tao_idl fails to compile const double variable assignments with math

[eliotlyon]

Version

ACE+TAO 6.5.15 & 7.0.4

Host machine and operating system

Red Hat Enterprise Linux 8.4

Compiler name and version (including patch level)

gcc 8.4.1 20200928 (Red Hat 8.4.1-1)

The $ACE_ROOT/ace/config.h file

config-linux.h

The $ACE_ROOT/include/makeinclude/platform_macros.GNU file

platform_linux.GNU

Contents of $ACE_ROOT/bin/MakeProjectCreator/config/default.features

(doesn't exist. I used release tarballs downloaded from this github)

AREA/CLASS/EXAMPLE AFFECTED:

tao_idl

Synopsis

6.5.15's and 7.0.4's tao_idl do not compile IDL that 6.5.9's tao_idl accepted

Description

I compiled ACE+TAO 7.0.4, installed it to /tmp/ACE-7.0.4, and set up paths appropriately. Then I ran commands as below:

user@box /tmp/test $ cat minimal.idl
module Minimal
{
    const double a = 6.0;
    const double b = 3.0;
    const double div = a / b;
    const double mul = a * b;
    const double add = a + b;
    const double sub = a - b;
};
user@box /tmp/test $ tao_idl minimal.idl
Error - /tmp/ACE-7.0.4/bin/tao_idl: "minimal.idl", line 5: coercion failure (Can not dump this type) to double
Error - /tmp/ACE-7.0.4/bin/tao_idl: "minimal.idl", line 6: coercion failure (Can not dump this type) to double
Error - /tmp/ACE-7.0.4/bin/tao_idl: "minimal.idl", line 7: coercion failure (Can not dump this type) to double
Error - /tmp/ACE-7.0.4/bin/tao_idl: "minimal.idl", line 8: coercion failure (Can not dump this type) to double
tao_idl: minimal.idl: found 4 errors

Commenting out lines 5-8 causes the compilation to succeed.

I compiled ACE 6.5.15 and its tao_idl behaves the same way.

The ACE+TAO 6.5.9 tao_idl accepts this IDL file and generates reasonable code.

[mitza-oci]

Not sure it matters, but change the end of the module to }; (this is different in IDL than the corresponding C++ feature)

[iguessthislldo]

I did try to refactor the code in ast_expression.cpp that evaluates expressions in #840, so I might have accidentally broken something. I will take a look at it.

[eliotlyon]

Not sure it matters, but change the end of the module to }; (this is different in IDL than the corresponding C++ feature)

My mistake. It's present in the actual file but I omitted it from the example originally.

[iguessthislldo]

I started playing around with it and float has the same problem. I also made it so the error message is a tad bit more useful:

Error - tao_idl: "1713/expr.idl", line 4: coercion failure (Can not dump none) to double
Error - tao_idl: "1713/expr.idl", line 5: coercion failure (Can not dump none) to double
Error - tao_idl: "1713/expr.idl", line 6: coercion failure (Can not dump none) to double
Error - tao_idl: "1713/expr.idl", line 7: coercion failure (Can not dump none) to double
Error - tao_idl: "1713/expr.idl", line 13: coercion failure (Can not dump none) to float
Error - tao_idl: "1713/expr.idl", line 14: coercion failure (Can not dump none) to float
Error - tao_idl: "1713/expr.idl", line 15: coercion failure (Can not dump none) to float
Error - tao_idl: "1713/expr.idl", line 16: coercion failure (Can not dump none) to float

[iguessthislldo]

Okay, so it is a result of #840 like I feared, but the existing code is passing something to my code that causes the problem, and I don't know why it's doing it this way. The main issue is that here:

ACE_TAO/TAO/TAO_IDL/ast/ast_expression.cpp

Lines 2338 to 2378 in 0bfb484

	// First, evaluate it, then try to coerce result type.
	// If already evaluated, return the result.
	switch (t)
	{
	case EV_int8:
	tmp = this->eval_internal (EK_int8);
	break;
	case EV_uint8:
	tmp = this->eval_internal (EK_uint8);
	break;
	case EV_short:
	tmp = this->eval_internal (EK_short);
	break;
	case EV_ushort:
	tmp = this->eval_internal (EK_ushort);
	break;
	case EV_long:
	tmp = this->eval_internal (EK_long);
	break;
	case EV_ulong:
	tmp = this->eval_internal (EK_ulong);
	break;
	case EV_longlong:
	tmp = this->eval_internal (EK_longlong);
	break;
	case EV_ulonglong:
	tmp = this->eval_internal (EK_ulonglong);
	break;
	case EV_octet:
	tmp = this->eval_internal (EK_octet);
	break;
	case EV_bool:
	tmp = this->eval_internal (EK_bool);
	break;
	case EV_fixed:
	tmp = this->eval_internal (EK_fixed_point);
	break;
	default:
	tmp = this->eval_internal (EK_const);
	break;
	}

floating point types result in EK_const getting passed to here in the place where the operations are happening:

ACE_TAO/TAO/TAO_IDL/ast/ast_expression.cpp

Lines 1739 to 1740 in 0bfb484

	ExprType const expr_type = eval_kind_to_expr_type (ek);
	if (expr_type == EV_none) return nullptr;

eval_kind_to_expr_type is something I added to simplify the code. It can't handle EK_const though, because it doesn't make it clear what type to actually use. I'm working on trying to modify it so that it passes more reasonable values in.

[mitza-oci]

Does the switch (t) need a case for EV_double?

[jwillemsen]

TAOX11 generates the following, so we don't calculate anything within the IDL compiler

namespace Minimal
{

  // generated from StubHeaderWriter#visit_const
  /// @copydoc const.idl::Minimal::a
  constexpr double a {6.0};

  // generated from StubHeaderWriter#visit_const
  /// @copydoc const.idl::Minimal::b
  constexpr double b {3.0};

  // generated from StubHeaderWriter#visit_const
  /// @copydoc const.idl::Minimal::div
  constexpr double div = (::Minimal::a / ::Minimal::b);

  // generated from StubHeaderWriter#visit_const
  /// @copydoc const.idl::Minimal::mul
  constexpr double mul = (::Minimal::a * ::Minimal::b);

  // generated from StubHeaderWriter#visit_const
  /// @copydoc const.idl::Minimal::add
  constexpr double add = (::Minimal::a + ::Minimal::b);

  // generated from StubHeaderWriter#visit_const
  /// @copydoc const.idl::Minimal::sub
  constexpr double sub = (::Minimal::a - ::Minimal::b);
} // namespace Minimal