syntax: Add enum for defined attributes

[klutzy]

This patch adds attr::DefinedAttr enum, which contains all attributes used by rust.
It also adds attr::contains_attr() function which is similar to attr::contains_name() but accepts enum variant instead of string.
So it becomes possible to catalog all attributes at one place.

cc #7318

[alexcrichton]

This seems pretty awesome to me!

We'd be putting a large emphasis on the initial attribute name rather than the inner attributes which is the only concern of mine. Right now attributes are consistent in that as you go inside attributes you continue to get more attributes, but this is changing that such that the outermost layer is an enum and the innermost layer is all strings.

I think I'm ok with this, but I'm curious what others' opinions on this are. I'm not entirely convinced that this is a huge win over the unified list inside of middle::lint that we have right now, but I could be convinced of such.

[klutzy]

Right, inner attributes are tricky and I feel quite uncomfortable to use contains_attr for outmost attribute but contains_name for inner ones.

I initially thought of adding some fields to DefinedAttr as like:

pub enum DefinedAttr {
    AttrCrateType(bool), // is_lib
    AttrFeature(~[AttrFeatureItem]), // enum AttrFeatureItem { Globs, MacroRules, ... }
    AttrNoUv, // no arg
    ...
}

and some function parsing MetaItem to Option<DefinedAttr>. But it seems too complex to use (contains_attr must be changed somehow).

Alternatively, we may just define e.g. AttrFeatureItem and use contains_attr(feature_attr.meta_item_list(), MacroRules) for feature(macro_rules).

[klutzy]

Rebased: replaced most uses of attr.name() == "..." by attr.is_defined_attr(...).

[huonw]

These should be lower case or the message will look like obsolete attribute: Use ....

(Also, auto_{en,de}code have be gone for a long time, and had a deprecation error message for a long time too, so I think we can just drop them at this point.)

[alexcrichton]

I've added this to the agenda for the meeting tomorrow, it's unclear to me whether this is the direction that we want to go in.

[alexcrichton]

It was decided in today's meeting that this is not the right direction for attributes to be going in. It is an explicit intention of attributes to be extensible, and having an enumeration of all known attributes goes against this intent.

I think that we still need a method of documenting attributes, but an enumeration in the compiler isn't necessarily the right direction to go in.