bitfield: sorry, maybe I didn't look closely enough here.
I had to patch it in 4.5, but maybe not in 4.7.
I had gotten warnings or errors about integers being truncated.
I'll look later.
gcc_unreachable: I don't think it is a bug, for a compiler
to not be gcc. That appears to be all it takes for gcc_unreachable
to be nothing special, just a function call.
/* Use gcc_unreachable() to mark unreachable locations (like an
unreachable default case of a switch. Do not use gcc_assert(0). */
#if (GCC_VERSION >= 4005) && !ENABLE_ASSERT_CHECKING
#define gcc_unreachable() __builtin_unreachable()
#else
#define gcc_unreachable() (fancy_abort (__FILE__, __LINE__, __FUNCTION__))
#endif
/* Redefine abort to report an internal error w/o coredump, and
reporting the location of the error in the source file. */
extern void fancy_abort (const char *, int, const char *) ATTRIBUTE_NORETURN;
#define abort() fancy_abort (__FILE__, __LINE__, __FUNCTION__)
#if (GCC_VERSION < 2007)
# define __attribute__(x)
#endif
#ifndef ATTRIBUTE_NORETURN
#define ATTRIBUTE_NORETURN __attribute__ ((__noreturn__))
#endif /* ATTRIBUTE_NORETURN */
- Jay
----------------------------------------
> Date: Sat, 9 Feb 2013 23:41:24 -0800
> Subject: Re: problems compiling 4.7, with Solaris cc and/or Solaris CC (C++)
> From: pins...@gmail.com
> To: jay.kr...@cornell.edu
> CC: gcc@gcc.gnu.org
>
> On Sat, Feb 9, 2013 at 3:49 PM, Jay K <jay.kr...@cornell.edu> wrote:
> > problems compiling 4.7, with Solaris cc and/or Solaris CC (C++)
> >
> >
> > 1) ENUM_BITFIELD is not portable. I've reported this before.
> >
> >
> > It should be more like:
> >
> >
> > #ifdef __cplusplus
> > #define ENUM_BITFIELD(TYPE, NAME, SIZE) enum TYPE NAME : SIZE
> > #elif (GCC_VERSION > 2000)
> > #define ENUM_BITFIELD(TYPE, NAME, SIZE) __extension__ enum TYPE NAME : SIZE
> > #else
> > #define ENUM_BITFIELD(TYPE, NAME, SIZE) unsigned int NAME : SIZE
> > #endif
>
>
> Could you explain why the current way is not correct?
>
> Which does:
> #ifdef __cplusplus
> #define ENUM_BITFIELD(TYPE) enum TYPE
> #elif (GCC_VERSION > 2000)
> #define ENUM_BITFIELD(TYPE) __extension__ enum TYPE
> #else
> #define ENUM_BITFIELD(TYPE) unsigned int
> #endif
>
> Which is no different from what you have above really.
>
>
> >
> >
> > and then modify the uses to fit -- three parameters.
> >
> >
> > It is likely that in 4.8 this is moot, as the C++ case will be the only one
> > remaining.
> >
> >
> > 2) given:
> >
> > int foo()
> > {
> > gcc_unreachable();
> > }
> >
> >
> > Solaris cc/CC gives a warning or maybe an error.
>
> That is a bug with Oracle's compilers as nothing after gcc_unreachable
> is reachable
>
> > It should be:
> >
> >
> > int foo(void)
> > {
> > gcc_unreachable();
> > return 0;
> > }
> >
> >
> > This occurs a few times e.g. in i386.md.
> >
> >
> > Despite me being uncertain between warnings and errors and cc and CC,
> > these do definitely show up as problems.
> > I've been building a few versions of gcc through the years from 4.2
> > and up (yeah, yeah, I know it goes back much further) with a variety
> > of host compilers, mostly a few versions of gcc/g++ but sometimes Solaris
> > cc/CC,
> > maybe others, and I've had to patch these repeatedly. Again, today, in 4.7.
> >
> >
> > I do use -disable-bootstrap, to drastically cut build times.
>
> Cut build times but depends on bugs in the compiler which compiles stage1.
>
>
> > It is possible that contributes to "the problem", but I also think these
> > are reasonable changes and -disable-bootstrap is really nice to use, it
> > saves a lot of time and these aren't awful ancient non-optimizing host
> > compilers.
>
> Except some host compilers have bugs. That is the point of the
> bootstrap rather than the non-optimizing host compilers. As show
> above Oracle's compilers have bugs already.
>
> Thanks,
> Andrew Pinski
