Re: Add a __nowarn__ keyword

[DJ Delorie <dj at redhat dot com>]

> and use the typedef instead?

What, you don't like all those "structs" around?  ;-)

> Is it correct that the "user interface" now is to say
>   
>     #pragma GCC push

$pragma GCC diagnostics push

> ?

Yes.

> If yes, is the "push" necessary?  Can we make it implicit?

So, a pop with no corresponding push, restores to... what?  The way
the code is now, if you don't push or save, it never even starts the
chain - it doesn't know the difference between a command line setting
and a #pragma without a push.  Thus, a pop wouldn't do anything
anyway.

I think what you want is, an unmatched pop restores the command line
settings, yes?  To do that, we'd need two things:

1. Something in gcc does the implicit save to start the chain, so that
   all #pragma changes are in the chain and not the table.

2. An unmatched pop points you back at the head of the chain, which is
   a no-op state created by that save.

But then, there would be no way to restore some project-wide settings
pulled in by a #include.

Or did you mean that a pop always pops just ONE #pragma?  The current
structure allows that, easier if we move the state struct to
diagnostics.h.  Then, you just restore to state->previous.  But is
that a pop, or an undo?