Pattern match checking across monadic bind

As noted by @barci2, it would be nice if the following kind of programs didn't cause any pattern match warnings to be emitted:

data T = T1 { fld :: Int } | T2

foo :: [T]
foo = do { T1 {} <- [a,b]; v <- [ b { fld = 4 }, a { fld = 3 } ]; return v }

bar :: Maybe T
bar = do { T1 {} <- Just c; return $ c { fld = 5 } }

In these examples, a, b, and c should all be considered to be scrutinees from the point of view of the function matchSinglePatVar. Doing so would then mean that the call to addCoreScrutTmCs inside that function will propagate the fact that a, b, and c are all of the form T1 _, which will allow GHC to see that the record updates are not partial.

To do this, in the compiler, we would need some kind of function:

bindStmtRhsScruts :: XBindStmtTc -> CoreExpr -> [CoreExpr]

which would be able to extract out scrutinees from an expression, e.g.

bindStmtRhsScruts ( BindStmt for List  ) ( Core of explicit list "[a,b]" ) = [a,b]
bindStmtRhsScruts ( BindStmt for Maybe ) ( Core of "Just c" )              = [c]

I'm not sure how we would implement this function in general, without implementing ad-hoc logic for known container types such as lists.

@sgraf812, do you have any thoughts?