struct padding ???

Collapse
This topic is closed.
X
X
 
  • Time
  • Show
Clear All
new posts
  • vikas talwar

    struct padding ???

    Hi All,

    Can you please explain me how the 'C' compiler allocate memory to
    'struct'.
    Please go thu the example below and pls suggest me the solution for my
    problem.

    Here is my structure definition

    struct my_dev {
    char abc;
    } ;

    and now when I do
    printf("size=%d \n", sizeof(struct my_dev));
    output is: 4

    But I expect the size of struct = '1'.

    Is there any trick to get the struct to occupy memory with pad of 1
    byte and not 4 bytes..


    --
    Thanks
    Vikas
  • Jens Thoms Toerring

    #2
    Re: struct padding ???

    vikas talwar <vikas.talwar@g mail.comwrote:
    Can you please explain me how the 'C' compiler allocate memory to
    'struct'.
    Please go thu the example below and pls suggest me the solution for my
    problem.
    Here is my structure definition
    struct my_dev {
    char abc;
    } ;
    and now when I do
    printf("size=%d \n", sizeof(struct my_dev));
    output is: 4
    But I expect the size of struct = '1'.
    Is there any trick to get the struct to occupy memory with pad of 1
    byte and not 4 bytes..
    The compiler is obviously putting in 3 padding bytes after the
    'abc' element of the structure. It does that to make sure when
    you create an array of such structures all the structures of
    the array end up on correctly aligned memory positions. If
    it wouldn't do that something simple as

    my_dev xxx[ 2 ];

    xxx.abc = 'c';

    could lead to a crash of the program, at least on your machine
    (on mine the sizeof for the structure is repoted as 1, so on
    my machine it doesn't seem necessary to insert padding bytes).
    For the same reason also padding bytes can appear between the
    elements of a structure.

    There is no C standard compliant way to elimintae padding bytes.
    Some compilers have some extra options to avoid the addition of
    padding bytes, for e.g. gcc you would have to add

    __attribute__(( packed))

    directly after the structure declaration) but, unless you know
    exactky what you're doing and the trouble you can get into with
    that I can only recommend not to use it. Perhaps you should
    tell what exactly you want to achieve that would require the
    elimination of padding bytes and perhaps there's a better way
    to do it.
    Regards, Jens
    --
    \ Jens Thoms Toerring ___ jt@toerring.de
    \______________ ____________ http://toerring.de

    Comment

    • Jens Thoms Toerring

      #3
      Re: struct padding ???

      Jens Thoms Toerring <jt@toerring.de wrote:
      ....it wouldn't do that something simple as
      my_dev xxx[ 2 ];
      xxx.abc = 'c';
      Sorry, I meant

      xxx[1].abc = 'c'

      here, the original version would be a syntax error...

      Regards, Jens
      --
      \ Jens Thoms Toerring ___ jt@toerring.de
      \______________ ____________ http://toerring.de

      Comment

      • Keith Thompson

        #4
        Re: struct padding ???

        jt@toerring.de (Jens Thoms Toerring) writes:
        vikas talwar <vikas.talwar@g mail.comwrote:
        >Can you please explain me how the 'C' compiler allocate memory to
        >'struct'.
        >Please go thu the example below and pls suggest me the solution for my
        >problem.
        >
        >Here is my structure definition
        >
        >struct my_dev {
        > char abc;
        >} ;
        >
        >and now when I do
        > printf("size=%d \n", sizeof(struct my_dev));
        >output is: 4
        >
        >But I expect the size of struct = '1'.
        >
        >Is there any trick to get the struct to occupy memory with pad of 1
        >byte and not 4 bytes..
        See questions 2.12 and 2.13 in the comp.lang.c FAQ,
        <http://www.c-faq.com/>.
        The compiler is obviously putting in 3 padding bytes after the
        'abc' element of the structure. It does that to make sure when
        you create an array of such structures all the structures of
        the array end up on correctly aligned memory positions. If
        it wouldn't do that something simple as
        >
        my_dev xxx[ 2 ];
        >
        xxx.abc = 'c';
        >
        could lead to a crash of the program, at least on your machine
        (on mine the sizeof for the structure is repoted as 1, so on
        my machine it doesn't seem necessary to insert padding bytes).
        For the same reason also padding bytes can appear between the
        elements of a structure.
        Actually, I can't think of any good reason why the compiler *needs* to
        pad "struct my_dev" to 4 bytes. It *could* internally treat a
        structure with a single member the same way it would treat an object
        of the member type.

        There is a requirement for all pointers to structures to have the same
        representation; on an implementation where byte pointers and word
        pointers are represented differently, that could force 4-byte
        alignment for all structures, even small ones. But most systems use
        the same representation for all pointers.

        Probably the compiler's developers decided that it would be simpler to
        force all structures to be word-aligned (where a "word" is probably 4
        bytes). They're allowed to do that. The standard allows padding
        after members of structures; it doesn't require that padding to be the
        minimum necessary. A perverse implementation could add as much
        padding as it likes, for no good reason at all.

        [...]

        For your particular problem, why do you need a structure at all?
        Rather than
        struct my_dev {
        char abc;
        };
        why not just use type char directly (with a typedef if you like)?
        (Using a structure does have the advantage of type safety; "struct
        my_dev" is incompatible with other types. But then you have to pay
        the price of letting the compiler add extra padding.)

        --
        Keith Thompson (The_Other_Keit h) kst-u@mib.org <http://www.ghoti.net/~kst>
        Nokia
        "We must do something. This is something. Therefore, we must do this."
        -- Antony Jay and Jonathan Lynn, "Yes Minister"

        Comment

        Working...