When using the IDL

struct AA;
typedef sequence<AA> AASeq;

struct BB
{
  AASeq aa_;
};
typedef sequence<BB> BBSeq;

struct AA
{
  BBSeq bb_;
};

And compile this to C++ code according to the mapping, with C++23 clang this doesn't compile, the mapped C++ code as far as I can determine should be:

#include <vector>

struct AA;

using AASeq = std::vector<::AA>;

struct BB
{
  ::AASeq aa_{};
};// BB

using BBSeq = std::vector<::BB>;

struct AA
{
  ::BBSeq bb_{};
};// AA

But with clang20 with C++23 enabled this gives

In file included from <source>:1:
In file included from /opt/compiler-explorer/gcc-14.2.0/lib/gcc/x86_64-linux-gnu/14.2.0/../../../../include/c++/14.2.0/vector:66:
/opt/compiler-explorer/gcc-14.2.0/lib/gcc/x86_64-linux-gnu/14.2.0/../../../../include/c++/14.2.0/bits/stl_vector.h:369:35: error: arithmetic on a pointer to an incomplete type 'AA'
  369 |                       _M_impl._M_end_of_storage - _M_impl._M_start);
      |                       ~~~~~~~~~~~~~~~~~~~~~~~~~ ^
/opt/compiler-explorer/gcc-14.2.0/lib/gcc/x86_64-linux-gnu/14.2.0/../../../../include/c++/14.2.0/bits/stl_vector.h:531:7: note: in instantiation of member function 'std::_Vector_base<AA, std::allocator<AA>>::~_Vector_base' requested here
  531 |       vector() = default;
      |       ^
<source>:9:14: note: in defaulted default constructor for 'std::vector<AA>' first required here
    9 |   ::AASeq aa_{};
      |              ^
<source>:3:8: note: forward declaration of 'AA'
    3 | struct AA;
      |        ^
1 error generated.
Compiler returned: 1

In the context of CORBA it is not legal to just compare object references with operator==/!= which are both provided by std:;shared_ptr. The fact that the local client proxy is different doesn't mean the called CORBA servant is different, in IDL to C++11 comparing two interfaces is illegal (see 6.7.2 of IDL to C++11). This specification should make it clear that omg::types::ref_type<T> should not be comparable

CORBA already have 'ami' as a keyword. I it does not need the @ami annotation.

DDS-RPC implements asynchronous methods as part of the API. It does not depend how the IDL interface was defined. So it does not need the @ami annotation or keyword in the interface.

The proposal is to remove the @ami annotation since neither DDS or CORBA need it. We will keep the ami keyword as it is.

IDL 4.0 contained specific grammar rules for where annotations can be applied. Issue IDL41-9 revised these rules and left the spec with the general statement that (7.4.15.4.2): "An annotation may be applied to any IDL constructs or sub-constructs."

This is problematic for the following reasons:

• The phrase "under annotation" is used in multiple places in the spec without it being defined
• Tool implementors using a grammar based on the one in the spec need to make their own enhanced grammar to include <annotation_appl>, which undermines the usefulness of having a grammar in the spec
• Documentation of specific annotations (whether standardized in IDL, standardized in other OMG documents such as XTYPES, or non-standard) has no systemic way of describing where the annotation should be used
• Users have little assurance that their IDL will be understood by all conforming tools, even when using standardized annotations

The spec says "An annotation may be applied to any IDL constructs or sub-constructs. Applying an annotation consists actually in adding the related meta-data to the element under annotation." However, there is no definition of "IDL construct" or "IDL sub-construct". Furthermore, the nonterminal <annotation_appl> does not appear on the right-hand side of any grammar rule. Thus it is not formally specified where <annotation_appl> may appear in an IDL file.

The specification says "[...] a constant expression (<const_expr>) matching (or automatically castable) to, the defined type
of the discriminator." but "automatically castable to" is never defined. This is also the only place in the specification in which the word "cast" appears.

For all c++ code in this chapter the destructor should be override, not virtual

The raise operation should be _raise, that way any derived user exception can have a raise member, similar to _name and _rep_id. Looks an oversight when adding the code from TAOX11, we use _raise always there

The spec clearly specifies that when a union has an explicit default member, then this member should be chosen in its default constructor. When this is not the case, it would pick the first available case that is listed for its first available member.
However, it makes sense that if there is an implicit default member, then it is picked by the default constructor for the union.
Finally, it should be stated that you cannot overrule in any way what member the default constructor of a union will pick by using the @default annotation on any specific member,

There is nothing explained what the expected behavior is when an annotation is applied at multiple levels, for example see the changed example from 8.3.3, what is the range for Foo:bar2, is it min=10,max=20 or min=5,max=10

@range (min=10, max=20)
typedef long MyLong;
...
struct Foo

;

Section 8.3.3
Has the normative example;

@range (min=10, max=20)
typedef long MyLong;

It is unclear to me if the annotation applies to the typedef keyword or to the type in the typedef definition.

The spec is not clear where optional may be applied, I wanted to use it with a typedef, see the test below, but it seems DDS XTypes disallowed it but IDL4 does seem to allow it, it says "This annotation may be used to set optionality on any element that makes sense to be optional.", my example makes sense about it. The spec should be more precise where this annotation (and others) maybe used

module Test {
@optional
typedef long OptLongFoo;
struct Bar

;
};

The spec is silent about the relationship between min/max/range annotations and the default value. When I have an unsigned long and define min=10/max=20, is that legal without specifying a default value annotation? The default of an unsigned long is zero, so what should be generated in the code now as default? Or should the spec say that when min/max/range are used and the default value of the type is not in the range of those default should be specified by the user?

It is unclear whether or how annotations such as "unit" and "range" apply to the values in an array or sequence. Other annotations may have similar ambiguity.

