Correctly handle float division in Scalar

[tromey]

Summary

Scalar's operator/ has an inverted test for the LHS being zero.

[tromey]

I've also noticed that int->float conversion in Scalar seems incorrect. At least, Scalar.h claims that Scalar follows C promotion rules, but int->float conversion is done using bitwise reinterpretation rather than preserving the value. I have a patch for this, but I don't know whether or not it's correct to change.

[clayborg]

Actually, we should add a test for this to ensure this doesn't regress.

[clayborg]

Seems like a unit test would work well in this case.

[davide]

Yes, this needs a test. Thanks!

[clayborg]

Bonus points for catching any other conversion errors that don't match the current C/C++ specs. The unit tests should be an easy place to test these kinds of things.

[tromey]

Add unit test

[davide]

LGTM, thanks. Do you have commit access or you need somebody to commit this on your behalf?

[tromey]

In D44693#1046925, @davide wrote:

LGTM, thanks. Do you have commit access or you need somebody to commit this on your behalf?

I do not have commit access.

[davide]

Allright, we got this one.

davide@Davidinos-Mac-Pro ~/w/l/l/lldb> git llvm push
Pushing 1 commit:

55f24c19d1c [Core] Correctly handle float division in Scalar.

Sending lldb/trunk/source/Core/Scalar.cpp
Sending lldb/trunk/unittests/Core/ScalarTest.cpp
Transmitting file data ..done
Committing transaction...
Committed revision 328649.
Committed 55f24c19d1c to svn.

source/Core/Scalar.cpp

case Scalar::e_uint256:
       if (b->m_integer != 0) {
         result.m_integer = a->m_integer.udiv(b->m_integer);
         return result;
       }
       break;
     case Scalar::e_float:
     case Scalar::e_double:
     case Scalar::e_long_double:
-      if (b->m_float.isZero()) {
+      if (!b->m_float.isZero()) {
         result.m_float = a->m_float / b->m_float;
         return result;
       }
       break;
     }
   }
   // For division only, the only way it should make it here is if a promotion
   // failed,

unittests/Core/ScalarTest.cpp

TEST(ScalarTest, GetValue) {
   EXPECT_EQ("1234567890123", ScalarGetValue<signed long long>(1234567890123LL));
   EXPECT_EQ("-1234567890123",
             ScalarGetValue<signed long long>(-1234567890123LL));
   EXPECT_EQ("1234567890123",
             ScalarGetValue<unsigned long long>(1234567890123ULL));
   EXPECT_EQ(std::to_string(std::numeric_limits<unsigned long long>::max()),
             ScalarGetValue(std::numeric_limits<unsigned long long>::max()));
 }
+
+TEST(ScalarTest, Division) {
+  Scalar lhs(5.0);
+  Scalar rhs(2.0);
+  Scalar r = lhs / rhs;
+  EXPECT_TRUE(r.IsValid());
+  EXPECT_EQ(r, Scalar(2.5));
+}