Re: [PATCH 01/17] bitfield: Add non-constant field_{prep,get}() helpers

[Jakub Kicinski]

On Wed, 24 Nov 2021 09:03:24 +0100 Johannes Berg wrote:
> On Tue, 2021-11-23 at 15:49 -0800, Jakub Kicinski wrote:
> > > Indeed.
> > > 
> > > Also as I said in my other mail, the le32/be32/... variants are
> > > tremendously useful, and they fundamentally cannot be expressed with the
> > > FIELD_GET() or field_get() macros. IMHO this is a clear advantage to the  
> > 
> > Can you elaborate?  
> 
> Well, the way I see it, the only advantage of FIELD_GET() is that it
> will auto-determine the type (based on the mask type.) This cannot work
> if you need be/le conversions, because the be/le type annotations are
> invisible to the compiler.
> 
> So obviously you could write a BE32_FIELD_GET(), but then really that's
> equivalent to be32_get_bits() - note you you have to actually specify
> the type in the macro name. I guess in theory you could make macros
> where the type is an argument (like FIELD_GET_TYPE(be32, ...)), but I
> don't see how that gains anything.

Ah, that's what you meant! Thanks for spelling it out.

FWIW I never found the be/le versions useful. Most of the time the data
comes from bus accessors which swap or is unaligned so you have to do
be/le_get_unaligned, which swaps. Plus if you access/set multiple
fields you'd swap them one by one which seems wasteful.

> > > typed versions, and if you ask me we should get rid of the FIELD_GETand
> > > FIELD_PREP entirely - difficult now, but at least let's not propagate
> > > that?  
> > 
> > I don't see why.  
> 
> Just for being more regular, in the spirit of "there's exactly one
> correct way of doing it" :)

Right now it seems the uppercase macros are more prevalent.

Could just be because of the way the "swapping ones" are defined.