Re: [PATCH 2/3] mem-pool: use more standard initialization and finalization

From: Junio C Hamano <gitster@pobox.com>
Date: Aug 14, 2020, 04:38 UTC
Message-ID: <xmqqpn7tzvtf.fsf@gitster.c.googlers.com>
In-Reply-To: <f13a52055cd975d457e0593cbabb70897e78024b.1597374135.git.gitgitgadget@gmail.com>

"Elijah Newren via GitGitGadget" <gitgitgadget@gmail.com> writes:

Show 9 quoted lines

> written.  mem_pool_init() does essentially the following (simplified
> for purposes of explanation here):
>
>     void mem_pool_init(struct mem_pool **pool...)
>     {
>         *pool = xcalloc(1, sizeof(*pool));
>
> It seems weird to require that mem_pools can only be accessed through a
> pointer.

Yup, if the _init() were to also allocate, I would expect it to be more like

	struct mem_pool *mem_pool_create(...)
	{
		struct mem_pool *pool = xcalloc(1, sizeof(*pool));
		...
		return pool;
	}

It also is OK to let the caller pass uninitialized region of memory, which is how we usually arrange _init() to work. It seems that that is the approach this patch takes.

Show 20 quoted lines

> -void mem_pool_init(struct mem_pool **mem_pool, size_t initial_size)
> +void mem_pool_init(struct mem_pool *mem_pool, size_t initial_size)
>  {
> -	struct mem_pool *pool;
> -
> -	if (*mem_pool)
> -		return;
> -
> -	pool = xcalloc(1, sizeof(*pool));
> -
> -	pool->block_alloc = BLOCK_GROWTH_SIZE;
> +	mem_pool->mp_block = NULL;
> +	mem_pool->pool_alloc = 0;
> +	mem_pool->block_alloc = BLOCK_GROWTH_SIZE;
>  
>  	if (initial_size > 0)
> -		mem_pool_alloc_block(pool, initial_size, NULL);
> -
> -	*mem_pool = pool;
> +		mem_pool_alloc_block(mem_pool, initial_size, NULL);

It used to be that this function both knew and took control of all the bits in *pool memory by using xcalloc(). Any field the function assigned to of course got explicitly the value the function wanted it to have, and all other fields were left to 0.

It may happen to be still the case (i.e. the assignments we see in this function cover all the fields defined), but don't we need some provision to make sure it will hold to be true in the future?

Starting it with "memset(pool, 0, sizeof(*pool)" would be one way.

You'd standardize to s/mem_pool/pool/ in [3/3]; shouldn't this be written with pool to begin with, instead of reintroducing mem_pool that is of different type from the original?

Show 5 quoted lines

> -	if (!*pool_ptr)
> -		mem_pool_init(pool_ptr, 0);
> +	if (!*pool_ptr) {
> +		*pool_ptr = xmalloc(sizeof(**pool_ptr));
> +		mem_pool_init(*pool_ptr, 0);

This one gives an uninitialized chunk of memory to the _init(); an example of the caller that the earlier comment may matter.

Show 10 quoted lines

> +	istate->ce_mem_pool = xmalloc(sizeof(*istate->ce_mem_pool));
>  	if (istate->version == 4) {
> -		mem_pool_init(&istate->ce_mem_pool,
> +		mem_pool_init(istate->ce_mem_pool,
>  				estimate_cache_size_from_compressed(istate->cache_nr));
>  	} else {
> -		mem_pool_init(&istate->ce_mem_pool,
> +		mem_pool_init(istate->ce_mem_pool,
>  				estimate_cache_size(mmap_size, istate->cache_nr));
>  	}

Likewise.

Thanks.