On Wed, Nov 29, 2023 at 12:25:37PM +0200, Ville Syrjälä wrote: > On Wed, Nov 29, 2023 at 12:12:59PM +0200, Pekka Paalanen wrote: > > On Tue, 28 Nov 2023 15:49:08 +0200 > > Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx> wrote: > > > > > Should we perhaps start to use the (arguably hideous) > > > - void f(struct foo *bar) > > > + void f(struct foo bar[static 1]) > > > syntax to tell the compiler we don't accept NULL pointers? > > > > > > Hmm. Apparently that has the same problem as using any > > > other kind of array syntax in the prototype. That is, > > > the compiler demands to know the definition of 'struct foo' > > > even though we're passing in effectively a pointer. Sigh. > > > > > > __attribute__((nonnull)) ? > > I guess that would work, though the syntax is horrible when > you need to flag specific arguments. I played around with this a bit (blindly cocci'd tons of drm and i915 function declarations with the nonnull attribute) and it's somewhat underwhelming unfortunately. It will trip only if the compiler is 100% sure you're passing in a NULL. There is no way to eg. tell the compiler that a function can return a NULL and thus anything coming from it should be checked by the caller before passing it on to something with the nonnull attribute. And I suppose error pointers would also screw that idea over anyway. Additionally the NULL device checks being being done in the drm_err/dbg macros trip this up left right and center. And hiding that check inside a function (instead of having it in the macro) is also ruined by the fact that we apparently pass different types of pointers to these macros :( Generics could be used to sort out that type mess I suppose, or the code that passes the wrong type (DSI code at least) should just be changed to not do that. But not sure if there's enough benefit to warrant the work. -- Ville Syrjälä Intel
