Re: [Tarantool-patches] [PATCH v3 02/10] xrow: introduce a PROMOTE entry

[Vladislav Shpilevoy via Tarantool-patches <tarantool-patches@dev.tarantool.org>]

I appreciate the work you did here!

See 2 comments below.

> diff --git a/src/box/xrow.c b/src/box/xrow.c
> index cc8e43ed4..5d515ce92 100644
> --- a/src/box/xrow.c
> +++ b/src/box/xrow.c
> @@ -884,28 +884,63 @@ xrow_encode_dml(const struct request *request, struct region *region,
>  	return iovcnt;
>  }
>  
> -void
> -xrow_encode_synchro(struct xrow_header *row,
> -		    struct synchro_body_bin *body,
> -		    const struct synchro_request *req)
> +static void
> +xrow_encode_synchro_body(struct synchro_body_bin *body,
> +		         const struct synchro_request *req)
>  {
>  	/*
> -	 * A map with two elements. We don't compress
> +	 * A map with two or three elements. We don't compress
>  	 * numbers to have this structure constant in size,
>  	 * which allows us to preallocate it on stack.
>  	 */
> -	body->m_body = 0x80 | 2;
> +	body->m_body = 0x80 | (req->type == IPROTO_PROMOTE ? 3 : 2);
>  	body->k_replica_id = IPROTO_REPLICA_ID;
>  	body->m_replica_id = 0xce;
>  	body->v_replica_id = mp_bswap_u32(req->replica_id);
>  	body->k_lsn = IPROTO_LSN;
>  	body->m_lsn = 0xcf;
>  	body->v_lsn = mp_bswap_u64(req->lsn);
> +}
> +
> +void
> +xrow_encode_synchro(struct xrow_header *row,
> +		    struct synchro_body_bin *body,
> +		    const struct synchro_request *req)
> +{
> +	assert(req->type == IPROTO_CONFIRM || req->type == IPROTO_ROLLBACK);
> +
> +	xrow_encode_synchro_body(body, req);
>  
>  	memset(row, 0, sizeof(*row));
> +	row->type = req->type;
> +	row->body[0].iov_base = body;
> +	row->body[0].iov_len = sizeof(*body);
> +	row->bodycnt = 1;
> +}
> +
> +static inline void
> +xrow_encode_promote_body(struct promote_body_bin *body,
> +			 const struct synchro_request *req)

1. I would propose to inline it. It is used in a single place,
and now it looks like if we would have more than 1 place where
we would need the promote body.

But more interestingly, it looks you could keep it a single
function xrow_encode_synchro. Although we wouldn't be able to
have a PACKED struct with predefined fields. Not a big deal
anyway.

The reasoning is similar to xrow_encode_dml(). It also uses
a single struct request for all kinds of DML, and conditionally
encodes the non-zero fields. I think your case is the same. It
would simplify some code, and remove branching from other
places. For example, from txn_limbo_write_synchro(), where you
branch between PROMOTE and non-PROMOTE when decide what to encode.
We even had the same issue when tried to encode CONFIRM and
ROLLBACK via separate functions.

> +{
> +	xrow_encode_synchro_body(&body->base, req);
> +
> +	body->k_term = IPROTO_TERM;
> +	body->m_term = 0xcf;
> +	body->v_term = mp_bswap_u64(req->term);
> +}
> +
>  
> +void
> +xrow_encode_promote(struct xrow_header *row, struct promote_body_bin *body,
> +		    const struct synchro_request *req)
> +{
> +	assert(req->type == IPROTO_PROMOTE);
> +
> +	xrow_encode_promote_body(body, req);
> +
> +	memset(row, 0, sizeof(*row));
>  	row->type = req->type;
> -	row->body[0].iov_base = (void *)body;
> +	row->body[0].iov_base = body;

2. Unnecessary change. But I don't mind, up to you.

>  	row->body[0].iov_len = sizeof(*body);
>  	row->bodycnt = 1;
>  }