Re: [cocci] mm/slab: prepare for struct page conversion by spatch

[Julia Lawall <julia.lawall@inria.fr>]

On Sun, 7 Nov 2021, Vlastimil Babka wrote:

> On 11/7/21 13:50, Markus Elfring wrote:
> >>> * Will any guidance help to find nicer solutions?
> >>
> >> Luis already proposed a perfectly reasonable solution.
> >
> > His proposal can look a bit too terse (for affected implementation details),
> > can't it?
> > https://sympa.inria.fr/sympa/arc/cocci/2021-11/msg00017.html
> > https://lore.kernel.org/cocci/YYVZva1hfwanXdyB@bombadil.infradead.org/
>
> I don't mind terse, I wasn't asking for complete fixup of my patches,
> but for hints, and I got plenty from all of you, thanks! Will try to
> apply everything tomorrow.
>
> > @@
> > @@
> > -struct page
> > +struct slab
>
> The above is itself incomplete to go from "struct page *page" to "struct
> slab *slab", but will results in "struct slab *page".
>
> So my attempts included:
> - struct page *page
> + struct slab *slab
>
> But that for some reason resulted in spatch reporting (IIRC) syntax

A variable declaration needs a semicolon.

> errors, so I had to change it to this:
>
> - struct page *page;
> + struct slab *slab;

If you first rename struct page as strict slab and then do

struct slab *
- page
+ slab
 ;

it might work even for the multiple declaration case.

julia

> And that worked in most cases, but didn't handle the multiple variables
> being declared nor declaration+assignment.
>
> But I assume Luis' suggestion is supposed to work together with the rule
> that does simply -page +slab. I'll see how it works out, and also
> Julia's suggestion "just have one rule to change all occurences of page
> to slab, and then have individual rules to change eg page_address(slab)
> to slab_address(slab)."
>
> Thanks,
> Vlastimil
>