[Bug c/16351] NULL dereference warnings

["manu at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=16351

--- Comment #37 from Manuel LÃpez-IbÃÃez <manu at gcc dot gnu.org> ---
(In reply to Jeffrey A. Law from comment #36)
> It's no different maybe uninitialized warnings.  

Well, one can silence uninitialized warnings by simply initializing the
variable. I don't see how this code can be refactored to silence the warning
(and the way I understood the warning is by looking at the GCC dumps, which
people will not do).

> Note that in this case, the compiler should be determining that MAP can't be
> NULL and thus linemap_check_ordinary can't return NULL and propagating that
> information.  So if we're still getting a warning, then something is wrong
> somewhere.

But how it can do such a thing if the definition of linemap_add were not
available?

(In this case it is, but GCC's IPA is not smart enough to figure it that it
cannot return NULL. The code injected by -f-isolate-paths survives until the
end).

In any case, I don't see how to keep this in -Wall or -Wextra if it breaks GCC
bootstrap. And options that are not enabled by either of those are rarely used
and tend to bit-rot or receive little attention.

Hum, perhaps, we do really need some kind of -Weverything that includes not
every warning, but those that are potentially useful but too noisy to not break
GCC bootstrap.