🚀 go-pugleaf

Steve Clamage (steve@taumet.com) wrote:
: msb@sq.sq.com (Mark Brader) writes:
: 
: >In particular, it could not modify padding at the end of a struct,
: >precisely *because* it would break the struct hack.
: 
: We seem to have some circular reasoning here.
: 
: I proposed an implemention which I claimed was conforming on which
: the struct hack would not work; therefore the struct hack is illegal.

 ...

: For example, if the struct hack is strictly conforming, what about this:
: 
: 	struct S {
: 	    double a[2];
: 	    double b;
: 	} s;
: 	s.a[2] = 1.0;
: 
: The array reference exceeds the declared bounds of s.a, but not the
: bounds of object s, because sizeof(s) can't be smaller than
: sizeof(a)+sizeof(b).  If a compiler can't add padding to the end of
: the struct for other than alignment, then surely it can't add padding
: between a and b, because it can't be required for alignment.  Therefore
: s.a[2] must be the same as s.b.

First off, the compiler is free to add padding for whatever reasons it
wants.. but it must not depend on the value of this padding information.
Second, in your example, a[2] would be equal to 1.0.  It says nothing 
about the value of b.  Because the contents of b may or may not be changed,
it's value is not definite.

--
  Trent Tobler