Do they apply to each contained value, or do do they somehow apply to the aggregate value? What about multi-dimensional arrays or nested sequences?

Does the range somehow apply to the sequence length, perhaps as a minimum?

Where do they belong in the array or sequence definitions – before the keyword, before the type name, or somewhere else?

Do the different locations accept different annotations? Do the annotations change meaning depending on location?

Some text for the annotations refers to element, but the introduction in 8.3.1 says "They may be applied on almost all kinds of elements", but what is an element. It looks for example logical that optional/value/min/max/unit/range can also be applied to a typedef, for example "@optional typedef boolean opt_bool1;". In a DDS world these annotations are mostly only used for types, but with CORBA/LwCCM adding them to a typedef could also mean that for example a optional is passed with a method call.

The spec should be more precise whether applying annotations to a typedef is legal.

7.4.15.4.2 on p.101 contains the rule

(225) <annotation_appl> ::= "@" <scoped_name> [ "(" <annotation_appl_params> ")" ]

The nature of the EBNF notation leaves room for interpreting that between the @ and the <scoped_name> there could appear spaces, tabs, or even comments.

On the other hand, 7.4.15.4.1 on p.100 contains:

(220) <annotation_header> ::= "@annotation" <identifier>

Here, it is clear that @annotation shall be treated as a single token without intermittent characters between @ and annotation.

I propose adding a clarification in the explanations at the beginning of p.102 (following rule 227):

Applying an annotation consists in prefixing the element under annotation with:

• The annotation name ( <scoped_name> ) prefixed with the at symbol( @ ), also known as commercial at. There shall be no intermittent spaces, tabs, or comments between @ and <scoped_name>.

The last sentence is my proposed addition.

There are currently a number of numeric annotations presented in the spec. These are annotations that impact the numerical behavior of a type (like its allowed ranges). Right now, the spec doesn't clearly specify where these annotations may be applied and where they may not be applied. It would be wise to extend the section that introduces these annotations with a paragraph that clearly lays out these restrictions.

The spec doesn't describe exactly to which element types @optional should be possible, leaving it to the vendor implementation. The spec should describe to which element type this annotation (but also the others) should be applied, struct members sounds logical, but what about union members, attributes, exception members?

Should the IDL4 grammar specify that @value annotations, if specified for one enum field, MUST be used on all enum fields?
Regardless of whether we make this restriction, the specification should clearly state what happens when one field is annotated, but the next one is not.

For all C++ struct members that do have a name the {} value initialization is missing to make sure they are initialized to their default value on default construction

The argument val of the method op2 of interface A is a basic type, so should be passed as int32_t by value in the example code

All methods having a return value should return by value, that is what the IDL4toCPP language mapping specifies, not as reference

Tthe traits template should be specialized on the <Bitmask> type, not on the <Bitmask>Bits type

The spec should also mandate that a default constructor is there which initialized all members to its default, so `MyStruct()` as constructor should be there

The mapping doesn't show or mention any initialization of the bitfields, which means that they get a random value at runtime, the spec should describe and show that the bitfields are all zero initalized

As struct/bitset support inheritance, at the moment there is a base a IDL::traits<>::base_type should be avilable for template meta programming

The text talks about struct_mapping, but shouldn't the text refer to cpp_mapping and say that the default is STRUCT_WITH_PUBLIC_MEMBERS, the internal struct_mapping with the default is just an internal thing (maybe rename struct_mapping to just value), when I look at IDL4.2, it has in 8.3.1.2 the autoid annotation, there AutoidKind is just not mentioned and the user uses @autoid, and not @AutoidKind

IDL to C++11 defines a set of IDL::traits, but this spec uses the old CORBA::traits, section A has to be updated/cleaned to use IDL::traits

The proposed enum mapping makes it harder to migrate from C+03 to C11 or newer, why not map this to a class with a _flags enum as member (same as bitmask does). The _flags enum contains the enum elements as defined in IDL, this way user code can use AnEnum::zero independent whether C03 or C+11 is used

The example has the following operations for a_long (same issue happens with the others)

void a_long(int32_t value)

int32_t a_long() const

a_long is a basic type, so according to the text it should have:

<MemberType> <MemberName>()
<MemberType> <MemberName>() const;

Both return by value, but the example has one returning by reference.

Also the setter for a_struct doesn't match the text, seems this chapter needs some review and tweaks

Constructed types are types that are created by an IDL specification.

All types are created by an IDL specification, this is not a characterization that distinguishes constructed types.

I propose
     Constructed types are types that have members.

This would provide distinction from other types.

The spec defines that a bitfield can be a maximum of 64bits, but it doesn't say anything about the maximum size of the bitset, can we create a bitset with 1000 fields of each 64bits? Is that legal, or is there some limit to the maximum size of a bitset

The bitmask mapping lacks the operator&, operator|, operator^, operator~, see https://en.cppreference.com/w/cpp/named_req/BitmaskType, a bitmask type in C++ should have these

The text says that the code should use a _flags enum, but the example usees MyBitMaskBits

The text says

An implementation of the !=, &=, and ^= bitwise operators

But that should use operator |= and not !=

The struct MyBitMask as example code has the following buts:

• After `flag1 = 0x01 <<` the 1 is missing
• operator|= has an incorrect first implementation line, should be `_value |= other;`
• operator&= has a wrong argument , should be `std::uint32_t other`
• operator uint32_t() can be const

This section has:

IDL biset types shall be mapped as defined in Clause 7.14.3.2 of this specification.

Notice the typo, biset should be bitset

The spec is not clear when a type annotated with @default/@default_literal will really use these annotations

For example when I have in IDL the following:

@default(5) typedef long mylong;

When I now use in C++

mylong myvalue {};

What would be the value of myvalue, would it be 5 (the default I specified), or will it be 0?

