That's true, but there is no better way to do it, short of resorting to a higher-precision floating-point type.

You forgot for double inprecision, basicly floating-point arithmetic, not a good idea.

Tests with incorrect order of brackets (like ")(") should be added.
Random tests with a random order of expected true/false answers should be added.
Potentially undeclared identifiers: malloc, rand etc.

create(p, 4096); // known issue: limit stack size

What?

'Seems you found your problem. Closing.

Nice one! Looks very clean :)

I think this solution it's better and elegant!

Inspired by Haskell solution. :)

We cannot assume limit number of cubic number in input.

so we have create a test case, which have many cubic number, such as

"153153153...153" (repeat 153 thousand or more times).

It's easy to construct this test case. We could add it.

I will not approve your translation. It needs random tests and sample test cases are missing too. You can check some other katas how random test cases are created there and just follow it.

Where to find more details about test case help function in <criterion/criterion.h>?

I will add more test cases.

Right!

This puzzle is smart!

I believe you can deduce such information from this line in the Kata description:

You can't add the characters "#&'+,-/;.

It's a puzzle after all, right?

OK, understand it.

Could you help declare this information "forbidden of typedef" in details of this kuta?
It's explicit.

You can't use typedef because you can't use extra ; in your code. It's intended as a part of this puzzle :)

Once that is thrown into the mix it's almost impossible to do anti-cheats, so I'd say it's okay. It's also not too much extra effort to not use it ;-)