Add ability to call user function from panic #256

New issue

Closed

opened 2021-03-14 11:00:01 -05:00 by DusterTheFirst · 5 comments

DusterTheFirst commented

2021-03-14 11:00:01 -05:00

(Migrated from github.com)

I have little expertise with C and all its linker magic/tools, but I was envisioning a feature much like the #[panic_handler] attribute of rust that would be called in place of the provided panic function.

If achieving this with a macro or some linker tomfoolery is not feasible or elegant, it could also be possible to let the user provide a function pointer that will be called first thing the panic function is entered. Something like:

// leaving out the implementation details of how the user would provide their function
// we could just use a simple setter function
void (*panic_handler)(const char *fmt, ...);

void __attribute__((noreturn)) __printflike(1, 0) panic(const char *fmt, ...) {
    if (panic_handler) {
        panic_handler(/* pass it the args  */);
    }

    // usual panic code
}

Panics are, of course, fatal errors, so it makes sense to halt the processor, but a lot of the time, panics are unexpected when they happen (They should get caught or prevented before actual deployment, but things always have a way of going not as planned) and leave the pico in an undefined state. Since the pico does not shut down and stays in the state it was before the panic, it can be useful to reset the pico's outputs or provide other functionality to safely enter the panicked state. I have little expertise with C and all its linker magic/tools, but I was envisioning a feature much like the `#[panic_handler]` attribute of rust that would be called in place of the provided panic function. If achieving this with a macro or some linker tomfoolery is not feasible or elegant, it could also be possible to let the user provide a function pointer that will be called first thing the panic function is entered. Something like: ```c // leaving out the implementation details of how the user would provide their function // we could just use a simple setter function void (*panic_handler)(const char *fmt, ...); void __attribute__((noreturn)) __printflike(1, 0) panic(const char *fmt, ...) { if (panic_handler) { panic_handler(/* pass it the args */); } // usual panic code } ```

kilograham commented

2021-03-14 11:48:50 -05:00

(Migrated from github.com)

yup, we should implement this... it's been on my list for a while, but not quite made it!

👍 1

Lyle-Alloy commented

2021-04-05 21:28:22 -05:00

(Migrated from github.com)

I actually had this same thought, but I was thinking it would be easiest to make the panic handler just be a __weak (or __attribute__((weak))) declaration.

That way the user can implement their own panic function that doesn't even do any of the usual panic code, which would be desirable for some.

I actually had this same thought, but I was thinking it would be easiest to make the panic handler just be a `__weak` (or `__attribute__((weak))`) declaration. That way the user can implement their own panic function that doesn't even do any of the usual panic code, which would be desirable for some.

👍 1

kilograham commented

2021-05-04 16:56:01 -05:00

(Migrated from github.com)

I actually had this same thought, but I was thinking it would be easiest to make the panic handler just be a __weak (or __attribute__((weak))) declaration.

That way the user can implement their own panic function that doesn't even do any of the usual panic code, which would be desirable for some.

Right now we only support gcc but with a mind to the use of clang, can someone confirm whether weak works with arm-none-eabi clang? I seem to recall trying it in the past, and it not working, but i don't quite remember.

> I actually had this same thought, but I was thinking it would be easiest to make the panic handler just be a `__weak` (or `__attribute__((weak))`) declaration. > > That way the user can implement their own panic function that doesn't even do any of the usual panic code, which would be desirable for some. Right now we only _support_ gcc but with a mind to the use of clang, can someone confirm whether weak works with arm-none-eabi clang? I seem to recall trying it in the past, and it not working, but i don't quite remember.

kilograham commented

2021-05-12 20:15:37 -05:00

(Migrated from github.com)