[PATCH] Generate diagnostic checks for attribute subjects

[Aaron Ballman]

On Tue, Nov 26, 2013 at 5:35 PM, Richard Smith <richard at metafoo.co.uk> wrote:
> Looks great! Just a couple of questions:
>
> 1) In CalculateDiagnostic, it looks like you sometimes pick a diagnostic
> that's not an exact match for the subjects list. What cases do we hit where
> we don't have an exact match, and have you considered extending the list to
> cover those too? (I also wonder whether we can do better here -- maybe by
> including Attr.td into DiagnosticSemaKinds.td and generating extra
> diagnostics from Attr.td or similar -- but I don't want that blocking this
> patch!).

Both of those cases kind of bug me, but there were a substantial
number of attributes using those values.  carries_dependency,
no_*_analysis, check lock, etc. I intend to resolve that sort of thing
on a case-by-case basis in future patches, hopefully resolving the
inconsistencies. I avoided it here so I wouldn't muddy the patch or
annoy DeLesley too much. ;-)

> 2) In GenerateAppertainsTo, you bail out for an attribute with custom
> parsing. I find that surprising -- that flag affects how the attribute is
> parsed into an AttributeList, but I don't see why it should have any
> implications on subject checking. Checking SemaHandler here would make sense
> to me, but that looks unnecessary. What's this for?

It may be unnecessary, but it's there because custom parsing
attributes generally have more complex subjects anyway (the only one
which kind of came close was type_tag_for_datatype, but the rest are
complex). Also, anything that's custom parsed automatically opts out
of common attribute handling as it stands. This was another area I was
intending to do future research on.

~Aaron