[PATCH] Distinguishing between user-written and implicit attributes

[Aaron Ballman]

Committed in r199378 with the changes to the ASM label attributes.
I'll look into giving them a keyword spelling soon.

~Aaron

On Wed, Jan 15, 2014 at 10:38 PM, Aaron Ballman <aaron at aaronballman.com> wrote:
> On Wed, Jan 15, 2014 at 9:58 PM, Richard Smith <metafoo at gmail.com> wrote:
>>
>>
>> On Wed Jan 15 2014 at 6:17:50 PM, Aaron Ballman <aaron at aaronballman.com>
>> wrote:
>>>
>>> This patch implements the ability to distinguish between attributes
>>> explicitly written at the request of the user, and attributes
>>> implicitly generated to assist in bookkeeping by the compiler. It
>>> additionally removes the optional nature of the spelling list index
>>> when creating attributes. These two things go hand in hand because the
>>> spelling of the attribute is semantically important in some cases.
>>>
>>> All attributes now receive a table-generated factory function named
>>> CreateImplicit. This function creates a new instance of the attribute,
>>> and flags it as being implicitly created. This attribute is then
>>> inserted into the AST as normal, but consumers of the AST are able to
>>> check whether the attribute was implicitly generated, if it suits
>>> them. Currently, no AST consumer cares (outside of the AST dumper, for
>>> testing purposes), but the expectation is that this will change.
>>>
>>>
>>>
>>> If the attribute has distinct spellings (where distinct is defined as
>>> "differing in more than just spelling varieties and leading/trailing
>>> underscores), it also receives a Spelling enumeration. This allows
>>> spelling-specific implicit attributes to be created without relying on
>>> fragile mechanisms.
>>
>>
>> This looks great, thanks!
>
> Thanks!
>
>>
>> It looks like this patch marks the AsmLabelAttr as always being implicit
>> (but maybe I'm misreading). I don't think that's the best choice; it
>> corresponds to something the user explicitly wrote in the source code, and
>> is our only AST representation of that thing. (It should probably also be
>> given a Keyword spelling.)
>
> Yes, that is what's currently happening. I'll change it up -- if you
> notice any others like that, please let me know and I'll modify them
> as well. I also suspect there are cases where we add attributes
> explicitly which should be implicit that I've missed.
>
> I'll likely commit in the morning, and we can tackle anything that's
> amiss as it comes up.
>
> Thanks, again!
>
> ~Aaron