Re: function prototype element ordering

From: Jani Nikula
Date: Wed Sep 22 2021 - 07:19:51 EST


On Wed, 22 Sep 2021, Alexey Dobriyan <adobriyan@xxxxxxxxx> wrote:
> On Tue, Sep 21, 2021 at 07:25:53PM -0700, Kees Cook wrote:
>> On Tue, Sep 21, 2021 at 04:45:44PM -0700, Joe Perches wrote:
>> > On Tue, 2021-09-21 at 16:37 -0700, Kees Cook wrote:
>> > > On Fri, Sep 10, 2021 at 10:23:48AM -0700, Linus Torvalds wrote:
>> > > > On Thu, Sep 9, 2021 at 8:10 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>> > > > >
>> > > > > +__alloc_size(1)
>> > > > >  extern void *vmalloc(unsigned long size);
>> > > > [...]
>> > > >
>> > > > All of these are added in the wrong place - inconsistent with the very
>> > > > compiler documentation the patches add.
>> > > >
>> > > > The function attributes are generally added _after_ the function,
>> > > > although admittedly we've been quite confused here before.
>> > > >
>> > > > But the very compiler documentation you point to in the patch that
>> > > > adds these macros gives that as the examples both for gcc and clang:
>> > > >
>> > > > + * gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-alloc_005fsize-function-attribute
>> > > > + * clang: https://clang.llvm.org/docs/AttributeReference.html#alloc-size
>> > > >
>> > > > and honestly I think that is the preferred format because this is
>> > > > about the *function*, not about the return type.
>> > > >
>> > > > Do both placements work? Yes.
>> > >
>> > > I'm cleaning this up now, and have discovered that the reason for the
>> > > before-function placement is consistency with static inlines. If I do this:
>> > >
>> > > static __always_inline void * kmalloc(size_t size, gfp_t flags) __alloc_size(1)
>> > > {
>> > > ...
>> > > }
>> > >
>> > > GCC is very angry:
>> > >
>> > > ./include/linux/slab.h:519:1: error: attributes should be specified before the declarator in a function definition
>> > >   519 | static __always_inline void *kmalloc_large(size_t size, gfp_t flags) __alloc_size(1)
>> > >       | ^~~~~~
>> > >
>> > > It's happy if I treat it as a "return type attribute" in the ordering,
>> > > though:
>> > >
>> > > static __always_inline void * __alloc_size(1) kmalloc(size_t size, gfp_t flags)
>> > >
>> > > I'll do that unless you have a preference for somewhere else...
>> >
>> > _please_ put it before the return type on a separate line.
>> >
>> > [__attributes]
>> > [static inline const] <return type> function(<args...>)
>>
>> Somehow Linus wasn't in CC. :P
>>
>> Linus, what do you want here? I keep getting conflicting (or
>> uncompilable) advice. I'm also trying to prepare a patch for
>> Documentation/process/coding-style.rst ...
>>
>> Looking through what was written before[1] and through examples in the
>> source tree, I find the following categories:
>>
>> 1- storage class: static extern inline __always_inline
>> 2- storage class attributes/hints/???: __init __cold
>> 3- return type: void *
>> 4- return type attributes: __must_check __noreturn __assume_aligned(n)
>> 5- function attributes: __attribute_const__ __malloc
>> 6- function argument attributes: __printf(n, m) __alloc_size(n)
>>
>> Everyone seems to basically agree on:
>>
>> [storage class] [return type] [return type attributes] [name]([arg1type] [arg1name], ...)
>>
>> There is a lot of disagreement over where 5 and 6 should fit in above. And
>> there is a lot of confusion over 4 (mixed between before and after the
>> function name) and 2 (see below).
>>
>> What's currently blocking me is that 6 cannot go after the function
>> (for definitions) because it angers GCC (see quoted bit above), but 5
>> can (e.g. __attribute_const__).
>>
>> Another inconsistency seems to be 2 (mainly section markings like
>> __init). Sometimes it's after the storage class and sometimes after the
>> return type, but it certainly feels more like a storage class than a
>> return type attribute:
>>
>> $ git grep 'static __init int' | wc -l
>> 349
>> $ git grep 'static int __init' | wc -l
>> 8402
>>
>> But it's clearly positioned like a return type attribute in most of the
>> tree. What's correct?
>>
>> Regardless, given the constraints above, it seems like what Linus may
>> want is (on "one line", though it will get wrapped in pathological cases
>> like kmem_cache_alloc_node_trace):
>>
>> [storage class] [storage class attributes] [return type] [return type attributes] [function argument attributes] [name]([arg1type] [arg1name], ...) [function attributes]
>>
>> Joe appears to want (on two lines):
>>
>> [storage class attributes] [function attributes] [function argument attributes]
>> [storage class] [return type] [return type attributes] [name]([arg1type] [arg1name], ...)
>>
>> I would just like to have an arrangement that won't get NAKed by
>> someone. ;) And I'm willing to document it. :)
>
> Attributes should be on their own line, they can be quite lengthy.
>
> __attribute__((...))
> [static] [inline] T f(A1 arg1, ...)
> {
> ...
> }
>
> There will be even more attributes in the future, both added by
> compilers and developers (const, pure, WUR), so let's make "prototype lane"
> for them.
>
> Same for structures:
>
> __attribute__((packed))
> struct S {
> };
>
> Kernel practice of hiding attributes under defines (__ro_after_init)
> breaks ctags which parses the last identifier before semicolon as object
> name. Naturally, it is ctags bug, but placing attributes before
> declaration will autmatically unbreak such cases.

git grep seems to suggest __packed is preferred over
__attribute__((packed)), and at the end of the struct declaration
instead of at front:

struct S {
/* ... */
} __packed;

And GNU Global handles this just fine. ;)


BR,
Jani.

--
Jani Nikula, Intel Open Source Graphics Center