Also when I have in IDL

enum Color

;

And in C++ I have

Color mycolor {};

What is the value of mycolor, RED or GREEN?

RTI support says these annotations are only used when the IDL type is used within a complex type (struct/union/exception), but if that is the case the IDL spec should mention this.

Some other related questions/thinking points

• What when using array/sequences/map of these types?
• what about @range?

As the class ValueFactoryBase has no base class the destructor should be virtual and not override, same error is made in more examples

The definition of value_type says `Defines type - Type to be used as return C++ type.`, but in what context is return handled. For example in the struct class mapping for type T a const and non-const accesor is defined. Shouldn't be value_type defined as `The template parameter T.`, just as A.1.1 does?

extensibility is heavily under specified in IDL4.2. It is listed as a general purpose annotation but it says nothing at the moment for the case appendable/mutable is specified and types are evolved. What are at that moment the rules for assignability/compatibility. Someone who just looks at IDL4 lacks a lot of required information

The specification is unclear of what the user can expect when using mutable and annotations like range, what when the old version of the type has a range of 1-10 and the new one range of 1-5. An old version is send with value 6, what does the receiver get? Very likely more annotations need to be extended in their description related to mutable

The specification just lists any as value for the min/max/default annotations, but that tells nothing about what is allowed as value, are for what about hex/octet values, floats, etc. Maybe make this more precise by referring to the value_expression rules `<const_expr>` as specified in 7.4.1.4.3 constants

Currently the key element type for maps is the same as the value element type, but that looks to be open, this allows for example a union, struct, reference, or other complex type as key type, all of them lack a comparison operator, very likely it is better to define a special map_key_type_spec just as we have for the union as switch_type_spec

The spec doesn't say anything about bitmask bit_value scoping. I think the spec currently allows

bitmask MyBitMask

;

This is problematic when a bitmask maps to an enum in C++. As a bitmask is pretty similar to an enum I do ask the following, in section 7.5.2 it says:

Enumeration value names are introduced into the enclosing scope and then are treated like any other declaration in that scope.

What about a bitvalue, shouldn't this also be done for a bitvalue (the members of a bitmask), that would allow a safe mapping to an enum in C++.

The spec should be clear about where bitmasks are introduced, within the bitmask scope itself, or in the enclosing scope

The spec says for bitfield " list of identifiers (<identifier>*).", shouldn't this be 1 or 0 identifiers, what if the user specifies 4 identifiers, how to separate them, what is the meaning?

In user code I want to use constexpr for IDL fixed values, for example

constexr F pi

;

But in order for that to work a constexpr constructor has to be available, the spec should define that

The spec says for structured types "A set of comparison operators, including at least "equal to" and "not equal to."", but that is not possible when the structure contains interfaces, valuetypes, valueboxes, abstract, all special types don't provide a operator==/!= so they can't be required when a structure contains any of these. This requirement has to be removed. CORBA doesn't provide equal for these special types because a local != doesn't mean that the references are not pointing to the same servant, object identity is not done through local pointers

CORBA::CustomerMarshal should be CORBA::CustomMarshal

If the numeric type is annotated by range:

    @range(min=1, max = 10) long f1;

or min

    @min(1) long f1;

or max

    @max(-1) long f1;

but does not specify a default, what is the default value since it cannot be 0.

the code examples use = and {} to initialize the constants, recommend to make this consistent by always using {}

The spec describes that with CORBA an invalid usage of union can result in a CORBA::BAD_PARAM exception, for bounded string/vector also a CORBA::BAD_PARAM, but when not using CORBA when IDL4CSP11-9 is resolved a different exception is thrown for an union (for vector it seems this still has to be added also). When a user now has appliation code that uses IDL defined data types with and without CORBA their exception code is different. In order to achieve portability of user code I think the spec should define a omg::bad_param alias type which maps to CORBA::BAD_PARAM or the selected std exception. That way user code can be written portable when the user has data types defined by IDL and wants to support use cases with and without CORBA

The spec in 7.4 says that only basic types/enum should be passed by value for an in argument, but that should also include all shared_ptr types, see https://herbsutter.com/2013/06/05/gotw-91-solution-smart-pointer-parameters/ for some information (more can be found on the internet).

I find it not very maintainable to extend in this spec silently the IDL to C++11 language mapping. In the v1.7 revision I think all is already mapped, so I propose to remove D.3 completely from this spec

A few remarks on 7.17.3:

• formatting of code in text is not done
• what is the impact of range when applied to a typedef of a basic type on all argument passing rules when that typedef is passed with a function, stored in an union or exception?
• how should min/max be implemented, shouldn't also omg::types::ranged be used?
• what should be done when min/max but also range are used as annotation?

Spec says:

If the union has a default case, the default constructor shall initialize the discriminator, and the selected
member field following the initialization rules described in Clause 7.2.4.3.1. If it does not, the default
constructor shall initialize the union to the first discriminant value specified in the IDL definition.

But when there is an implicit default member, that should be selected, for example, the example below (

  enum DataType
  {
    dtEmpty,
    dtLong,
    dtShort,
    dtString,
    dtPoint,
    dtTrack,
    dtGlobal
  };

  union Data switch (DataType)
    {
      case dtLong: long longData;
      case dtShort: short shortData;
      case dtString: string stringData;
      case dtPoint: string pointData;
      case dtTrack: string trackData;
      case dtGlobal: string globalData;
      // by default (implicit), empty union
    };

Prior to C++20, the following template cannot be instantiated:

template <typename T, const T thingy>
class MyTemplate {};

MyTemplate<double, 5.0> x;

Not sure where (not necessarily in the noted paragraph) should include a reference to an electronic mapping file as an example. Could also include test sample IDL and corresponding C++ output. This should probably be informational and could also be in an Annex. The key reason is to show the actual transformation as well as a test/verification of the mappings. Perhaps also include versions of tools/compiler used in which the transformations were validated. In addition unit tests could be referenced or included as informative content to the spec.

