Re: [tarantool-patches] Re: [PATCH v1 4/4] box: refactor tuple_init_field_map to use bitmap

[Vladimir Davydov <vdavydov.dev@gmail.com>]

On Sat, Dec 29, 2018 at 03:58:34PM +0300, Kirill Shcherbatov wrote:
> Refactored tuple_init_field_map to fill a local bitmap and

It's not refactoring, it's a rework.

> compare it with template required_fields bitmap containing
> information about required fields. Each field is mapped to
> bitmap with field:id - unique field identifier.
> This approach to check the required fields will work even after
> the introduction of JSON paths, when the field tree becomes
> multilevel.
> 
> Needed for #1012

> @@ -440,15 +484,30 @@ tuple_init_field_map(struct tuple_format *format, uint32_t *field_map,
>  			 (unsigned) format->exact_field_count);
>  		return -1;
>  	}
> -	if (validate && field_count < format->min_field_count) {
> -		diag_set(ClientError, ER_MIN_FIELD_COUNT,
> -			 (unsigned) field_count,
> -			 (unsigned) format->min_field_count);
> -		return -1;
> -	}
>  
>  	/* first field is simply accessible, so we do not store offset to it */
>  	struct tuple_field *field = tuple_format_field(format, 0);
> +	/*
> +	 * Allocate fields_bitmap - a copy of the initialized
> +	 * format:required_fields bitmap. The field:id bits would
> +	 * be nullified for founded fields during tuple parse to
> +	 * raise an error when some required field is missing.
> +	 */
> +	struct region *region = &fiber()->gc;
> +	char *fields_bitmap = NULL;
> +	uint32_t fields_bitmap_sz = bitmap_size(format->total_field_count);
> +	if (validate) {
> +		fields_bitmap = region_alloc(region, fields_bitmap_sz);
> +		if (fields_bitmap == NULL) {
> +			diag_set(OutOfMemory, fields_bitmap_sz, "calloc",

s/calloc/region

Memory allocated on the region isn't freed when the function returns...

> +				"required_fields");
> +			return -1;
> +		}
> +		memcpy(fields_bitmap, format->required_fields,
> +		       fields_bitmap_sz);
> +		if (field_count > 0)
> +			bit_clear(fields_bitmap, field->id);
> +	}
>  	if (validate &&
>  	    !field_mp_type_is_compatible(field->type, mp_typeof(*pos),
>  					 tuple_field_is_nullable(field))) {

If field_count is 0, then you'll access uninitialized memory here ^^^

I'm kinda tired of reviewing this simple patch so I fixed these issues
by myself, fixed grammar in the comments, and pushed the patch to 2.1.