On 09/01/2015 11:29 AM, Joseph Myers wrote:
On Tue, 1 Sep 2015, Martin Sebor wrote:
Attached is an updated patch that avoids diagnosing taking the address
of implicitly declared library builtins like abs, bootstrapped and
tested on ppc64le with no regressions.
The tweak below was added to reject_gcc_builtin make it possible.
Since the expression is in c-family/c-common.c, the more descriptive
C_DECL_IMPLICIT that exists for this purpose is not available (it's
defined in c/c-tree.h).
+ && (c_dialect_cxx () || !DECL_LANG_FLAG_2 (expr))
I don't think hardcoding DECL_LANG_FLAG_2 here is a good idea; rather,
some sort of c_decl_implicit hook should be defined (that would just
return false for C++) - existing practice is various functions that have
different definitions for C and C++.
I've made the suggested change and tested it by bootstrapping
and running the two tests.
Before I post another revision of the patch for review let me
make sure there are no other change requests, and that introducing
a new c_decl_implicit hook here is what everyone wants.
The original patch had a distinct function for each of the C and
C++ front ends to issue the diagnostics because each had slightly
different tests. I've since merged the two functions into one on
Jason's suggestion and removed most of the differences. Adding
a hook seems to be step in the opposite direction.
Having now made this change, I don't think the added complexity
of three declarations and two trivial definitions of the new
c_decl_implicit function across five files is an improvement,
especially since we're still checking for c_dialect_cxx().
(The change simply replaces:
&& (c_dialect_cxx () || !DECL_LANG_FLAG_2 (expr))
with
&& (c_dialect_cxx () || c_decl_implicit (expr))
as suggested.)
Having also looked at the existing code in c-family/common.c
that does things differently between C and C++ I see examples
of both hooks for more complex computations, and conditionals
similar to the one I just replaced for simpler things.
I also noticed uses of DECL_LANG_FLAG_4 in the definitions of
what appear to be C-specific macros in c-family/c-common.h,
and then uses of the same macro in definitions of a C++-specific
macro in cp/cp-tree.h.
In this light it seems to me that leaving the test for the flag
as it was would be both in keeping with existing practice and
preferable to introducing the hook.
Let me know.
Thanks
Martin
