[lkml]   [2004]   [Jun]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [PATCH][2.6.6-rc3] gcc-3.4.0 fixes
    H. Peter Anvin writes:
    > Followup to: <>
    > By author: Mikael Pettersson <>
    > In newsgroup:
    > >
    > > 'ptr' _is_ a char pointer, and the code (visible in the part of
    > > the patch you didn't include) already performed pointer arithmetic
    > > on it relying on it being a char pointer. The old code had no
    > > sane reason at all for updating 'ptr' via a cast-as-lvalue.
    > >
    > > cast-as-lvalue is not proper C, has dodgey semantics, and can
    > > always be replaced by proper C.
    > >
    > I don't see how it has dodgey semantics for cast of pointers or
    > [u]intptr_t to pointers.

    You're assuming pointers have uniform representation.
    C makes no such guarantees, and machines _have_ had
    different types of representations in the past.
    Some not-so-obsolete 64-bit machines in effect use fat
    representations for pointers to functions (descriptors),
    but they usually cheat and use pointers to the descriptors
    instead. However, a C implementation could legally
    represent a function pointer as a 128-bit value, while
    data pointers remain 64 bits.

    A cast fundamentally involves an assignment conversion,
    a copy to a temporary, and it yields an rvalue.
    Even if we allow its use as an lvalue, the semantics
    would still be to assign the copy not the original.
    So cast-as-lvalue as gcc implemented it changed two
    major aspects of the semantics. Call me conservative
    if you like, but that's simply not C any more.

    Other gcc extensions, such as __inline__, __attribute__,
    and __asm__, do provide useful and sensible features.
    The issue with cast-as-lvalue is that it is neither
    necessary nor does it promote maintainable and portable code.

    Remember "the world's not a VAX" and "the world's not
    a 68K with 24-bit addresses" lessons of the 80s,

    > IMNSHO the fact that it breaks C++ isn't a good reason to outlaw a
    > long-documented extension for C.

    I couldn't care less about C++. There are ample reasons
    why it's a bad idea in C itself.

    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2005-03-22 14:03    [W:0.022 / U:59.728 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site