[SYCL] Let tablegen handle mutually exclusive decl attrs

[AaronBallman]

The new MutualExclusions tablegen facilities allow us to remove a bunch
of custom diagnostic checking code. This also fixes an issue with the
upstream implementation of the MutualExclusions facilities for
DeclOrStmtAttr attributes.

[smanna12]

Thank you.

[premanandrao]

@AaronBallman, isn't the second bool the same as:
!Attr.isSubClassOf("StmtAttr") || Attr.isSubClassOf("DeclOrStmtAttr");

Wouldn't this match anything that is not StmtAttr?

[AaronBallman]

That's correct.

There are three kinds of attributes (four-ish if you count pragmas, but those never get here): type, stmt, and decl. We bail out early if it's a type attribute on line 3664, so by this point it should either be a statement attr, a decl attr, or both.

[premanandrao]

Thanks!

[premanandrao]

How do you want to handle this PR? Is there a separate PR to address statement attributes? I just don't want us to lose track of it. I will approve this one, assuming that one won't fall through the cracks.

[smanna12]

Is there a separate PR to address statement attributes?

I think so. Here is the discussion:
Originally posted by @AaronBallman in #3507 (comment)

[AaronBallman]

Statement attributes need more love; I'm working on that refactoring currently. If it looks like that refactoring has to be set down for some reason, I'll file an issue about it.

[bader]

@premanandrao, @AaronBallman, is this conversation resolved or I should wait for PR updates?

[AaronBallman]

I'm not expecting to change this PR further unless I've missed something or there are other concerns.

[premanandrao]

@premanandrao, @AaronBallman, is this conversation resolved or I should wait for PR updates?

Sorry, I should have been more explicit. I considered this resolved.