Re: [PATCH] eal: provide rte attribute macro for GCC attribute

[Thomas Monjalon <thomas@monjalon.net>]

20/02/2024 19:06, Tyler Retzlaff:
> On Sun, Feb 18, 2024 at 04:31:50PM +0100, Thomas Monjalon wrote:
> > 18/02/2024 15:51, Mattias Rönnblom:
> > > On 2024-02-18 13:24, Thomas Monjalon wrote:
> > > > 15/02/2024 23:20, Tyler Retzlaff:
> > > >> Provide a new macro __rte_attribute(a) that when directly used
> > > >> compiles to empty for MSVC and to __attribute__(a) when using GCC/LLVM.
> > > >>
> > > >> Replace direct use of __attribute__ in __rte_xxx macros where there is
> > > >> existing empty expansion of the macro for MSVC allowing removal of
> > > >> repeated #ifdef RTE_TOOLCHAIN_MSVC per macro to expand empty.
> > > > 
> > > > I'm not sure it makes sense.
> > > > I prefer seeing clearly what is empty with MSVC.
> > > 
> > > Anything __rte_attribute() is empty on MSVC. You could rename it 
> > > __rte_attribute_ignored_by_msvc() for clarity.
> > 
> > Yes it would bring more clarity.
> > But I still prefer #ifdef which may work with more compilers.
> > 
> > > One could note that on the ignore list are things like "may_alias" and 
> > > "packed", so whatever comes out of a MSVC build should not be expected 
> > > to actually work.
> > > 
> > > Unless I'm missing something, for all the attributes that will have 
> > > MSVC-propriety equivalent, the usage pattern would have to change, since 
> > > the syntax is different in incompatible ways.
> > > 
> > > Wouldn't it be better to ask the MSVC team to add support GCC 
> > > attributes? ICC and LLVM managed, so why not Microsoft. Then you would 
> > > solve this issue for all Open Source projects, not only DPDK.
> > 
> > We can expect MSVC to improve.
> > That's another reason why I prefer to keep #ifdef to keep track easily.
> 
> MSVC is committed to provide functionality where something simply cannot
> be done at all with their toolset and standard C.

That makes sense.

> They will not make changes to their toolset for functionality they already have.

So you mean there is already a way to insert zero-size markers in a struct?