With C++20 it would be nice when the mapping would specify a module mapping (see https://en.cppreference.com/w/cpp/language/modules).

Keys of maps can be any IDL type. How should the generated C++ code support key comparison of any type?

As a OMG IDL user, I have messages that treat NaN or infinity as a significant value. I would like to specify this as a default for a field.

Currently, the OMG IDL specification says that a floating point value must include an integer part. It does not mention how to set or use special floating point values.

Per IEEE-754, there are a few types of NaN. It would be desirable for OMG DDS IDL to interpret NaN as a quiet NaN, which matches C++ and Python.

In ROS-IDL, it would look like this:
float32 float32_nan NaN
float64 float64_nan nan

In ROS 2, I proposed an implementation that sets NaN. In the IDL generator, it creates the following IDL file, which is not supposedly valid according to the standard:

module rosidl_generator_tests {
module msg {
module NanValueConstant_Constants

;
struct NanValueConstant

;
};
};

Can the specification be amended to support NaN, +inf, and -inf? This is blocking implementation in ROS 2.

Currently only TRUE or FALSE are allowed boolean values, but why not allow true and false (all lower case) as values

The annotation optional impacts the language mapping, for example in C it results in a pointer, with C++11 it could be a IDL::optional<>. The usage of an annotation is very weak in terms of semantics, it is much better to use a new optional keyword as optional heavily impacts the type presented to the programmer. Adding optional will break existing user code, it is not only something that is checked by a concrete middleware.

The

The IDL 4.2 page has a download link for the spec as published as an ISO: https://www.omg.org/spec/IDL/ISO/19516/PDF

All (?? - verification left as an exercise but there's 926 instances) the internal document links look to be screwy.

Search for e.g. "Reference source not found"

An example (from page 102):

Annex A
Consolidated IDL Grammar
This annex gathers all the rules from all the building blocks.
Error! Reference source not found.:
Error! Reference source not found. Error! Reference source not found.
Error! Reference source not found. Error! Reference source not found.
Error! Reference source not found. Error! Reference source not found.
... etc...

The data retrieval of described rules of constants entity during document compilation failed.

With a bitset/bitfield I assume you can use @default to specify a default value, would be nice if a binary constant could be used, as with C++, a binary constant consists of a sequence of ‘0’ and ‘1’ digits, prefixed by ‘0b’ or ‘0B’ for example 0b101010

IDL 4.2 says for bitset the following, but to our idea the IDL spec goes to far, how the data is organized in memory is something IDL can't enforce, probably not all programming languages provide such a direct way to control the memory layout. Middleware could control how a bitset is transferred on the wire, but I think IDL should not enforce the layout

Bit sets are sequences of bits stored optimally and organized in concatenated addressable pieces called bit fields,
themselves stored optimally. "Stored optimally" means that one bit uses just one bit in memory. "Concatenated" means
that each bit field will be placed in memory just after its predecessor within the bit set (no alignment considerations
apply).

I assume that in a map keys may not be unique (this should be added to 7.4.13.4.3.1), in case that is a restriction for a map, I propose to add a multimap which allows keys to be duplicate.

Now that with IDL4 the user can use @default to specify a specific default it would be good to also specify the default for all basic types when no @default is used. That way the user (or tool) can make a decision whether it must use @default in IDL or not

In the footnote the link to http://www.bipm.org/en/measurement-units/ is mentioned, but this seems to be dead, the link should probably be https://www.bipm.org/en/measurement-units/

Add ruby as language, there is a formal OMG language mapping for that, maybe just add all programming languages for which the OMG already has a formal language mapping

For int8 a new signed_type_int is defined, but I think it should just do the following so that anywhere where octet is used now int8 can be used

<octet_type> ::+ "int8",

IDL 4.2 syntax supports inheritance of only interfaces, valuetypes, and structures.

There are multiple scenarios where it would be useful to inherit other types, specifically Unions and Enumerations.

(1) Inheritance of unions

This would be very useful in some specs where unions are automatically generated. For example unions are created interface definitions in DDS-RPC and derived interfaces would like to use unions that "inherit" the unions in the base interfaces.

This would be useful when a "derived" union wants to add some more cases to the base union.

The derived union would have to use the same discriminator type. All it can do is add new discriminator cases that do not conflict with the non-default discriminator in the base union.

It can add a default case if not present in the base union.

Proposed syntax would be:

union DerivedUnion : BaseUnion {
    case 1 :
       Case1Type case1meber;
...
};

(2) Inheritance of enumerations

This is useful to add literals to an existing enumeration without touching the original definition.

The literals on the derived union would be constrained to not conflict the literals in the base class.

Proposed syntax would be:

enum DerivedEnum : BaseEnum {
    AdditionalLiteral1,
    ...
};

For bitmask the spec says "By default, the size of a bit mask is 32", but what unit is 32, I assume the spec wants to say "By default, the size of a bit mask is 32 bits"

All constructed types map to a C+11 class with accessors, but the new bitset mapping is an exception now, it delivers no accessor methods which gives a different API to the programmer. We propose to map to a class with accessors, similar like struct, or fastdds does in their implementation (see https://fast-dds.docs.eprosima.com/en/latest/fastddsgen/dataTypes/dataTypes.html#bitsets). Also the IDL inheritance is gone in C11, propose to use a derived class, so that in the C+11 code there is also an inheritance

There is a general need to document the IDL. However the IDL language does not provide any standard way to do this. As a consequence people may be forced to define custom annotations which would preclude the use across projects and/or development of general tools that are documentation aware.

