Fix CTASSERT issue in a more clean way

[ngie]

Summary

• Replace all CTASSERT macro instances with static_assert's
• Localize all static_assert's need their relevant structures being tested

MFC after: 1 week
X-MFC with: r302364
Reviewed by: grehan (maintainer)
Submitted by: ed
Sponsored by: EMC / Isilon Storage Division

[ed]

pci_emul.c

758

We originally had to place these CTASSERT()s in source files as opposed to header files, as our version of the macro was implemented with __LINE__ instead of __COUNTER__. Could you please move these compile-time assertions next to the actual structure declarations (i.e. into pci_emul.h)? That way they're less likely to be removed accidentally due to refactoring, tested for every use, etc.

Also: why the indirection? _Static_assert() is part of the C standard and supported by FreeBSD 9 and higher, even if you're using an ancient compiler (GCC 4.2). Strongly consider removing CTASSERT() altogether. There is no use for it. Looking at pci_emul.h, it already includes <assert.h>, so you can even just use static_assert().

[ngie]

pci_emul.c

758

Done.

[ngie]

• Replace all CTASSERT macro instances with static_assert's
• Localize all static_assert's need their relevant structures being tested

[grehan]

Thanks, much nicer.

[ed]

Looks good! For bonus points, you could also customise the error strings to something like "Required by PCI standard abc paragraph xyz", but I don't care about that too strongly. Future versions of _Static_assert() may even make the error message optional anyway:

http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3928.pdf

[ngie]

In D7130#148571, @ed wrote:

Looks good! For bonus points, you could also customise the error strings to something like "Required by PCI standard abc paragraph xyz", but I don't care about that too strongly. Future versions of _Static_assert() may even make the error message optional anyway:

http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3928.pdf

"" seems to be a secondary option that's used elsewhere in the tree. I kind of lack the patience to look up the PCI standard spec that states that the structure must be packed in a particular way :P, and eventually this will become optional.. it's just not available in this particular standard version (and I don't want to go chasing down more issues right now.. this works "well enough" for ^/stable/11).

[ngie]

rS302373