Re: parsing variable arg lists via va_list pointers (any gurus here?)

Collapse
This topic is closed.
X
X
 
  • Time
  • Show
Clear All
new posts
  • =?ISO-8859-1?Q?Tom=E1s_=D3_h=C9ilidhe?=

    Re: parsing variable arg lists via va_list pointers (any gurus here?)

    On Jun 3, 3:23 am, Jesse Ziser <d...@spam.diew rote:
    > ANSI says that when you pass a va_list object to
    a function and call va_arg() on it within that function, the value of
    the va_list object is undefined upon return.

    The relevant paragraph from the Standard is:

    ---- Begin Quote ----
    The type declared is va_list
    which is an object type suitable for holding information needed by the
    macros
    va_start, va_arg, va_end, and va_copy. If access to the varying
    arguments is
    desired, the called function shall declare an object (generally
    referred to as ap in this
    subclause) having type va_list. The object ap may be passed as an
    argument to
    another function; if that function invokes the va_arg macro with
    parameter ap, the
    value of ap in the calling function is indeterminate and shall be
    passed to the va_end
    macro prior to any further reference to ap.
    ---- End Quote ----

    At first reading, I thought this was just saying the following:
    * C passes arguments by value, i.e. the object in the calling
    function remains unchanged
    * The va_* macros alter their va_list argument
    * You can only use the most up-to-date va_list, you can't use an
    old out-of-date one

     Behold:
    >
    void parse_half_of_a rg_list( va_list args )
    {
       ...
       foo_t x = va_arg( args, foo_t );
       bar_t y = va_arg( args, bar_t );
       ...
    >
    }
    >
    void superfunc( int a, ... )
    {
       va_list args;
    >
       va_start( args, a );
    >
       parse_half_of_a rg_list( args );
       /* Uh-oh!  ANSI says args is undefined now,
          but there's still more to parse */
       parse_half_of_a rg_list( args );
    >
       va_end( args );
    >
    }

    The va_list object in "main" remains unchanged, therefore it is no
    longer "up to date" when you go to try use another va_* macro on it.

    So, I reasoned, why not pass a va_list * instead of a va_list?  Surely
    that can't do any harm... I mean, since va_list is a data type, it
    should be possible to create a pointer to it, and *(&x) should be the
    same as x for any named type and for any operations, right?  Lo:
    >
    void parse_half_of_a rg_list( va_list *args )
    {
       ...
       foo_t x = va_arg( *args, foo_t );
       bar_t y = va_arg( *args, bar_t );
       ...
    >
    }
    >
    void superfunc( int a, ... )
    {
       va_list args;
    >
       va_start( args, a );
    >
       parse_half_of_a rg_list( &args );
       parse_half_of_a rg_list( &args );
    >
       va_end( args );
    >
    }
    >
    I haven't been able to find anything in any standard that explicitly
    allows or prohibits this, but there aren't many ways I can think of that
    some devious library author could screw this up.  The thing is, this
    code will exist for a long time and will quite possibly need to run
    under every semi-major system and C compiler that will come into
    existence in the next 20 years.  As of right now it will need to
    immediately work on Sun, Mac, Cygwin, and Linux and compile under gcc,
    pgcc, and icc.

    I find the original paragraph from the Standard to be very strange. I
    don't see why it went to the bother of mentioning calling and called
    functions when it just could have said "the va_list must always be up-
    to-date". This kind of leads me to believe that maybe there's some
    other requirement than the va_list being up to date... maybe something
    to do with the stack.

    So, I guess the gist is that I want to do something really weird but I
    want it to be really portable, and I'm afraid that might be an
    impossibility.  Not being able to do this would mean I'd have to use a
    bunch of structs or arrays with void pointers everywhere, which is a
    considerable burden as this function will be called very frequently.
    >
    I am well aware that this is a really bizarre thing to have to do.
    >
    Please let me know if you know a system that breaks this, or if you have
    good cause to believe that this will always work.
    >
    Thanks in advance.

    Wait and see what others have to say.
  • christian.bau

    #2
    Re: parsing variable arg lists via va_list pointers (any gurus here?)

    On Jun 3, 4:28 am, Tomás Ó hÉilidhe <t...@lavabit.c omwrote:
    I find the original paragraph from the Standard to be very strange. I
    don't see why it went to the bother of mentioning calling and called
    functions when it just could have said "the va_list must always be up-
    to-date". This kind of leads me to believe that maybe there's some
    other requirement than the va_list being up to date... maybe something
    to do with the stack.
    There are some interesting architectures out there that do interesting
    things in the va_arg macros. Sparc machines come to mind, or Itanium.
    There may be compiler support needed (even the x86-64 ABI looks like a
    major pain).