To remediate this the suggestion is to add some built-in annotations that maybe used to document the IDL. For example the @documentation annotation below:

@annotation documentation {
    string value default "TBD";
};

This annotation could applied to a type declaration or to a member / element, as in:

@documentation("This documents the type Foo")
struct Foo {
    @documentation("documentation for member m1")
    long m1;
    @documentation("documentation for member m2")
    long m2;
 };

@documentation("This documents the type MyEnum")
enum MyEnum {
    @documentation("documentation for literal L1")
    L1,

    @documentation("documentation for literal L2")
    L2
 };

When map/bitset/bitmask are added it looks formal_parameter_type is not extended to allow these new types to be used as templated module arguments.

The spec gives:

const <MemberType>& <MemberName>() const

But for basic types/int this will result in a warning, for basic/enum it should be

<MemberType> <MemberName>() const

The spec says:

A public constant accessor method with the name of the union member that returns a constant reference to
its value:
const <MemberType>& <MemberName>() const;
Accessing an invalid union member may result in an undefined error.

A language mapping should be precise and very clear to my idea, the user should know what happens if an invalid union member is accessed, this could happen in some cases, the spec should specify which exception will be thrown so that the programmer can react on this, saying it is undefined is not enough.

Section 7.4.13.4.3.1 defines IDL Maps but doesn't specify:

• Which types may be keys
• If maps need to de-duplicate equivalent entries
• If maps need to preserve order

The most essential of these is the first. It's currently unrestricted but this causes problems for language mappings where some additional operations (like less-than or equal-to) must be defined for these types. In order to prevent different language mappings adding different (conflicting) restrictions, the IDL spec should define this.

One approach to this would be that each building block should define how its types can (or can't) be used in maps. For example: everything in core data types is allowed; interfaces are not allowed; valuetypes are not allowed; etc.

The section Imports contains

The effects of an import statement are as follows:

• [...]
• Importing a name scope recursively imports all name scopes nested within it.

This looks counter intuitive to me.

Example:

module commontypes {
   module nested1 {
      enum RGBColor { RED, GREEN, BLUE };
   }
   module nested2 {
      enum TrafficLight { READ, YELLOW, GREEN };
   }
};

What happens if another module does "import commontypes"?

If, as the standard says, all name scopes nested within "commontypes" are recursively imported then this would lead to errors:
The enum values RED and GREEN would be directly visible and would be in conflict with each other.

The 'Group of Annotations General Purpose' includes the @autoid annotations which can optionally specify a HASH as in @autoid(HASH).

To be useful this must be augmented with the @hashid("MyString") annotation that allows an ID to be explicitly assigned by computing the hash of a string using the same algorithm that @autoid(HASH) uses.

ptc/19-07-02 section 7.2.4.2.1 defines the mapping of sequences of basic and non basic types in terms of Java interfaces.
However, the classes implementing the interfaces are not defined.
The implementation classes are required for actual programming.
If it is intended that the user shall provide own implementations then this should be expressly stated.

Table 7-13 (Integer types) should have a little bit wider column 1 and the font size "See Building Block Extended Data" should match the other font in this table

Extend the grammar for Constants to allow constants for all Core Data Types that are currently not allowed:

• sequence
• structure
• union
• array

Section 7.4.1.4.4.4.4 Constructed Recursive Types and Forward Declarations (in the Core Data Types BB) has special semantic rules for sequences. The intent of Maps (in the Extended Data Types BB) seems to be that maps should have these same semantic rules.

There are a lot of more possibilities to be used with C++11 compared to C++, there is a specific OMG language mapping for C++11, so we would like to propose to add c++11 as defined value

Currently when we want to declare a type in a deeply nested module we have to start with all higher level modules, for example

module A {
module B {
module C

7.3 Any says:

"The IDL any type shall be mapped to org.omg.type.Any type."

A.1.4 Any says:

"The IDL type any maps to a public class named org.omg.CORBA.Any with the following definition

package org.omg.CORBA;
public class Any {
..."

Suggested solution:

Replace class declaration with:

"package org.omg.CORBA;
public class Any implements org.omg.type.Any {
..."

IDL 4.2 section 7.4.1.4.3 Constants consistency rules on page 32 contains

• An octet constant can be defined using an integer literal or an integer constant expression but values outside the range 0…255 shall be treated as an error.

ptc/20-05-17 section 7.2.4.1.6 maps IDL octet to Java byte.
For mapping IDL const octet values 128 to 255 to Java, it should be permissible to apply promote_integer_width to octet.

Anonymous types should be moved to their own section

Suggest:

//IDL
typedef long myLongArray[2][3];

// C# example

The annotations should be covered in a separate section of the document and NOT be part of a clean, canonical example.

Suggest:
// IDL
enum E

;

// C#
public enum E

;

Also should enums inherit uint?

This should be changed to a "may" or a "could". Shall seems overly prescriptive.

How could a compliant implementation of Equals<T> be distinguished from Equals(Object) from the callers/users perspective?

Since IDL is a specification language why have an IDL interface that cannot define data types it wishes to encapsulate and use as parameters for methods and attribute types? It looks like someone had a programming language in mind. I guess in the end the reason I think basic interfaces are a bad idea is that they can only be identified heuristically. The absence of contained types is a poor indicator of the IDL authors intent regarding whether the interface is basic or not.

Currently the specification says: Note – The default value (TRUE) is significant when the annotation is present (this means that using the compact form @nested will set the element as nested, which is what is expected intuitively). It does not mean that by default (i.e.,when no annotation is present) an element is nested.

The case when no nested is present should be specified, it should be the same as TRUE or FALSE and not left open. We observed that different DDS vendors take different decisions making portability of user IDL which uses partly @nested broken. Some vendors assume default nested TRUE, others FALSE which in practice means when the user wants portable IDL he has to specify it which each IDL type.

