Re: [PATCH 3/4] Avoid a stupid 'const object should have initializer' warning

[Paolo Bonzini]

On 04/28/2010 12:00 AM, Bruno Haible wrote:
> Andreas Gruenbacher wrote:
> > diff --git a/lib/getdate.y b/lib/getdate.y
> > index 445865b..dcfe3cc 100644
> > --- a/lib/getdate.y
> > +++ b/lib/getdate.y
> > @@ -152,7 +152,7 @@ typedef struct
> >   #if HAVE_COMPOUND_LITERALS
> >   # define RELATIVE_TIME_0 ((relative_time) { 0, 0, 0, 0, 0, 0, 0 })
> >   #else
> > -static relative_time const RELATIVE_TIME_0;
> > +static relative_time const RELATIVE_TIME_0 = { 0, 0, 0, 0, 0, 0, 0 };
> >   #endif
> >
> >   /* Information passed to and from the parser.  */
>
> This patch makes the code less efficient: An allocation of n bytes in the
> 'data' segment causes n bytes to be read from disk. An allocation of
> n bytes in the 'bss' segment does not.

Recent and not-so-recent GCC (the behavior was introduced in 3.3) put a 
zero initializer in bss.  The difference between "int f;" and "int f = 
0;" is only whether the resulting symbol is "common" or not:

$ cat f.c
int f;
int g = 0;
int h = 1;
$ gcc f.c -o - -S
        .file   "f.c"
        .comm   f,4,4
.globl g
        .bss
        .align 4
        .type   g, @object
        .size   g, 4
g:
        .zero   4
.globl h
        .data
        .align 4
        .type   h, @object
        .size   h, 4
h:
        .long   1

... which doesn't even matter for static objects so the "= 0" doesn't 
even change the generated assembly:

$ cat g.c
static int f;
static int i = 0;
int x(){return i+f;}        /* avoid GCC optimizing out the variables */
int y(int j){return i=f=j;}
$ gcc g.c -o - -S
        .file   "f.c"
        [snip]
        .local  i
        .comm   i,4,4
        .local  f
        .comm   f,4,4

Paolo