[sdcc-devel] [ sdcc-Feature Requests-2991122 ] C99 designated initializer support

[SourceForge.net <no...@so...>]

Feature Requests item #2991122, was opened at 2010-04-22 17:39
Message generated for change (Comment added) made by cananian
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=350599&aid=2991122&group_id=599

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: None
Group: None
Status: Open
Priority: 5
Private: No
Submitted By: C. Scott Ananian (cananian)
Assigned to: Nobody/Anonymous (nobody)
Summary: C99 designated initializer support

Initial Comment:
Add support for C99 "designated initializers", like:

struct {
   int x, y;
} point = { .y = 3, .x = 2 };

----------------------------------------------------------------------

>Comment By: C. Scott Ananian (cananian)
Date: 2010-04-22 21:18

Message:
I attached an updated patch which honors initializer order, as the spec
says we ought.

For yr reference, here's my test case; I've examined the assembly output
for mcs51, pic, and pic16 and convinced myself it's correct:

/** Tests for designated initializers */
struct point { int x, y; };
//char __code simpletest[] = "foo";
int a1[6] = { [4] = 29, 30, [2] = 15 };
int __code a2[6] = { [4] = 29, 30, [2] = 15 };
unsigned char widths1[] = { [2] = 2, [20] = 3, [2] = 1 };
unsigned char __code widths2[] = { [2] = 2, [20] = 3, [2] = 1 };

int foo(int xvalue, int yvalue) {
    struct point p = { .y = yvalue, .x = xvalue };
    return p.x + p.y;
}
struct point p1 = { .y = 2, .x = 1 };
struct point __code p2 = { .y = 2, .x = 1 };

struct point pa1[] = { { 1 } };

union foo { int i; float d; };
union foo f1 = { .d = 4 };
union foo __code f2 = { .d = 4 };
union foo __code f3 = { .i = 4 };

int whitespace1[256]
       = { [' '] = 1, ['\t'] = 1, ['\h'] = 1,
           ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 };
int __code whitespace2[256]
       = { [' '] = 1, ['\t'] = 1, ['\h'] = 1,
           ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 };

struct test {
    int x;
    int y1 : 2;
    int    : 2;
    int y2 : 3;
    int z;
};
struct test stest1 = { .y1 = 1, 3, 4, .x = 2, .z = 5 };
struct test __code stest2 = { .y1 = 1, 3, 4, .x = 2, .z = 5 };

const char *foo1 = "bar";
char * __code foo2 = "bat";

struct point ptarray1[10] = { [2].y = 5, [2].x = 6, [0].x = 7 };
struct point __code ptarray2[10] = {
    [2].y = 5,
    [2].x = 6,
    [0].x = 7,
    [2].x = 8 };

void main(void)
{
}


----------------------------------------------------------------------

Comment By: C. Scott Ananian (cananian)
Date: 2010-04-22 18:43

Message:
The standard seems to use "designation" to mean "a sequence of one or more
designators".  I tried to use this definition in the code as well.

Good catch on the order of initialization.  My original patches followed
the initializer order more closely, but I reworked it to try to remove some
duplicate code.  I can go back to the version that respects the initializer
order.

We do in fact omit initialization for holes; the code you're looking at is
for 'code' segment initializations, where we emit a static set of '.db'
statements -- obviously we can't skip holes here.

----------------------------------------------------------------------

Comment By: Philipp Krause (spth)
Date: 2010-04-22 18:31

Message:
- Should the term "designation initializer" be used in place of "designated
initalizer"? The first term is the one used in the C standard, while the
second oine is used in the rationale.

- Does the standard mandate that the initializations occour in a given
order? E.g. we have a volatile struct, inilialized using a designation
initalizer that lists the second member before the second do we have to
really initialize the second member first? Verse 1670-1672 seems to suggest
so to me (who doesn't really know anything about designation
initializers).

- The has a comment "If this is a hole, substitute an appropriate
initializer." Unless the object is static, can we increase efficiency by
not doing the initialization for these holes instead of initializing them
to 0?

Philipp


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=350599&aid=2991122&group_id=599