Suggestion is to have one section or appendix where possible naming mappings could be covered.

Change the IDL example naming to match the C# community expectations if that makes it easier to specify

Beside basic types no other canonical example is included. I suggest removing the basic types and having one example.

// IDL
struct foo

;

Something like this.

Types such as sbyte and int32 are included in this section and should not be.

These types should be moved to the extended types section.

The section for these types is missing from the document.

For the CORBA specifications as part of the scope chapter there are no hyperlinks, these should be added

This building block section is missing (e.g. sbyte).

It seems every section has this deviation of alternative identifier mappings. I find that it maps the document more difficult to read. The document at large should provide canonical text and examples. These alternatives should be placed in a section that caters to those alternatives with easy links to this section elsewhere in the document.

I could not find a place that discussed what is expected if an IDL compiler encounters an annotation that it can parse but does not understand.

The section on annotations should contain examples of how to use the various standard annotations.

This section needs to be filled out with examples of anonymous IDL types.

The example uses annotation to define the type. The mapping should be clean/pure without annotations. The annotation should be moved to the sections on annotations.
// IDL
enum AnEnum

;

// Java
public enum AnEnum {
X,
Y;
private int value;
private AnEnum(int value)

public int getValue()

public static AnEnum valueOf(int v)

7.17 Standardized Annotations
}

Same as the description

Ensure the correct separation of building block basic types (e.g. sbye)

There needs to be a section on IDL interfaces being mapped to Java 8 interfaces which provide the capability to be a proper container. This removes the need for the old fooPackage id foo were defined in an IDL interface named foo.

Now that IDL is modular with building blocks, IDL authors need a way of guarding IDL files so they are compatible with tools that implement different subsets of the building blocks. See discussion in IDL43-21 for a motivating example.

See C++ feature test macros for a possible implementation pattern.
https://en.cppreference.com/w/cpp/feature_test

Another option is to have an "IDL_VERSION" macro that can be used as a pre-processor condition.

At present, annotation syntax must be placed in the IDL file, usually right before or after the syntax being annotated.

This works very well for annotations that define shared interface constraints. It does not work well for annotations that control internal implementation details, such as the selection of an API option. These internal details should be separated from the shared interface and so do not belong in the shared IDL file.

This proposal is to add a new syntax for annotating IDL structures. The basic idea is to define an annotation that takes two parameters, a "place" in the IDL and an annotation, and has the effect of inserting the annotation at the given place. For example @insert(Foo.x, @some(value)) would be equivalent to placing @some(value) by the definition of Foo.x.

With this mechanism, end users could create a custom IDL that includes the shared IDL and then adds the custom annotations.

7.4.13.4 "Explanations and Semantics" gives a synoptic list of the complements,

Those complements are:

• Additions to structure definition in order to support single inheritance and void content (no members).
• Ability to discriminate a union with other types (wide char and octet).
• An additional template type (maps).
• Additional constructed types (bitsets and bitmasks).

The complement "Integers restricted to holding 8 bits of information" (7.4.13.4.4) should be added there.

Predecessor: CPP1116-3

For mapping the switch of unions, the language mappings choose different fixed names.
It would be desirable to permit names that reflect the user's application domain.
Example:

  enum environment_t { air, water, land };

  @switchname("environment")
  union env_info_t switch (environment_t) {
    case air:
      air_info_t air_info;
    case water:
      [...]
  };

The @switchname annotation would cause the discriminator to be mapped using the given name.

This may replace the name chosen by the language mapping (such as in the Ada mapping), or it may supplement the mapping chosen name (such as in the C++ mapping), depending on the nature of the mapped union.

The provided name may not overlap with an enum value given in a case and it may also not overlap with a branch member name.

Currently the spec says:

The << binary operator indicates that the value of the left operand shall be shifted left the number of bits specified by the right operand, with 0 fill for the vacated bits. The right operand shall be in the range 0 <= right operand < 64.•The >> binary operator indicates that the value of the left operand shall be shifted right the number of bits specified by the right operand, with 0 fill for the vacated bits. The right operand shall be in the range 0 <= right operand < 64

But this should be more restricted, the range should be dependent on the number of bits of the underlying type, when it is for example a 32bit long, the operand should be in the range of 0 <= 32

IDL 4.2 allows the @value on an enum to set a specific value for example

enum Test_E

;

It would be more clear and logical when IDL would allow

enum Test_E

;

The section that describes the default value does not include a discussion
of what to do when a default is not present. In this case a method called _default or
__default should be defined to set the discriminant to a non-switched field value.
This should be fixed before December.

The use of the Omg.Types name should be changed to OMG.Types.

Every scope containing a constant declaration shall contain a public static partial class.

The "public static partial class" is inconsistent with the examples in 7.2.3.1 above.

The use of "partial" is suspect and cannot span assemblies. It may be useful for forward interface declarations though whose definition appears in another IDL file.

At the moment model generated IDL files are used it could happen that in a certain configuration setting an empty IDL module appears in an IDL file or when doing some prototyping the users uses an empty IDL file. Currently seciton 7.4.1.4.2 of the IDL specification requires an IDL module to have at least 1 definition. We propose to more relax this, to allow an empty IDL module.

Rule 3 should be changed to

<module_dcl> ::= "module" <identifier> "

" A module

And in the text

A list of at least one definition (<definition>+) enclosed within braces ({}). Those definitions form the module body.

Could be changed to

A list of zero or more definitions (<definition>*) enclosed within braces ({}). Those definitions form the module body

It is possible to put an annotation on an attribute with multiple declarators, like this:

struct Foo {
  @key long x, y;
};

What is the effect of this? Does the annotation apply to both declarators? In that case, how do you interpret the following:

struct Foo {
  @fieldid(0x01000) long x, y;
};

The spec should say how to handle this case.

