Fix a couple gotchas in LLSDArray, LLSDParam, llsd_equals().

Nested LLSDArray expressions, e.g.:
LLSD array_of_arrays(LLSDArray(LLSDArray(17)(34))
                              (LLSDArray("x")("y")));
would quietly produce bad results because the outermost LLSDArray was being
constructed with the compiler's implicit LLSDArray(const LLSDArray&) rather
than LLSDArray(const LLSD&) as the reader assumes. Fixed with an explicit copy
constructor to Do The Right Thing.
Generalized LLSDParam<float> specialization into a macro to resolve similar
conversion ambiguities for float, LLUUID, LLDate, LLURI and LLSD::Binary.
Added optional bits= argument to llsd_equals() to permit comparing embedded
Real values using is_approx_equal_fraction() rather than strictly bitwise.
Omitting bits= retains current bitwise-comparison behavior.

2 files changed, 81 insertions, 28 deletions

diff --git a/indra/llcommon/llsdutil.cpp b/indra/llcommon/llsdutil.cpp
index f8f9ece058..803417d368 100644
--- a/indra/llcommon/llsdutil.cpp
+++ b/indra/llcommon/llsdutil.cpp
@@ -41,6 +41,7 @@
 
 #include "llsdserialize.h"
 #include "stringize.h"
+#include "is_approx_equal_fraction.h"
 
 #include <map>
 #include <set>
@@ -571,7 +572,7 @@ std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::str
     return match_types(prototype.type(), TypeVector(), data.type(), pfx);
 }
 
-bool llsd_equals(const LLSD& lhs, const LLSD& rhs)
+bool llsd_equals(const LLSD& lhs, const LLSD& rhs, unsigned bits)
 {
     // We're comparing strict equality of LLSD representation rather than
     // performing any conversions. So if the types aren't equal, the LLSD
@@ -588,6 +589,20 @@ bool llsd_equals(const LLSD& lhs, const LLSD& rhs)
         // Both are TypeUndefined. There's nothing more to know.
         return true;
 
+    case LLSD::TypeReal:
+        // This is where the 'bits' argument comes in handy. If passed
+        // explicitly, it means to use is_approx_equal_fraction() to compare.
+        if (bits >= 0)
+        {
+            return is_approx_equal_fraction(lhs.asReal(), rhs.asReal(), bits);
+        }
+        // Otherwise we compare bit representations, and the usual caveats
+        // about comparing floating-point numbers apply. Omitting 'bits' when
+        // comparing Real values is only useful when we expect identical bit
+        // representation for a given Real value, e.g. for integer-valued
+        // Reals.
+        return (lhs.asReal() == rhs.asReal());
+
 #define COMPARE_SCALAR(type)                                    \
     case LLSD::Type##type:                                      \
         /* LLSD::URI has operator!=() but not operator==() */   \
@@ -596,10 +611,6 @@ bool llsd_equals(const LLSD& lhs, const LLSD& rhs)
 
     COMPARE_SCALAR(Boolean);
     COMPARE_SCALAR(Integer);
-    // The usual caveats about comparing floating-point numbers apply. This is
-    // only useful when we expect identical bit representation for a given
-    // Real value, e.g. for integer-valued Reals.
-    COMPARE_SCALAR(Real);
     COMPARE_SCALAR(String);
     COMPARE_SCALAR(UUID);
     COMPARE_SCALAR(Date);
@@ -617,7 +628,7 @@ bool llsd_equals(const LLSD& lhs, const LLSD& rhs)
         for ( ; lai != laend && rai != raend; ++lai, ++rai)
         {
             // If any one array element is unequal, the arrays are unequal.
-            if (! llsd_equals(*lai, *rai))
+            if (! llsd_equals(*lai, *rai, bits))
                 return false;
         }
         // Here we've reached the end of one or the other array. They're equal
@@ -644,7 +655,7 @@ bool llsd_equals(const LLSD& lhs, const LLSD& rhs)
             if (rhskeys.erase(lmi->first) != 1)
                 return false;
             // Both maps have the current key. Compare values.
-            if (! llsd_equals(lmi->second, rhs[lmi->first]))
+            if (! llsd_equals(lmi->second, rhs[lmi->first], bits))
                 return false;
         }
         // We've now established that all the lhs keys have equal values in
@@ -657,7 +668,7 @@ bool llsd_equals(const LLSD& lhs, const LLSD& rhs)
         // We expect that every possible type() value is specifically handled
         // above. Failing to extend this switch to support a new LLSD type is
         // an error that must be brought to the coder's attention.
-        LL_ERRS("llsd_equals") << "llsd_equals(" << lhs << ", " << rhs << "): "
+        LL_ERRS("llsd_equals") << "llsd_equals(" << lhs << ", " << rhs << ", " << bits << "): "
             "unknown type " << lhs.type() << LL_ENDL;
         return false;               // pacify the compiler
     }
diff --git a/indra/llcommon/llsdutil.h b/indra/llcommon/llsdutil.h
index 58ccc59f5e..c873b17112 100644
--- a/indra/llcommon/llsdutil.h
+++ b/indra/llcommon/llsdutil.h
@@ -123,8 +123,10 @@ LL_COMMON_API BOOL compare_llsd_with_template(
  */
 LL_COMMON_API std::string llsd_matches(const LLSD& prototype, const LLSD& data, const std::string& pfx="");
 
