On Mon, Dec 17, 2018 at 10:43:40PM -0700, Andreas Dilger wrote:
> I don't think that it is verboten to use binary flag values in an enum,
> if you explicitly specify the values, which is why I used "0x01", "0x02"
> to make it more clear these are binary values. IMHO, using a named enum
> is a good way to annotate the function parameters rather than a generic
> "int flag" parameter that is ambiguous unless you look at the function
> code to see what the values of "flag" might be.
I tend to only use enums in this kind of way:
enum classification_levels {
FOR_OFFICIAL_USE_ONLY,
CONFIDENTIAL,
SECRET,
TOP_SECRET,
};
I think the reason why I've never used it for type checking is because
for gcc and sparse, it doesn't work. For the below example, "gcc
-Wall foo.c" won't complain at all. Sparse complains only about the
"return a | b;" line, because we're combining two different enum
types. Sparse doesn't say boo that I passed EXT4_IGET_NORMAL where a
classification_levels, and secret where an ext4_iget_flags is
expected:
enum ext4_iget_flags {
EXT4_IGET_RESERVED = 0x00, /* just guessing, see further below */
EXT4_IGET_NORMAL = 0x01,
EXT4_IGET_HANDLE = 0x02
};
int combine(enum classification_levels a, enum ext4_iget_flags b)
{
return a | b;
}
int main(int argc, char **argv)
{
printf("%d\n", combine(EXT4_IGET_NORMAL, secret));
exit(1);
}
Then again, llvm does correctly complain, and at least for Android
configs, llvm will complain kernels correctly (although I'm not sure
enterprise distros trust LLVM just yet), and I do agree that it's
useful from a documentation perspective.
Cheers,
- Ted