Proposal - annotation applies to all declared members. In some cases this causes an error (same ID on 2 members). In other cases, tools could decide to issue warnings.

Structure inheritance is described by rules (45) and (48), page 27. The constraints on using forward-declared structures are given in section 7.4.1.4.4.4.4, page 40. For clarity, it would be better if this section made explicit the prohibited uses of forward declarations. This is done for value types in section 7.4.5.4.2 on page 57, "It is illegal to inherit from from a forward-declared value type not previously defined". Though this is a clarification it has been shown to be needed as this illegal use for forward-declared structures has occurred in external (to OMG) specifications that use IDL.

Programming languages that have the concept of modules/namespaces also include the concept of "using" a namespace type such that one can ommit the explicit reference to the 'used' namespace.

The following is an example for C++:

Assume the declaration:

namespace MyNamespace
{
    class MyClass
    {
    public:
        void my_operation() {}
    };
    void MyFunction1(MyClass) {}
}

Normally these types need to be accessed using their fully qualified names:

MyNamespace::MyClass obj;
obj.my_operation();
MyNamespace::MyFunction1(obj);

This can be cumbersome when having to reference a lot of types in the same namespace, specially for deep nested namespaces.

The C++ using declaration can bring the name space or individual types into scope simplifying the coding as in:

using MyNamespace::MyClass;
MyClass obj;
obj.my_operation();
MyFunction1(obj);

Or alternatively bringing every type in the name space into the scope:

using namespace MyNamespace;
MyClass obj;
obj.my_operation();
MyFunction1(obj);

Java and C# provide similar facilities:
In Java the "import" keyword is used to import packages or types into the scope. The syntax is:

import package.name.ClassName;   // To import a certain class only
import package.name.*;

In C# the "using" keyword is used to import namespaces or types into the scope. The syntax is:

using  MyNamespaceName.MyClassName;   // To import a certain class only
using MyNamespaceName;

This issue request that a similar facility is added to IDL

This was originally reported as https://issues.omg.org/browse/CORBA34-1

In 7.5.1, the 3rd bullet point lists "interface, struct, union, exception." Instead the same language from 7.5.2 ("The following kinds of definitions...") should be in effect.

In section 7.4.4.4.2 it is explained that an annotation may be used in a shortened form in the following conditions:

• There is no member or only one member with a default value.
  • In that it is allowed to apply the annotation by using just the annotation name.
• There is only one member.
  • In that case you may apply the name of the annotation and the value of its member, without explicitly mentioning the member name and the equals sign.

However, the second bullet may break backward compatibility when annotations get extended with additional members in future spec releases. Consider the following example:

@bit_bound(32)
bitmask DataRepresentationMask {
   @position(0) XCDR, 
   @position(1) XML,
   @posiiton(2) XCDR2
}

@annotation data_representation {
    DataRepresentationKind allowed_kinds;
};

Now to annotate an IDL type to support only the XCDR2 annotation, I can apply a shortened annotation like this:

@data_representation(XCDR | XCDR2)
struct Foo {
   long my_long;
};

Now suppose in a future version of the spec, we extend the @data_representation annotation with an additional field to specify the supported endianness, like this:

enum EndiannessKind {
    BIG_ENDIAN,
    LITTLE_ENDIAN,
    NATIVE_PLATFORM
};

@annotation data_representation {
    DataRepresentationKind allowed_kinds;
    EndiannessKind endianness;
};

In this case, the addition of the new member suddenly invalidates existing IDL files that use the shorthand notation.

What I want to propose is to allow the shorthand notation to be used for implicitly referring to the first available member. That is fully compatible with the current rule, but will also still apply when new members are added afterwards.

The example for 'References to a Template Module' contains the line:
alias MyTemplModule2<S1, S2, m> MyTemplModule;

According to rule 193, 'MyTemplModule2' should be a scoped_name that refers to a previously defined Template Module. There is nothing else (successfully) defined as 'MyTemplModule2'. I think this line should be:

alias MyTemplModule<S1, S2, m> MyTemplModule2;

The grammar rule 195 (for struct_def) indicates that a structure may either inherit from a base or may be empty, but not both. The intent based on the text around the rule seems to be to allow structs that are both empty and inheriting.

8.3.3.1 "@default" and 8.3.6.2 "@oneway" specify standardized annotations that a user is not allowed to make use of because grammar rule 225 (in 7.4.15.4.2) states that a <scoped_name> is used to identify the applied annotation. From rule 4, a <scoped_name> is built out of <identifiers> which can't be keywords (see 7.2.1, 7.2.3, 7.2.4).

Section 7.4.11.4.3 on page 85 contains:

(181) <connector_inherit_spec> ::= ":" <scoped_name>

The rule number should be 182.

Page 112 top continues the bullet list started on p.111 bottom:

• May precise on which elements the annotation is valid in that specific context. That list may be only a subset of all the possible ones.

Replace "precise" by a verb, e.g. specify, detail, elaborate, or expound.

• Shall indicate the default behavior [...]. This is because the later values are intended to be the most logical values when the annotation is present.

"later" -> latter

How are bitsets that use inheritance mapped?

Something I noticed while looking at the union mapping for CORBA. Why is it that the IDL4 mapping uses two underscores prepended to the default method (e.g. __default v.s. _default)?

If there is no technical reason for it I think the single underscore is preferred.

The ptc/19-07-02 section 7.2.4.4 proposes a weak mapping for IDL arrays in that they are mapped directly to Java arrays.
This means that the fixed size nature of IDL arrays cannot be enforced at the location of usage (e.g. index checking and handling of variable size operations of java.util.List).

