fe_sqr tests could be improved

[Coding-Enthusiast]

secp256k1/src/tests.c

Lines 3289 to 3303 in efe85c7

	static void run_sqr(void) {
	secp256k1_fe x, s;
	{
	int i;
	secp256k1_fe_set_int(&x, 1);
	secp256k1_fe_negate(&x, &x, 1);
	for (i = 1; i <= 512; ++i) {
	secp256k1_fe_mul_int(&x, 2);
	secp256k1_fe_normalize(&x);
	secp256k1_fe_sqr(&s, &x);
	}
	}
	}

This method doesn't test anything and after looking through commits it was added in 59447da and got removed in the next commit 21f81a8. I'm not sure what the "problem" mentioned in the first commit is but if it is checking it using the SECP256K1_FE_VERIFY method, it is pointless to call run_sqr without this build configuration.

[real-or-random]

it is pointless to call run_sqr without this build configuration.

Well, there's nothing wrong with it, except that it adds a few milliseconds to the noverify_tests. I claim that adding #ifdefs isn't worth the complexity and the development time. So this is not an issue.

The test itself could perhaps be improved. We could compare with the result of another method, e.g., fe_mul. We can't use fe_mul to compute x*x directly (see docs) but we could, e.g., check that (x+y)*(x-y) = x^2 - y^2 for some random values y != 0.

[real-or-random]

@Coding-Enthusiast

we could, e.g., check that (x+y)*(x-y) = x^2 - y^2 for some random values y != 0.

Do you want to give it a try and work on a PR?

[Coding-Enthusiast]

Sadly my C skills are readonly 😊

[real-or-random]

We all started small... ;)

[real-or-random]

This has been resolved by #1490.