Messages in this thread | | | From | Nick Desaulniers <> | Date | Tue, 14 Jun 2022 11:08:33 -0700 | Subject | Re: [PATCH] [RFC] Initialization of unused function parameters |
| |
On Tue, Jun 14, 2022 at 10:24 AM Linus Torvalds <torvalds@linux-foundation.org> wrote: > > On Tue, Jun 14, 2022 at 10:11 AM Nick Desaulniers > <ndesaulniers@google.com> wrote: > > > > Maybe a new function parameter attribute would be nice? > > Right, exactly something like this seems reasonable. > > > #define __must_init __attribute__((must_init)) > > int init (int * __must_init x) { > > // ^ warning: function parameter x marked '__attribute__((must_init))' > > not unconditionally initialized > > if (stars_dont_align) { > > return -42; > > } > > *x = 42; > > return 0; > > } > > void foo (void) { int x; init(&x); /* use of x without fear */ }
Thinking more about your snprintf example which is potentially more costly than initializing just one value...here's another case for us to consider.
int maybe_init (char* buf) { if (stars_align) return -42; buf[42] = 0; return 0; }
char foo (void) { char buf [PATH_MAX]; maybe_init(buf); return char[42]; }
I'm thinking the attribute would have to go on the pointed to type not the pointer, i.e. `__must_init char *` not `char * __must_init`. Similarly, you'd need to unconditionally initialize all of buf which might be painful. __must_init would not give you the specificity to differentiate between "this whole buffer must be initialized" vs "only index 42 need be initialized." I _think_ that's fine, perhaps "only index 42 need be initialized" is YAGNI and you could just _not_ use __must_init for such a case.
One thing I'm curious about; if you have an aggregate in C (struct or array) and don't fully initialize the whole object, just members/sub-objects, but only use those I assume that's not UB? (Which is what my maybe_init example does). I think that's fine.
Another thing that makes me uncertain about my maybe_init example above is decay-to-pointer, and the compiler's ability to track things like __builtin_object_size precisely (or across translation units); Kees is having a dog of a time with __builtin_object_size of structs that contain flexible array members for example. If a function accepts a `__must_init struct foo*`, can we know if we were passed an array of struct foo* vs just one? What if `struct foo` is an opaque type; then the callee can't verify that all members of the struct have been initialized (at least designated initialized wouldn't work; memcpy should though...can you get the sizeof an opaque type though?)
But maybe I'm getting ahead of myself and am just describing good unit tests when building such a feature. Probably worth prototyping to get a sense of the ergonomics and limitations. But it doesn't sound too controversial, which is probably good enough to get started. I'm sure Alexander and team will have additional ideas or proposals for achieving the same outcome. -- Thanks, ~Nick Desaulniers
| |