Re: [PATCH v2 2/2] oidmap: rework iterators to return typed pointer

[Jeff King <peff@peff.net>]

On Wed, Apr 08, 2020 at 12:33:46PM +0530, Abhishek Kumar wrote:

> 87571c3f (hashmap: use *_entry APIs for iteration, 2019-10-06) modified
> hashmap_iter_next() to return a hashmap_entry pointer instead of void
> pointer.
> 
> However, oidmap_iter_next() is unaware of the struct type containing
> oidmap_entry and explicitly returns a void pointer.
> 
> Rework oidmap_iter_next() to include struct type and return appropriate
> pointer. This allows for compile-time type checks.

Yes, I think returning a pointer to an oidmap_entry makes sense. And
then we get type safety, and anybody who wants embed an oidmap_entry can
use container_of() to get back to their original struct.

But...

> +/*
> + * Returns the next entry, or NULL if there are no more entries.
> + *
> + * The entry is of @type (e.g. "struct foo") and has a member of type struct
> + * oidmap_entry.
> + */
> +#define oidmap_iter_next(iter, type) \
> +	(type *) hashmap_iter_next(&(iter)->h_iter)

This cast is turning a hashmap_entry into whatever type the caller
passed in.  But it's doing it with a straight cast. We know that
hashmap_entry and oidmap_entry pointers are equivalent, but we don't
know where the oidmap_entry is with respect to the user's type.

I think oidmap_iter_next() should continue to be a function that returns
an oidmap_entry pointer (and use container_of_or_null() to get to it
from the hashmap_entry, even though we know the offset is 0).

And then the caller can either use container_of() to get to their
original struct, or we can provide a helper macro. See the difference
between hashmap_iter_first() and hashmap_iter_first_entry().

There's no hashmap_iter_next_entry(). There could be, but instead it
skipped straight to hashmap_for_each_entry(), which uses an offset
within the variable rather than the type. But likewise, we could add
oidmap_for_each_entry() here.

> diff --git a/t/helper/test-oidmap.c b/t/helper/test-oidmap.c
> index 0acf99931e..a28bf007a8 100644
> --- a/t/helper/test-oidmap.c
> +++ b/t/helper/test-oidmap.c
> @@ -96,7 +96,7 @@ int cmd__oidmap(int argc, const char **argv)
>  
>  			struct oidmap_iter iter;
>  			oidmap_iter_init(&map, &iter);
> -			while ((entry = oidmap_iter_next(&iter)))
> +			while ((entry = oidmap_iter_next(&iter, struct test_entry)))
>  				printf("%s %s\n", oid_to_hex(&entry->entry.oid), entry->name);

This works because "test_entry" has the oidmap_entry at the start.

But it wouldn't work with a struct where that wasn't the case, nor would
it provide any compile-time safety (because of the cast).

Note that if we do want to support that and get type safety (and I think
it is worth doing), oidmap_get() would need similar treatment (it
returns a void pointer, but it is really a pointer to an oidmap_entry).
And I guess oidmap_put() and oidmap_remove(), which returns pointers to
existing entries.

-Peff