fix bad assumption of strncmp return value

[swalk-cavium]

This corrects the issue described in
https://bugs.php.net/bug.php?id=74082
The test case was making an incorrect assumption about the return
values from strncmp. The standard library implementation only has to guarantee <,=,>
to zero.

[cmb69]

Thanks for the fix! However, I still think that this test makes no sense, since it claims to test binary strings, but it tests binary numbers, i.e. strings consisting of 0s and 1s only, so it misses its goal (but wastes CPU cycles).

[swalk-cavium]

@cmb69 - I admit I was more concerned about making the test pass on aarch64 than
validating the test. Removing it would also achieve that result. Do you have a replacement
test to add?

[nikic]

I believe this one can just be dropped without replacement. We have other strncmp tests, and this one looks completely bogus to me.

Independently of that, I think it would be good to normalize the return value of strncmp and friends to -1,0,1 (rather than only documenting that the specific value is not guaranteed).

[cmb69]

+1 on dropping this test.

± on normalizing the result; I don't mind, but apparently others do.

[swalk-cavium]

So, update the PR to remove the test?

The underlying library calls to the string functions have been standardized
for quite a while (C89 and earlier). It's unlikely those will ever change.

[nikic]

So, update the PR to remove the test?

Yup!

[cmb69]

Thanks! Applied as ea1ea33 (sorry, missed to sqash).