Re: c++ language question

[Jason Merrill <jason at redhat dot com>]

>>>>> "Eric" == Eric Christopher <echristo@cygnus.com> writes:

    >> Does this actually buy us much?  (I.e., were there conforming
    >> programs that didn't work before, but will work now?)

    Eric> Well.. it'll at least stop programs like:

    Eric> #include <stdlib.h> int main (void) { int wchar_t = 1; }

    Eric> from working.

    >> I'm afraid that will simply screw up people who have typedef's
    >> for wchar_t in their header files, without helping them much.
    >> Jason?

    Eric> It depends on what the policy is for conformance really.
    Eric> Ultimately we need to make the compiler conform to the
    Eric> standard, but there's room for flexibility.

My goal has always been to make the compiler conform to the standard,
but still DTRT with a pedwarn on ill-formed code (if reasonable).
This is particularly important in system header files, where the
pedwarn should be omitted.

This is why the __wchar_t hack was implemented in the first place;
many system headers had

 typedef int wchar_t;

which breaks badly if wchar_t is a keyword.  On the other hand, we
chose to solve the related problem for bool using fixincludes; that
may well be adequate for wchar_t as well.

If we don't go with your patch, we will need to extend the current
code to check for redeclaration of wchar_t to consider declarations in
other scopes as well.

So go ahead: put in the patch and see what breaks.

Jason