The spec has the following sentence:

The _d discriminator modifier can only be used to set the discriminant to a value within the same union member.

This is not really clear and should be improved, maybe

The _d discriminator modifier can only be used to set the discriminant to a value that has the same union member as currently selected.

This chapter says:

The Fixed(std::string&) constructor converts a string representation of a fixed-point literal, with an optional leading sign (+ or -) and an optional trailing ‘d’ or ‘D’, into a real fixed-point value.

But there is no constructor that accepts a non-const std::string&, the text should say

The Fixed(const std::string&) constructor converts a string representation of a
fixed-point literal, with an optional leading sign (+ or -) and an optional trailing ‘d’ or ‘D’, into a real fixed-point value.

In the specification an addition was made about the default argument when the union member has multiple legal discriminator values:

Union members associated with the "default" label and those associated with multiple labels may have multiple legal discriminator values. For these members, the modifiers (those that return void) have a second parameter which has the discriminator's type. This parameter has a default argument. For members associated with multiple labels, the value of the default argument is the value of the case label that appears first in IDL. For members associated with the default label, the value of the default argument is an implementation-defined value that selects the member.

There is nothing in the text that describes what happens when a discriminator value is passed which doesn't belong to the member, this is implicitly in the example code further down "u.obj(a, 2)", but this should be explicit in the text also, passing a not allowed discriminator should result in a BAD_PARAM exception

For some use cases it would be more efficient to map a IDL map to a std::unordered_map (see https://en.cppreference.com/w/cpp/container/unordered_map), maybe standardize an annotation which can be used to override the mapping to std::map to the type as specified by the annotation?

The section says:

"For all primitive types, enums, and
reference types, an in argument A of type P, that argument is passed as P. For all other types, an in argument A of type P is
passed as const P&. For an inout and out argument it is passed as P&. If we return a type of P, it is returned as P."

This is the only use of "primitive types" in the spec. "Basic Data Types" should be preferred. Would suggest:

"For all Basic Data Types [link/ref to 6.6 Mapping for Basic Data Types], enums [link/ref to 6.9 Mapping for Enums], and
reference types [link/ref to 6.7.1 Reference Types], an in argument A of type P, that argument is passed as P. For all other types, an in argument A of type P is
passed as const P&. For an inout and out argument it is passed as P&. If we return a type of P, it is returned as P."

using M2 = IDL::bounded_map<std::string, T, 20;

should be

using M2 = IDL::bounded_map<std::string, T, 20>;