Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!bloom-beacon!adam.pika.mit.edu!scs
From: scs@adam.pika.mit.edu (Steve Summit)
Newsgroups: comp.lang.c
Subject: Re: C pointers
Summary: dynamically allocated arrays of structures
Keywords: C array struct malloc
Message-ID: <9234@bloom-beacon.MIT.EDU>
Date: 13 Feb 89 06:03:32 GMT
References: <1707@shaffer.UUCP>
Sender: daemon@bloom-beacon.MIT.EDU
Reply-To: scs@adam.pika.mit.edu (Steve Summit)
Lines: 75

In article <1707@shaffer.UUCP> jcrowe@shaffer.UUCP (Joe Crowe) writes:
>I have a structure which contains among
>other things a pointer to an array of structures declared as follows:
>typedef struct XYZ
>{
>     struct diag_cmd    (*p_cmd_blk)[];
>} t_xyz;
>
>     I am allocating storage for the above structures dynamically and
>setting the pointer p_cmd_blk manually.  All of this works.

While your declaration of a pointer to an array of structures is
correct, it has two problems:

     1.	The array size isn't specified, making this pointer even
	less useful than most pointers to arrays.

     2.	It probably isn't what you want.

I suspect that the intent was that an XYZ structure contain an
arbitrary number of diag_cmd structures, with the number
unknown at compile time.  That is, you wanted something like

	struct diag_cmd p_cmd_blk[n];

where n is determined at run time (n is probably another member
of the XYZ structure).  What you want is

	struct diag_cmd *p_cmd_blk;

which can be initialized with

	t_xyz.p_cmd_blk =
		(struct diag_cmd *)malloc(n * sizeof(struct diag_cmd));

and dereferenced with

	t_xyz.p_cmd_blk[i].cmd_blk_field

The key is that a pointer to a structure (or any kind of pointer,
for that matter) need not point to only one instance of the
structure.  In this case, it "points at" an "array" of n
structures, although the pointer's type is "pointer to structure"
rather than "pointer to array of structure."  For the usual
cases, "pointer to struct" is in fact preferable; pointer
arithmetic (and, by extension, array-like subscripting) works as
you probably want it to: selecting among the various single
structures in the "array" of n of them.  (In this paragraph, I
have used the terms "points at" and "array", in quotes, in a
loose way -- struct diag_cmd *p_cmd_blk neither declares an
array, nor, formally, points at one, but once allocated, it acts
like an array for all practical purposes.)

"pointer to array of struct" would, on the other hand, increment
or index over entire arrays of structures at once, which is why
it's a fairly useless concept if the size of the array is
unspecified.  (Many compilers get pointers to arrays wrong, which
is both unsurprising, since they're so confusing, and benign,
since they're so rarely useful.  I suspect that "pointer to
array" with the size of the array unspecified should elicit a
warning or error message.  pcc apparently pretends that the size
is zero, because incrementing the pointer adds 0 to it.  My C
interpreter says "missing array dimension.")  Although true pointers
to arrays could be made to work in this example, they would
require extradereferencing and punctuation, and generally only
confuse the issue.

                                            Steve Summit
                                            scs@adam.pika.mit.edu

P.S. This example helps to illustrate why, if something of type
"array of X" appears in a C expression, it is converted into a
pointer with type "pointer to X" (rather than "pointer to array
of X").  The resultant pointer "acts like" the original array;
i.e. after saying p=a, p[i] references a[i].