Implement Base.cmp

This operation is useful for implementing `:(==)`, `:(<)`, and `:(<=)`.
The more efficient `cmp` is, the more efficient these comparisons are.

Author: barucden

src/equals.jl

@@ -1,45 +1,85 @@
-# Equality
+_sign(x::Decimal) = x.s ? -1 : 1
 
-# equals() now depends on == instead
-# of the other way round.
-function Base.:(==)(x::Decimal, y::Decimal)
-    # return early on zero
-    x_is_zero = iszero(x)
-    y_is_zero = iszero(y)
-    if x_is_zero || y_is_zero
-        return x_is_zero === y_is_zero
+function Base.cmp(x::Decimal, y::Decimal)
+    if iszero(x) && iszero(y)
+        return 0
+    elseif iszero(x) # && !iszero(y)
+        return -_sign(y)
+    elseif iszero(y) # && !iszero(x)
+        return _sign(x)
     end
 
-    a = normalize(x)
-    b = normalize(y)
-    a.c == b.c && a.q == b.q && a.s == b.s
-end
+    # Neither x nor y is zero here
 
-function Base.:(<)(x::Decimal, y::Decimal)
-    # return early on zero
-    if iszero(x) && iszero(y)
-        return false
+    if x.s != y.s
+        # x and y have different signs, so
+        #  if x < 0, then return -1 (because y is positive)
+        #  if x > 0, then return +1 (because y is negative)
+        return _sign(x)
     end
 
-    # avoid normalization if possible
-    if x.q == y.q
-        return isless(x.s == 0 ? x.c : -x.c, y.s == 0 ? y.c : -y.c)
+    cmp_c = cmp(x.c, y.c)
+    cmp_q = cmp(x.q, y.q)
+
+    # If both x.c and x.q is greater (or equal, or less) than y.c and y.q,
+    # then x is greater (or equal, or less) than y
+    if cmp_c == cmp_q
+        return cmp_c
     end
 
-    diff = y - x
+    # Adjusted exponent of x and y
+    # It is the position of the most significant digit with respect to
+    # the decimal point
+    expx = ndigits(x.c) + x.q - 1
+    expy = ndigits(y.c) + y.q - 1
 
-    farther_from_0 = diff.c > 0 || (iszero(diff.c) && diff.q > 0)
+    # If expx > expy, then abs(x) > abs(y)
+    # If expx < expy, then abs(x) < abs(y)
+    #
+    # Then we need to consider the sign, which is the same for x and y here
+    #
+    # Overall:
+    #  -1  if expx > expy and they are negative
+    #  +1  if expx > expy and they are positive
+    #  -1  if expx < expy and they are positive
+    #  +1  if expx < expy and they are negative
+    if expx != expy
+        s = _sign(x) # same as _sign(y)
+        return ifelse(expx > expy, s, -s)
+    end
 
-    if diff.s == 1
-        return !farther_from_0
+    #  cmp(x, y) = sign(x - y)
+    #            = sign(sign(x) * abs(x) - sign(y) * abs(y))
+    #
+    # We know that x and y have the same sign here:
+    #
+    #  cmp(x, y) = sign(sign(x) * (abs(x) - abs(y)))
+    #            = sign(x) * sign(abs(x) - abs(y))
+    #            = sign(x) * sign(x.c * 10^x.q - y.c * 10^y.q)
+    #
+    # Now, for the latter sign:
+    #
+    #   sign(x.c * 10^x.q - y.c * 10^y.q)
+    # = sign(x.c * 10^(x.q - y.q) - y.c) * 10^y.q
+    # = sign(x.c - y.c * 10^(y.q - x.q)) * 10^x.q
+    #                                      ^^^^^^ positive
+    #
+    # So, we just need to return
+    #
+    #  sign(x) * sign(x.c * 10^(x.q - y.q) - y.c)  if x.q ≥ y.q,
+    #  sign(x) * sign(x.c - y.c * 10^(y.q - x.q))  if x.q < y.q
+    if x.q ≥ y.q
+        q = x.q - y.q
+        return _sign(x) * cmp(x.c * big(10) ^ q, y.c)
     else
-        return farther_from_0
+        q = y.q - x.q
+        return _sign(x) * cmp(x.c, y.c * big(10) ^ q)
     end
 end
 
-function Base.:(<=)(x::Decimal, y::Decimal)
-    return x < y || x == y
-end
+Base.:(==)(x::Decimal, y::Decimal) = iszero(cmp(x, y))
+Base.:(<)(x::Decimal, y::Decimal) = cmp(x, y) < 0
+Base.:(<=)(x::Decimal, y::Decimal) = cmp(x, y) <= 0
 
 # Special case equality with AbstractFloat to allow comparison against Inf/Nan
 # which are not representable in Decimal

test/runtests.jl

@@ -23,4 +23,6 @@ include("test_hash.jl")
 include("test_norm.jl")
 include("test_round.jl")
 
+include("test_compare.jl")
+
 end