On Wed, Jan 9, 2019 at 11:01 AM Ramsay Jones
<ramsay@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> Ahem, so yes, both gcc and clang allow this (I just tested it), but
> my reading of the C99 standard lead me to believe that this is not
> valid C. (Not that it really matters).
Interesting. Looking at the kernel, we actually have a lot of them, ie
[torvalds@i7 linux]$ git grep 'struct.*= {[:space:]*}' | wc
1105 6766 81616
although admittedly we also have a fair amount of the { 0 } kind:
[torvalds@i7 linux]$ git grep 'struct.*= {[:space:]*0[:space:]*}' | wc
616 3722 50034
and I didn't realize that the empty one might not even be standard.
We used to (long long) ago have code like this:
/*
* gcc versions before ~2.95 have a nasty bug with empty initializers.
*/
#if (__GNUC__ > 2)
typedef struct { } spinlock_t;
#define SPIN_LOCK_UNLOCKED (spinlock_t) { }
#else
typedef struct { int gcc_is_buggy; } spinlock_t;
#define SPIN_LOCK_UNLOCKED (spinlock_t) { 0 }
#endif
but that was iirc not strictly about the empty initializers, it was
also about the structure itself being empty, and that confusing gcc.
But my memory might be dodgy on this.
(And in fact, that "before ~2.95" was apparently "egcs-1.1". So we're
talking truly old versions of gcc, this is pre-2000).
Linus
