sys/test_utils/print_stack_usage: work with small stacks

[kaspar030]

Contribution description

#17706 introduced printing of stack metrics. But, it printed them in the exiting thread, causing stack overflows due to printf(), for very small stacks.

This PR fixes the issue two-fold:

1. print_str() is used, if fmt is linked in.
2. the actual stack size is checked, and nothing is printed if the stack is too small

Testing procedure

Issues/PRs references

[fjmolinas]

ACK

[benpicco]

Instead of making the code more cluttered with fmt why not just

USEMODULE += picolibc

?

[kaspar030]

Instead of making the code more cluttered with fmt why not just

USEMODULE += picolibc

?

Because it changes the test.

(And, picolibc printf() still uses way more stack than print_str().)

[maribu]

See inline. I wonder if we should just let it depend on fmt, though.

Somewhat unrelated: I always wanted to start using fmt in the panic handlers (at least for newlib). Because the very moment the panic handler uses printf I get an "ISR stack overflown", but that was caused by the panic handler's use of printf. If that would get upstream, there would be fmt anyway whenever newlib is used.

[kaspar030]

See inline. I wonder if we should just let it depend on fmt, though.

I thought about it, but it does add >200 bytes to ROM is fmt is not used otherwise. Which would in turn kick tests over the limit.

[fjmolinas]

I thought about it, but it does add >200 bytes to ROM is fmt is not used otherwise. Which would in turn kick tests over the limit.

+1 I think since this is a utility it's good to make it just use what is available, and not have it change the footprint of an application outside of the code size added by the function itself.