Default implemention of Tolerace<> is faulty

[Alpar Juttner]

The default implementation of Tolerance<> is faulty (it answers false to everything). Probably, it should not exist at all except the documentation.

[Alpar Juttner]

Replying to alpar:

The default implementation of Tolerance<> is faulty (it answers false to everything). Probably, it should not exist at all except the documentation.

So, we have two choices here:

1. Enable the non-specialized version only in the documentation. Then using tolerance with a non-specialized type will result in a compile time error.
2. Implement meaningful return values (for exact computation) in generic version. Then tolerance doesn't have to be specialized for integral types (all non-specialized type will be treated as an integral type).

Which one do you prefer?

[Alpar Juttner]

The attached tolerance-int.patch​ implements the second approach. You can apply it on the top of deba's patch at ticket:230 without the --exact hg option.

[Peter Kovacs]

It seems much better than the former code.

I only found a small mistake in the doc. You wrote "otherwise it a specialized version must be ...", here "it" should be removed.

[Alpar Juttner]

Replying to kpeter:

It seems much better than the former code.

I only found a small mistake in the doc. You wrote "otherwise it a specialized version must be ...", here "it" should be removed.

It's fixed and went to both the 1.0 and the main branches, see [7d7d9debb29a].