Unions and structures implementation in C

[Jason Curl]

Hello,

Does the C standard specify (and if so, how, or possibly a reference to
the standard - I don't have a copy). I am also assuming that CHAR_BIT is
8 (for Posix systems)

a) How elements in a union overlap
b) limitations on how elements in a structure are padded

This will help me in writing (hopefully) portable C code that can
communicate across different architectures, that may be faster and more
memory efficient than copying from one buffer to another.

Or is the only practical way to memcpy() taking into account endianness
of the machine into a "char *buffer"?

Thanks,
Jason.

[pemo]

"Jason Curl" <j_********@mot orola.com> wrote in message
news:di******** **@engnntp2.cig .mot.com...

Hello,

Does the C standard specify (and if so, how, or possibly a reference to
the standard - I don't have a copy). I am also assuming that CHAR_BIT is 8
(for Posix systems)

a) How elements in a union overlap
b) limitations on how elements in a structure are padded

This will help me in writing (hopefully) portable C code that can
communicate across different architectures, that may be faster and more
memory efficient than copying from one buffer to another.

Or is the only practical way to memcpy() taking into account endianness of
the machine into a "char *buffer"?

You can get a copy of the c99 std from the ANSI website for $18 - might be
worth investing?

[Eric Sosman]

Jason Curl wrote:

Hello,

Does the C standard specify (and if so, how, or possibly a reference to
the standard - I don't have a copy). I am also assuming that CHAR_BIT is
8 (for Posix systems)

a) How elements in a union overlap
All the union's elements start at the same address.
More formally, a pointer to the union can be converted
to a pointer to any of its elements, and vice versa.
b) limitations on how elements in a structure are padded
Padding may appear after any element of the struct.
(People often say "Between any elements, and at the end
of the struct," but that's just a longer way of saying
the same thing.) The struct's first element starts at
the struct's address; more formally, a pointer to the
struct can be converted to a pointer to its first element,
and vice versa.
This will help me in writing (hopefully) portable C code that can
communicate across different architectures, that may be faster and more
memory efficient than copying from one buffer to another.

Or is the only practical way to memcpy() taking into account endianness
of the machine into a "char *buffer"?

memcpy() might not be enough. For example, you may
want to communicate a `long' value, but `long' on machine A
is 32 bits while on machine B it's 64. Or perhaps you want
to move a `long double' value between a machine were it is
actually more precise than plain `double' and a machine
where it's really just a plain `double' with a different name.
Mere memcpy() cannot reconcile such differences.

What's usually needed is a careful specification of the
external format -- the way the information looks "on the
wire" or "in the file." Then for each machine you write a
pair of functions (or a set of pairs): one to convert the
machine's idiosyncratic internal data representation to the
external form, and one to digest the external form and convert
it to the internal representation. A visit to someplace like
wotsit.org may give you some ideas about how to specify an
external format that has a reasonable chance at portability:
study the way JFIF or WAV files (for example) are described,
and see if you can steal, er, "be inspired by" their techniques.

--
Eric Sosman
es*****@acm-dot-org.invalid

[Emmanuel Delahaye]

Jason Curl a écrit :

Does the C standard specify (and if so, how, or possibly a reference to
the standard - I don't have a copy). I am also assuming that CHAR_BIT is
8 (for Posix systems)

a) How elements in a union overlap
b) limitations on how elements in a structure are padded

Are your facing such questions probably means that you have a design
error. Unions are used to save memory, but certainely not to do parallel
memory mappings.

Arrays of unsigned char and bitwise operators are your friends.

[Default User]

Jason Curl wrote:

Hello,

Does the C standard specify (and if so, how, or possibly a reference
to the standard - I don't have a copy). I am also assuming that
CHAR_BIT is 8 (for Posix systems)

a) How elements in a union overlap
b) limitations on how elements in a structure are padded

This will help me in writing (hopefully) portable C code that can
communicate across different architectures, that may be faster and
more memory efficient than copying from one buffer to another.

Or is the only practical way to memcpy() taking into account
endianness of the machine into a "char *buffer"?

There's no way you're going to achieve what you want with unions and
memcpy and such. Cross-platform communication has many potential traps
and pitfalls. Besides structure padding, there are issues with size and
endianess of basic data types.

Look up the topic of serialization. You will find much prior art on the
subject. The FAQ from another newsgroup, comp.lang.c++ has some
information.
http://www.parashift.com/c++-faq-lit....html#faq-36.1

Brian