From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <vdavydov.dev@gmail.com>
Date: Sat, 1 Dec 2018 20:20:57 +0300
From: Vladimir Davydov <vdavydov.dev@gmail.com>
Subject: Re: [PATCH v5 7/9] box: tune tuple_field_raw_by_path for indexed data
Message-ID: <20181201172057.ewodzt376fb4hp6r@esperanza>
References: <cover.1543229303.git.kshcherbatov@tarantool.org>
 <42e5b2becbb0ad788d2664f329ddfa9bd29f5922.1543229303.git.kshcherbatov@tarantool.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <42e5b2becbb0ad788d2664f329ddfa9bd29f5922.1543229303.git.kshcherbatov@tarantool.org>
To: Kirill Shcherbatov <kshcherbatov@tarantool.org>
Cc: tarantool-patches@freelists.org, kostja@tarantool.org
List-ID: <tarantool-patches.dev.tarantool.org>

On Mon, Nov 26, 2018 at 01:49:41PM +0300, Kirill Shcherbatov wrote:
> We don't need to parse tuple in tuple_field_raw_by_path if
> required field has been indexed. We do path lookup in field
> tree of JSON paths and return data by it's offset from field_map
> instead of whole tuple parsing.
> 
> Part of #1012
> ---
>  src/box/tuple_format.c     | 34 ++++++++++++++++++++++++----------
>  test/engine/tuple.result   |  5 +++++
>  test/engine/tuple.test.lua |  2 ++
>  3 files changed, 31 insertions(+), 10 deletions(-)
> 
> diff --git a/src/box/tuple_format.c b/src/box/tuple_format.c
> index 193d0d8..be89764 100644
> --- a/src/box/tuple_format.c
> +++ b/src/box/tuple_format.c
> @@ -956,15 +956,12 @@ tuple_field_raw_by_path(struct tuple_format *format, const char *tuple,
>  		goto error;
>  	switch(token.key.type) {
>  	case JSON_TOKEN_NUM: {
> -		int index = token.key.num;
> -		if (index == 0) {
> +		fieldno = token.key.num;
> +		if (fieldno == 0) {
>  			*field = NULL;
>  			return 0;
>  		}
> -		index -= TUPLE_INDEX_BASE;
> -		*field = tuple_field_raw(format, tuple, field_map, index);
> -		if (*field == NULL)
> -			return 0;
> +		fieldno -= TUPLE_INDEX_BASE;
>  		break;
>  	}
>  	case JSON_TOKEN_STR: {
> @@ -982,10 +979,9 @@ tuple_field_raw_by_path(struct tuple_format *format, const char *tuple,
>  			name_hash = field_name_hash(token.key.str,
>  						    token.key.len);
>  		}
> -		*field = tuple_field_raw_by_name(format, tuple, field_map,
> -						 token.key.str, token.key.len,
> -						 name_hash);
> -		if (*field == NULL)

After this patch tuple_field_raw_by_name and tuple_field_by_name are not
used. Please remove them in this patch.

> +		if (tuple_fieldno_by_name(format->dict, token.key.str,
> +					  token.key.len, name_hash,
> +					  &fieldno) != 0)
>  			return 0;
>  		break;
>  	}
> @@ -994,6 +990,24 @@ tuple_field_raw_by_path(struct tuple_format *format, const char *tuple,
>  		*field = NULL;
>  		return 0;
>  	}
> +	/* Optimize indexed JSON field data access. */
> +	assert(field != NULL);
> +	struct tuple_field *indexed_field =
> +		unlikely(fieldno >= tuple_format_field_count(format)) ? NULL :
> +		tuple_format_field_by_path(format,
> +					   tuple_format_field(format, fieldno),
> +					   path + lexer.offset,
> +					   path_len - lexer.offset);
> +	if (indexed_field != NULL &&
> +	    indexed_field->offset_slot != TUPLE_OFFSET_SLOT_NIL) {
> +		*field = tuple + field_map[indexed_field->offset_slot];
> +		return 0;
> +	}
> +
> +	/* No such field in index. Continue parsing JSON path. */
> +	*field = tuple_field_raw(format, tuple, field_map, fieldno);
> +	if (*field == NULL)
> +		return 0;

This code essentially does the same thing as tuple_field_by_part_raw.
Please reuse.