-/// Deep equality
-LL_COMMON_API bool llsd_equals(const LLSD& lhs, const LLSD& rhs);
+/// Deep equality. If you want to compare LLSD::Real values for approximate
+/// equality rather than bitwise equality, pass @a bits as for
+/// is_approx_equal_fraction().
+LL_COMMON_API bool llsd_equals(const LLSD& lhs, const LLSD& rhs, unsigned bits=-1);
 
 // Simple function to copy data out of input & output iterators if
 // there is no need for casting.
@@ -163,6 +165,31 @@ public:
     LLSDArray():
         _data(LLSD::emptyArray())
     {}
+
+    /**
+     * Need an explicit copy constructor. Consider the following:
+     *
+     * @code
+     * LLSD array_of_arrays(LLSDArray(LLSDArray(17)(34))
+     *                               (LLSDArray("x")("y")));
+     * @endcode
+     *
+     * The coder intends to construct [[17, 34], ["x", "y"]].
+     *
+     * With the compiler's implicit copy constructor, s/he gets instead
+     * [17, 34, ["x", "y"]].
+     *
+     * The expression LLSDArray(17)(34) constructs an LLSDArray with those two
+     * values. The reader assumes it should be converted to LLSD, as we always
+     * want with LLSDArray, before passing it to the @em outer LLSDArray
+     * constructor! This copy constructor makes that happen.
+     */
+    LLSDArray(const LLSDArray& inner):
+        _data(LLSD::emptyArray())
+    {
+        _data.append(inner);
+    }
+
     LLSDArray(const LLSD& value):
         _data(LLSD::emptyArray())
     {
@@ -264,6 +291,39 @@ private:
 };
 
 /**
+ * Turns out that several target types could accept an LLSD param using any of
+ * a few different conversions, e.g. LLUUID's constructor can accept LLUUID or
+ * std::string. Therefore, the compiler can't decide which LLSD conversion
+ * operator to choose, even though to us it seems obvious. But that's okay, we
+ * can specialize LLSDParam for such target types, explicitly specifying the
+ * desired conversion -- that's part of what LLSDParam is all about. Turns out
+ * we have to do that enough to make it worthwhile generalizing. Use a macro
+ * because I need to specify one of the asReal, etc., explicit conversion
+ * methods as well as a type. If I'm overlooking a clever way to implement
+ * that using a template instead, feel free to reimplement.
+ */
+#define LLSDParam_for(T, AS)                    \
+template <>                                     \
+class LLSDParam<T>                              \
+{                                               \
+public:                                         \
+    LLSDParam(const LLSD& value):               \
+        _value(value.AS())                      \
+    {}                                          \
+                                                \
+    operator T() const { return _value; }       \
+                                                \
+private:                                        \
+    T _value;                                   \
+}
+
+LLSDParam_for(float,        asReal);
+LLSDParam_for(LLUUID,       asUUID);
+LLSDParam_for(LLDate,       asDate);
+LLSDParam_for(LLURI,        asURI);
+LLSDParam_for(LLSD::Binary, asBinary);
+
+/**
  * LLSDParam<const char*> is an example of the kind of conversion you can
  * support with LLSDParam beyond native LLSD conversions. Normally you can't
  * pass an LLSD object to a function accepting const char* -- but you can
@@ -308,22 +368,4 @@ public:
     }
 };
 
-/**
- * LLSDParam<float> resolves conversion ambiguity. g++ considers F64, S32 and
- * bool equivalent candidates for implicit conversion to float. (/me rolls eyes)
- */
-template <>
-class LLSDParam<float>
-{
-private:
-    float _value;
-
-public:
-    LLSDParam(const LLSD& value):
-        _value(value.asReal())
-    {}
-
-    operator float() const { return _value; }
-};
-
 #endif // LL_LLSDUTIL_H