PR34431 and friends

[Paul Richard Thomas]

Tobias,

Just to let you know of progress:

In the version of the patch that you last saw, this was an (untested
in the testsuite!) regression:

real (kind(0d0)) function foo ()
  foo = real (kind (foo))
end function

where the error was an "unexpected character in name".

I have spent the last week trying to understand the problem, only to
realise that, as in decode statement, the re-matching of the typespec
must be done with error buffering on and that the errors must be
cleared afterwards.  The annoying thing is that I had already been
through this before in

I am going to do some further checks and cleaning up today, with a
view to submit tonight.... phew!

Cheers

Paul