Re: [PATCH v2 18/19] OFFSETOF_VAR macro to simplify hashmap iterators

From: Junio C Hamano <gitster@pobox.com>
Date: Sep 30, 2019, 16:58 UTC
Message-ID: <xmqqy2y5hgau.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <20190924010324.22619-19-e@80x24.org>

Eric Wong <e@80x24.org> writes:

> While we cannot rely on a `__typeof__' operator being portable
> to use with `offsetof'; we can calculate the pointer offset
> using an existing pointer and the address of a member using
> pointer arithmetic.

> +/*
> + * like offsetof(), but takes a pointer to type instead of the type

It actually takes "a pointer to a variable of the type". I had to read the above twice to guess what is going on.

Show 5 quoted lines

> + * @ptr is subject to multiple evaluation since we can't rely on TYPEOF()
> + */
> +#define OFFSETOF_VAR(ptr, member) \
> +	((uintptr_t)&(ptr)->member - (uintptr_t)(ptr))
> +

This unfortunately has funny interactions with using uninitialized variables. "make CC=clang config.o" gives something like this:

config.c:1944:50: error: variable 'entry' is uninitialized when used here
      [-Werror,-Wuninitialized]
        hashmap_for_each_entry(&cs->config_hash, &iter, entry,
                                                        ^~~~~
./hashmap.h:453:20: note: expanded from macro 'hashmap_for_each_entry'
                                                OFFSETOF_VAR(var, member)); \
                                                             ^~~
./git-compat-util.h:1346:16: note: expanded from macro 'OFFSETOF_VAR'
        ((uintptr_t)&(ptr)->member - (uintptr_t)(ptr))
                      ^~~
./hashmap.h:445:61: note: expanded from macro 'hashmap_iter_first_entry_offset'
        container_of_or_null_offset(hashmap_iter_first(map, iter), offset)
                                                                   ^~~~~~
config.c:1939:34: note: initialize the variable 'entry' to silence this warning
        struct config_set_element *entry;
                                        ^
                                         = NULL

Personally, I feel the workaround suggested by the compiler is just as bogus as the warning itself X-<. Casting NULL as if it were a pointer to an object of type typeof(*ptr), treating it as if it is a valid address, and doing raw address arith is just as bogus as doing the same raw address arith on an uninitialized ptr, isn't it?

Anyway...