    Also note that va_list could be defined either as an array or as a
    struct. All the macros will work if you use them as the C Standard
    says, but the effect of applying an address operator or having a
    pointer to a va_list could be very different in both cases.

    Comment

    • Jesse Ziser

      #3
      Re: parsing variable arg lists via va_list pointers (any gurus here?)

      christian.bau wrote:
      On Jun 3, 4:28 am, Tomás Ó hÉilidhe <t...@lavabit.c omwrote:
      >
      >I find the original paragraph from the Standard to be very strange. I
      >don't see why it went to the bother of mentioning calling and called
      >functions when it just could have said "the va_list must always be up-
      >to-date". This kind of leads me to believe that maybe there's some
      >other requirement than the va_list being up to date... maybe something
      >to do with the stack.
      >
      There are some interesting architectures out there that do interesting
      things in the va_arg macros. Sparc machines come to mind, or Itanium.
      There may be compiler support needed (even the x86-64 ABI looks like a
      major pain).
      >
      Also note that va_list could be defined either as an array or as a
      struct. All the macros will work if you use them as the C Standard
      says, but the effect of applying an address operator or having a
      pointer to a va_list could be very different in both cases.
      My problem is precisely that the C standard _doesn't_ really say how to
      use them. It only outlines the basics.

      Could you explain how address operators and pointers could behave
      differently depending on whether va_list is a struct or array type? To
      my knowledge, if stdarg.h contains either of these:

      typedef weird_t va_list[42];
      typedef struct weird va_list;

      And I declare:

      va_list args;

      Then, in either case, *(&args) will still behave exactly the same as
      args. Is there some case I'm not thinking of in which this is not true?

      Comment

      • Richard Heathfield

        #4
        Re: parsing variable arg lists via va_list pointers (any gurus here?)

        Jesse Ziser said:

        <snip>
        >
        Could you explain how address operators and pointers could behave
        differently depending on whether va_list is a struct or array type? To
        my knowledge, if stdarg.h contains either of these:
        >
        typedef weird_t va_list[42];
        typedef struct weird va_list;
        >
        And I declare:
        >
        va_list args;
        >
        Then, in either case, *(&args) will still behave exactly the same as
        args. Is there some case I'm not thinking of in which this is not true?
        No. * and & "cancel".

        --
        Richard Heathfield <http://www.cpax.org.uk >
        Email: -http://www. +rjh@
        Google users: <http://www.cpax.org.uk/prg/writings/googly.php>
        "Usenet is a strange place" - dmr 29 July 1999

        Comment

        • Bartc

          #5
          Re: parsing variable arg lists via va_list pointers (any gurus here?)


          "Richard Heathfield" <rjh@see.sig.in validwrote in message
          news:xMmdnZ5Sq_ SIZdjV4p2dnAA@b t.com...
          Jesse Ziser said:
          >
          <snip>
          >>
          >Could you explain how address operators and pointers could behave
          >differently depending on whether va_list is a struct or array type? To
          >my knowledge, if stdarg.h contains either of these:
          >>
          >typedef weird_t va_list[42];
          >typedef struct weird va_list;
          >>
          >And I declare:
          >>
          >va_list args;
          >>
          >Then, in either case, *(&args) will still behave exactly the same as
          >args. Is there some case I'm not thinking of in which this is not true?
          >
          No. * and & "cancel".
          So in other words, *(&args) is the same as args like he said?

          (There is a slight difference: args has to be an lvalue for *(&args) to
          work, so *(&1234) and 1234 are not the same since the former won't
          compile -- assuming a suitable context.)

          --
          Bartc


          Comment

          • Richard Heathfield

            #6
            Re: parsing variable arg lists via va_list pointers (any gurus here?)

            Bartc said:
            >
            "Richard Heathfield" <rjh@see.sig.in validwrote in message
            news:xMmdnZ5Sq_ SIZdjV4p2dnAA@b t.com...
            >Jesse Ziser said:
            <snip>
            >>Then, in either case, *(&args) will still behave exactly the same as
            >>args. Is there some case I'm not thinking of in which this is not
            >>true?
            >>
            >No. * and & "cancel".
            >
            So in other words, *(&args) is the same as args like he said?
            Yes. My "no" was in response to "is there some case I'm not thinking of in
            which this is not true".
            (There is a slight difference: args has to be an lvalue for *(&args) to
            work, so *(&1234) and 1234 are not the same since the former won't
            compile -- assuming a suitable context.)
            Oops, you're right - I missed that case. Thanks for the correction.


            --
            Richard Heathfield <http://www.cpax.org.uk >
            Email: -http://www. +rjh@
            Google users: <http://www.cpax.org.uk/prg/writings/googly.php>
            "Usenet is a strange place" - dmr 29 July 1999

            Comment

            Working...