Re: Proposal: block-based vector allocator

[Paul Eggert]

Thanks for working on this.  Here are a few low-level comments.

On 06/07/2012 03:03 AM, Dmitry Antipov wrote:
> +#define VECTOR_FREE_LIST_FLAG                                \
> +  ((~0) & ~(ARRAY_MARK_FLAG | PSEUDOVECTOR_FLAG              \

There's no need for the "(~0) &".

> +  zero_vector = (struct Lisp_Vector *)
> +    pure_alloc (header_size, Lisp_Vectorlike);

No need for the cast; "zero_vector = pure_alloc (header_size, Lisp_Vectorlike)"
will work just fine.  In general, it's better to avoid casts when they're
not needed, as casts are error-prone.  (Admittedly the code doesn't always
follow this rule, but this was partly motivated by pre-c89 compilers.)

> +#define ADVANCE(v,nbytes) (struct Lisp_Vector *) ((char *) (v) + (nbytes))

Space after comma (here and elsewhere).
Need to parenthesize the right-hand-side.

> +#define VBLOCK_BYTES_MIN (vroundup (sizeof (struct Lisp_Vector)))
> +
> +/* Size of the largest vector allocated from block.  */
> +
> +#define VBLOCK_BYTES_MAX                                     \
> +  (vroundup ((VECTOR_BLOCK_BYTES / 2) - sizeof (Lisp_Object)))

No need to parenthesize either right-hand side.

> +  block = xmalloc (sizeof (struct vector_block));
> +  if (!block)
> +    memory_full (sizeof (struct vector_block));

The last two lines are not needed, as 'xmalloc'
already does overflow checking.