/git/kani/target/kani/bin/kani-compiler exited with status exit status: 1

Re-opening.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment