Merge lp://qastaging/~vanvugt/compiz-core/fix-940066 into lp://qastaging/compiz-core

Gah, truly bitten by this I see :/

It seems odd to use "return memcmp (a, b, sizeof (double) * 6);" without at least a compile-time assert that sizeof (decor_matrix_t) == sizeof (double) * 6).

#define STATIC_ASSERT(condition, label) _Static_assert(condition, #label)
STATIC_ASSERT(sizeof (decor_matrix_t) == sizeof (double) * 6, decor_matrix_t is 6 doubles);

That assertion would be incorrect, which is why I avoided sizeof (decor_matrix_t).

Structure alignment rules guarantee that decor_matrix_t contains 6 doubles tightly packed at offset 0. However I do not think they guarantee how much padding there is after that. Thus, you cannot assume sizeof (decor_matrix_t) == 6 * sizeof (double).

While I agree there is no C standard requirement that there is no tail
padding, the universal implementation of tail padding is for the size
of a struct to reach the smallest possible multiple of the alignment
required by the contained types.  (This is required by every ABI that
I've seen.)
If you feel a need to be paranoid I don't believe one can show that
all bits in a double are significant.

----- Original Message -----
From: <email address hidden>
To:"Daniel van Vugt"
Cc:
Sent:Sat, 25 Feb 2012 03:19:16 -0000
Subject:Re: [Merge] lp:~vanvugt/compiz-core/fix-940066 into
lp:compiz-core

 That assertion would be incorrect, which is why I avoided sizeof
(decor_matrix_t).

 Structure alignment rules guarantee that decor_matrix_t contains 6
doubles tightly packed at offset 0. However I do not think they
guarantee how much padding there is after that. Thus, you cannot
assume sizeof (decor_matrix_t) == 6 * sizeof (double).
 --
https://code.launchpad.net/~vanvugt/compiz-core/fix-940066/+merge/94502
[1]
 Your team Compiz Maintainers is subscribed to branch lp:compiz-core.

Links:
------
[1]
https://code.launchpad.net/~vanvugt/compiz-core/fix-940066/+merge/94502

I think all the bits in a double are significant to an equality operation:
http://en.wikipedia.org/wiki/Double_precision_floating-point_format

But I'm happy to remove the memcmp for argument's sake.