Relax current limitations for UDA syntax parsing.

[9rnsr]

Issue 11677 - user defined attributes must be first
Issue 11678 - user defined attributes cannot appear as postfixes
Issue 11679 - user defined attributes not allowed for local auto declarations
Issue 11680 - user defined attributes for type inference

Additionally, I added test cases for ddoc unittest feature to avoid silent breaking.

[WalterBright]

Being an enhancement, I'm going to leave this for 2.066!

[IgorStepanov]

BTW, why UDAs don't allowed for module declaration?

@reflect!basic module test; //determine the reflection strategy for the "test" module
import my.reflection: reflect, basic;
...

[9rnsr]

@WalterBright Thanks.

@IgorStepanov Because it is not allowed in grammar. Instead you can use @reflect!basic: immediately after the module declaration.

[IgorStepanov]

Because it is not allowed in grammar.

May be we should allow it? Does fundamental objections aganist it exists?

Instead you can use @reflect!basic: immediately after the module declaration.

No, I need to annotate module, not all module symbols. In my example, I collect data for all symbols to use it in runtime. But I start collect only if scoped symbol annotated with special attribute.
E.g.

@reflect!basic class MyClass //Reflection is on. Information about MyClass members will be saved in special structure
{
   ...
}

@reflect!none class MyClass2 //Reflection is off. Reflection data structure for MyClass2 will be created but will not store any data.
{
   ...
}

class MyClass3 //Reflection is off (by default).
{
   ...
}

class MyClass4 //Reflection is off and MyClass will not know about his members (include MyNestedClass), but Reflection is on for MyNestedClass and it will know about his members
{
   @reflect!basic class MyNestedClass
   {

   }
}

Thus I need to annotate module to say that it should be reflected (or shouldn't).
Of course, this issue (module's UDA) doesn't limited by my task. All times when user need to annotate a class, he may need to annotate the module, because "module" is the same language object as the "class".

[ghost]

Should we try to use the milestone feature to tag this pull request for 2.066? I'm wondering whether adding some kind of tagging would enable us to quickly see e.g. which pulls need reviewing, which are stalled for later, etc.

[mihails-strasuns]

@AndrejMitrovic Will also reduce amount of "will it be in next release?" questions. I'd love to see a milestone for every release, but is something that needs to be done by commiters/reviewers.

[9rnsr]

Added this enhancement in the list of 2.066 targets on Agenda page.

[WalterBright]

BTW, I don't think we should allow prefixes on module statements. The appearance of module as the first token in the file is handy for detecting D source code.

[dnadlinger]

… except that a module statement isn't mandatory?

[ghost]

Might as well make it:

if (!decl) 
    return;

[9rnsr]

Because the style if (decl) { ... } is consistent with other functions style in attrib.c.

[ghost]

It looks ok, but will probably need another reviewer as there's quite a few changes.

[9rnsr]

Rebased.

[9rnsr]

Thank you!