Rewrite documentation for floating point matchers

docs/matchers.md

@@ -36,7 +36,17 @@ REQUIRE_THAT( str,
 ```
 
 ## Built in matchers
-Catch currently provides some matchers, they are in the `Catch::Matchers` and `Catch` namespaces.
+Catch2 provides some matchers by default. They can be found in the
+`Catch::Matchers::foo` namespace and are imported into the `Catch`
+namespace as well.
+
+There are two parts to each of the built-in matchers, the matcher
+type itself and a helper function that provides template argument
+deduction when creating templated matchers. As an example, the matcher
+for checking that two instances of `std::vector` are identical is
+`EqualsMatcher<T>`, but the user is expected to use the `Equals`
+helper function instead.
+
 
 ### String matchers
 The string matchers are `StartsWith`, `EndsWith`, `Contains`, `Equals` and `Matches`. The first four match a literal (sub)string against a result, while `Matches` takes and matches an ECMAScript regex. Do note that `Matches` matches the string as a whole, meaning that "abc" will not match against "abcd", but "abc.*" will.
@@ -57,10 +67,30 @@ These are
 
 
 ### Floating point matchers
-The floating point matchers are `WithinULP` and `WithinAbs`. `WithinAbs` accepts floating point numbers that are within a certain margin of target. `WithinULP` performs an [ULP](https://en.wikipedia.org/wiki/Unit_in_the_last_place)-based comparison of two floating point numbers and accepts them if they are less than certain number of ULPs apart.
+Catch2 provides 3 matchers for working with floating point numbers. These
+are `WithinAbsMatcher`, `WithinUlpsMatcher` and `WithinRelMatcher`.
 
-Do note that ULP-based checks only make sense when both compared numbers are of the same type and `WithinULP` will use type of its argument as the target type. This means that `WithinULP(1.f, 1)` will expect to compare `float`s, but `WithinULP(1., 1)` will expect to compare `double`s.
+The `WithinAbsMatcher` matcher accepts floating point numbers that are
+within a certain distance of target. It should be constructed with the
+`WithinAbs(double target, double margin)` helper.
 
+The `WithinUlpsMatcher` matcher accepts floating point numbers that are
+within a certain number of [ULPs](https://en.wikipedia.org/wiki/Unit_in_the_last_place)
+of the target. Because ULP comparisons need to be done differently for
+`float`s and for `double`s, there are two overloads of the helpers for
+this matcher, `WithinULP(float target, int64_t ULPs)`, and
+`WithinULP(double target, int64_t ULPs)`.
+
+The `WithinRelMatcher` matcher accepts floating point numbers that are
+_approximately equal_ with the target number with some specific tolerance.
+In other words, it checks that `|lhs - rhs| <= epsilon * max(|lhs|, |rhs|)`,
+with special casing for `INFINITY` and `NaN`. There are _4_ overloads of
+the helpers for this matcher, `WithinRel(double target, double margin)`,
+`WithinRel(float target, float margin)`, `WithinRel(double target)`, and
+`WithinRel(float target)`. The latter two provide a default epsilon of
+machine epsilon * 100.
+
+> `WithinRel` matcher was introduced in Catch X.Y.Z
 
 ### Generic matchers
 Catch also aims to provide a set of generic matchers. Currently this set