On the other hand, 7.2.4.2.1.1 does define an abstraction for "Sequence of Basic Types", see Table 7.4 where e.g. sequence<boolean> is mapped to the interface BooleanSeq.
A similar mapping could be defined for IDL arrays.

Example:

  // IDL
  enum Color { RED, GREEN, BLUE };

  typedef Color ColorArray[3];

could be mapped to

  public enum Color {
    // [...]
  }

  public class ColorArray implements org.omg.type.Array<Color> {
    public ColorArray() {
      elementData = new Color[3];   // initialize private member
    }
    // [...] most of the operations can be implemented in the superclass
  }

The generic abstract class org.omg.type.Array would provide getter, setter, and other methods. It could also throw an exception on use of java.util.List operations that undermine the fixed size. I can provide a demo implementation of the class if desired.

If it is desired to maintain the possibility of using plain Java arrays then the mapping could be made switchable via an annotation.

Examples given throughout ptc/19-07-02 make indiscriminate use of anonymous types; see

• 7.2.4.2.1.2 struct MyType
• 7.2.4.4 struct S2
• 7.2.4.6 struct MyType
• 7.14.3.1 struct S4

For IDL compatibility with other mappings that do not support anonymous types (see e.g. C++11 mapping version 1.4 stion 6.2), the examples should preferably use typedefs.
Further, it should be mentioned that anonymous types can incur portability problems when used with other language mappings.

Interfaces - Full should define a FooOperations interface like we have done in the IDL to C# Language Mapping.

Also, public abstract void op1(S sIn); should be changed to void op1(S sIn);.

The revised submission contains a number of typos and inconsistencies that should be fixed:

• In Clause 7.2.4.2.1.1, Sequence of Basic Types, IntarfaceName should be InterfaceName.

• In Clause 7.2.4.2.1.2, Sequence of non Basic Types, the text describing when to throw an exception is somewhat repetitive.

• In Clause 7.2.4.3.2 Unions, the requirement for implementing java.io.Serializable is repeated.

In section 7.2.4.2.1.1 on page 13 top of page after end of fixed font Java code:

Where:
• <IntarfaceName> is the interface name indicated in Table 7.4.

<IntarfaceName> should be <InterfaceName>

In Clause 7.2.4.2.1.2, Sequence of non Basic Types, the text describing when to throw an exception is somewhat repetitive.

In Clause 7.2.4.3.2 Unions, the requirement for implementing java.io.Serializable is repeated.

In section 7.2.4.3.2 on page 16 paragraph before the union U1 example:

[...] The first modifier method shall
take no arguments, return void, and setsthe discriminant to the first available default value starting from a 0 index of the
discriminant type.

setsthe should be set the.

In section 8.1.3 table 8.2 column "IDL Construct", repeated typo:
Accesor should be Accessor.
In particular, on page 32:

Structure Member Name in
Accesor/Modifier Methods

Union Member Name in
Accesor/Modifier Methods

and on page 33:

Interface Attribute Name in
Accesor/Modifier Methods

Exception Member Name in
Accesor/Modifier Methods

Bitfield Name in Bitset
Accesor/Modifier Methods

Throughout the document URIs starting with "http://" should start with "https://" instead.

In section 7.4.11.4.3 page 87 and in Annex A page 132, <connector_inherit_spec> should have the rule number 182.

7.2.6.2.1 on page 23 contains:

A character literal is one or more characters enclosed in single quotes, as in:
const char C1 = ’X’;

Please use the ASCII apostrophe:
const char C1 = 'X';

If an implementation wishes to claim support for the "Building Block Extended Data-Types", it must support all of its complements, including the section 7.4.13.4.4 "Integers restricted to holding 8-bits of information".

However, if the targeted programming language does not natively support int8 or uint8 then it is not obvious how the "unsupported" type may be mapped.

For example, the Java language has a native type byte which is signed.
The question then arises how the implementation maps the type uint8.

The implementation could choose to

• Maintain bit size:
  In that case, it would map uint8 to Java byte.
  The question then arises how values exceeding the positive signed maximum (127) are handled.
• Maintain numeric range:
  In that case, it could map uint8 e.g. to Java short (16 bit quantity) or int (32 bit quantity).
  The implementation must then address the interoperability with other languages where uint8 is
  mapped to an 8 bit quantity.
• Use yet a different mapping, for example a holder class (cf. Short, Integer, etc.)

I therefore propose adding the requirement that the implementation shall document its mapping choice:

If the targeted programming language does not natively support int8 or uint8 then an implementation
shall include information about how it maps the types which lack the native language support.

This is not an issue in the strict sense, just a usability observation.
If I copy/paste the word <porttype_def> of rule 174, it comes out as

>‎ yedeepyotrop <

In section 4 "Terms and Definitions" on page 7, the last sentence is:

> [...] Building blocks are described in
> clause 0,

The sentence seems to end on the comma; furthermore, I cannot find a "clause 0".

There is a stray > at the end of the title,

7.4.1.4.1 IDL Specification>

Page 133 last item of "From Building Block Extended Data-Types" :

(1) <unsigned_longlong_int> ::+ “uint64”

The rule number should be 215.

As an aside, the quotation marks used in rules 208 to 215 are fancy Unicode characters while the rest of the grammar uses the regular ASCII code 34 decimal.

The title of 7.4.13.4 says:

7.4.13.4 <unsigned_longlong_int> ::+ “uint64”Explanations and Semantics

This is a formatting error. The {{ <unsigned_longlong_int> ::+ “uint64”}} should appear before the title as one of the rules following:

(214) <unsigned_long_int> ::+ “uint32”

Note that this problem appears only in the generated PDF. The word document is fine.

Section 7.4.1.4.4.4.3 Enumerations says:

An enumeration must contain at least one enumerator and no more than 232.

The 232 is a typo. It was meant to be 2 ³²