c/1027: slightly misleading printf format warning

[Kaveh R. Ghazi]

 > From: Gabriel Dos_Reis <gdosreis@sophia.inria.fr>
 > 
 > |  > I've already put in the necessary machinery for so doing in
 > |  > diagnostics.*.  What remains to be done is to replace error_with_decl()
 > |  > and warning_with_decl() to error() and warning() (maybe error_at())
 > |  > with use of those specifiers.
 > |  > -- Gaby
 > | 
 > | If we change error/warning to allow non-standard specifiers then we
 > | can't use attribute printf on these functions and we'll lose this
 > | valuable error checking in gcc sources, right?
 > 
 > I have trouble to get your point. Could you elaborate?

Here's my point.  If I understand your messages correctly, you wish to
change error/warning to accept the %T modifier just like cp_error
(etc), do.  In cp-tree.h, it specifically says that cp_error and
friends aren't suitable for attribute printf because it accepts
non-standard specifiers, namely %T.

So if error/warning start to accept %T, then we must remove the
attribute from them and then specifier-vs-arg errors could creep back
in because we no longer check all code paths at compile-time.

Perhaps I've misunderstood what you intend, if so I apologize.
However if I'm correct, then I have concerns over this loss of
compile-time sanity checking.

Many errors were detected when we started putting attribute printf on
our diagnostic functions.  I suspect some errors remain in the cp_*
style ones, precisely because they can't do this.

		--Kaveh
--
Kaveh R. Ghazi			Engagement Manager / Project Services
ghazi@caip.rutgers.edu		Qwest Internet Solutions