Dear all,

the attached patch picks up an observation by Tobias that we did
not specify the RESTRICT qualifier for optional arguments even
if that was allowed.  In principle this might have prevented
better optimization.

While looking more closely, I found and fixed an issue with CLASS
dummy arguments that mishandled this.  This revealed a few cases
in the testsuite that were matching the wrong patterns...

Regtested on x86_64-pc-linux-gnu.  OK for mainline?

Thanks,
Harald