[tarantool-patches] Re: [PATCH v2 1/1] sql: remove ephemeral Expr from AddDefaultValue

[Vladislav Shpilevoy <v.shpilevoy@tarantool.org>]

Hello. Please, see my 4 comments below.

On 17/04/2018 18:01, Kirill Shcherbatov wrote:
> As we don't need Expr tree in DDL we could store only
> string representation that's better to understand.
> 
> Needed for #3051.
> ---

1. Put here branch and issue links, as it pointed in guideline:
https://tarantool.io/en/doc/1.9/dev_guide/developer_guidelines.html#how-to-submit-a-patch-for-review


2. Do not create separate branch for this - you can push this commit
on the main branch, linked with the issue.

>   src/box/sql.c       | 10 ++++------
>   src/box/sql/build.c | 32 ++++++++++++++------------------
>   2 files changed, 18 insertions(+), 24 deletions(-)
> 
> diff --git a/src/box/sql/build.c b/src/box/sql/build.c
> index 394ee3a..4406955 100644
> --- a/src/box/sql/build.c
> +++ b/src/box/sql/build.c
> @@ -399,6 +399,8 @@ deleteTable(sqlite3 * db, Table * pTable)
>   	if (pTable->def) {
>   		/* fields has been allocated on separate region */
>   		struct field_def *fields = pTable->def->fields;
> +		for (uint32_t i = 0; i < pTable->def->field_count; ++i)
> +			sqlite3DbFree(sql_get(), fields[i].default_value);

3. I see, that fields deletion becomes more and more complex. Lets
move this complexity into table building - into EndTable function.
EndTable will move the def spread over different mallocs on a single
big malloc by calling space_def_new again like this:

old_def = pNewTable->def;
pNewTable->def = space_def_new(<put here old_def memebers>).

space_def_new here does this compaction for you. Then delete
old_def in the same way as you do it in deleteTable now. It
makes table->def be exactly the same as space->def, and in
deleteTable you just call space_def_delete.
> @@ -889,27 +891,21 @@ sqlite3AddDefaultValue(Parse * pParse, ExprSpan * pSpan)
>   					"default value of column [%s] is not constant",
>   					pCol->zName);
>   		} else {
> -			/* A copy of pExpr is used instead of the original, as pExpr contains
> -			 * tokens that point to volatile memory. The 'span' of the expression
> -			 * is required by pragma table_info.
> -			 */
> -			Expr x;
>   			assert(p->def);
>   			struct field_def *field =
>   				&p->def->fields[p->nCol - 1];
> -			sql_expr_free(db, field->default_value_expr, false);
> -
> -			memset(&x, 0, sizeof(x));
> -			x.op = TK_SPAN;
> -			x.u.zToken = sqlite3DbStrNDup(db, (char *)pSpan->zStart,
> -						      (int)(pSpan->zEnd -
> -							    pSpan->zStart));
> -			x.pLeft = pSpan->pExpr;
> -			x.flags = EP_Skip;
> -
> -			field->default_value_expr =
> -				sqlite3ExprDup(db, &x, EXPRDUP_REDUCE);
> -			sqlite3DbFree(db, x.u.zToken);
> +
> +			sqlite3DbFree(db, field->default_value);

4. How can it be not NULL here, if DEFAULT is not created yet?

> +			field->default_value =
> +				sqlite3DbStrNDup(db, (char *)pSpan->zStart,
> +						 (int)(pSpan->zEnd -
> +						       pSpan->zStart));
> +			if (field->default_value == NULL) {
> +				assert(db->mallocFailed);
> +				pParse->rc = SQLITE_NOMEM_BKPT;
> +				pParse->nErr++;
> +				return;
> +			}
>   		}
>   	}
>   	sql_expr_free(db, pSpan->pExpr, false);
>