[tarantool-patches] Re: [PATCH v1 1/2] sql: restrict nullable action definitions

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

Hello. See 5 comments below.

1. Assertion.

tarantool> format = {}
---
...

tarantool> format[1] = {'field1', 'unsigned', nullable_action = "undefined"}
---
...

tarantool> s = box.schema.create_space('test', {format = format})
Assertion failed: (field->nullable_action != ON_CONFLICT_ACTION_UNDEFINED), function field_def_decode, file /Users/v.shpilevoy/Work/Repositories/tarantool/src/box/alter.cc, line 393.
Process 13519 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = signal SIGABRT
     frame #0: 0x00007fff51b71b6e libsystem_kernel.dylib`__pthread_kill + 10
libsystem_kernel.dylib`__pthread_kill:
->  0x7fff51b71b6e <+10>: jae    0x7fff51b71b78            ; <+20>
     0x7fff51b71b70 <+12>: movq   %rax, %rdi
     0x7fff51b71b73 <+15>: jmp    0x7fff51b68b00            ; cerror_nocancel
     0x7fff51b71b78 <+20>: retq
Target 0: (tarantool) stopped.

On 12/07/2018 19:34, Kirill Shcherbatov wrote:
> This patch dissallows define multiple "NULL", "NOT NULL"
> options per column and fixes silent implicit behavior
> for invalid "NULL PRIMARY KEY" construction.
> 
> Closes #3473.
> ---
>   src/box/alter.cc              |  1 +
>   src/box/field_def.c           |  1 +
>   src/box/field_def.h           |  3 ++-
>   src/box/sql/build.c           | 46 +++++++++++++++++++++++++++++++++++++++----
>   src/box/sql/parse.y           |  2 +-
>   test/sql/on-conflict.result   | 13 ++++++++++++
>   test/sql/on-conflict.test.lua |  6 ++++++
>   7 files changed, 66 insertions(+), 6 deletions(-)
> 
> diff --git a/src/box/field_def.h b/src/box/field_def.h
> index 05f80d4..c35e0b8 100644
> --- a/src/box/field_def.h
> +++ b/src/box/field_def.h
> @@ -60,7 +60,8 @@ enum field_type {
>   };
>   
>   enum on_conflict_action {
> -	ON_CONFLICT_ACTION_NONE = 0,
> +	ON_CONFLICT_ACTION_UNDEFINED = 0,

2. Please, remove this new action, because
* it is not really action, but just a flag;
* it can be replaced with on_conflict_action_MAX or
   ON_CONFLICT_ACTION_DEFAULT.

> +	ON_CONFLICT_ACTION_NONE,
>   	ON_CONFLICT_ACTION_ROLLBACK,
>   	ON_CONFLICT_ACTION_ABORT,
>   	ON_CONFLICT_ACTION_FAIL,
> diff --git a/src/box/sql/build.c b/src/box/sql/build.c
> index 53c20a6..946b10c 100644
> --- a/src/box/sql/build.c
> +++ b/src/box/sql/build.c
> @@ -701,9 +701,22 @@ sqlite3AddNotNull(Parse * pParse, int onError)
>   	p = pParse->pNewTable;
>   	if (p == 0 || NEVER(p->def->field_count < 1))
>   		return;
> -	p->def->fields[p->def->field_count - 1].nullable_action = (u8)onError;
> -	p->def->fields[p->def->field_count - 1].is_nullable =
> -		action_is_nullable(onError);
> +	struct field_def *field = &p->def->fields[p->def->field_count - 1];
> +	if (field->nullable_action != ON_CONFLICT_ACTION_UNDEFINED) {
> +		/* Prevent defining nullable_action many times. */
> +		const char *err_msg =
> +			tt_sprintf("NULL declaration for column '%s' of table "
> +				   "'%s' has been already set to '%s'",
> +				   field->name, p->def->name,
> +				   on_conflict_action_strs[field->
> +							   nullable_action]);
> +		diag_set(ClientError, ER_SQL, err_msg);
> +		pParse->rc = SQL_TARANTOOL_ERROR;
> +		pParse->nErr++;
> +		return;
> +	}
> +	field->nullable_action = (u8)onError;

3. Why do you cast to u8 though nullable_action is enum?

> +	field->is_nullable = action_is_nullable(onError);
>   }
>   
>   /*
> @@ -1728,6 +1741,31 @@ sqlite3EndTable(Parse * pParse,	/* Parse context */
>   		p->def->id = SQLITE_PAGENO_TO_SPACEID(p->tnum);
>   	}
>   
> +	/* Set default on_nullable action if required. */

4. This still works:

     box.sql.execute("CREATE TABLE test (a int, b int NULL, c int, PRIMARY KEY(a, b, c))")

And 'NULL' here is ignored somehow. It is incorrect and should be caught
by core, in alter.cc, because it is not a grammar issue.

> +	for (uint32_t i = 0; i < p->def->field_count; i++) {
> +		if (p->def->fields[i].nullable_action ==
> +		    ON_CONFLICT_ACTION_UNDEFINED) {
> +			p->def->fields[i].nullable_action =
> +				ON_CONFLICT_ACTION_NONE;
> +			p->def->fields[i].is_nullable = true;
> +		} else if (p->iPKey == (int)i &&
> +			    p->def->fields[i].nullable_action ==
> +			    ON_CONFLICT_ACTION_NONE) {
> +			/*
> +			 * PRIMARY KEY can not be defined with
> +			 * ON_CONFLICT_ACTION_NONE.
> +			 */
> +			const char *err_str =
> +				tt_sprintf("Cannot define PRIMARY KEY "
> +					   "constraint on nullable column in "
> +					   "table '%s'", p->def->name);
> +			diag_set(ClientError, ER_SQL, err_str);
> +			pParse->rc = SQL_TARANTOOL_ERROR;
> +			pParse->nErr++;
> +			return;
> +		}
> +	}
> +
>   	if (!p->def->opts.is_view) {
>   		if ((p->tabFlags & TF_HasPrimaryKey) == 0) {
>   			sqlite3ErrorMsg(pParse,
> diff --git a/src/box/sql/parse.y b/src/box/sql/parse.y
> index ac935fd..6463019 100644
> --- a/src/box/sql/parse.y
> +++ b/src/box/sql/parse.y
> @@ -276,7 +276,7 @@ ccons ::= DEFAULT id(X).              {
>   // In addition to the type name, we also care about the primary key and
>   // UNIQUE constraints.
>   //
> -ccons ::= NULL onconf.
> +ccons ::= NULL onconf.           {sqlite3AddNotNull(pParse, ON_CONFLICT_ACTION_NONE);}

5. I do this:

     box.sql.execute("CREATE TABLE test (a int PRIMARY KEY, b int NULL ON CONFLICT IGNORE)")

But ON CONFLICT makes no sense for nullable columns.

>   ccons ::= NOT NULL onconf(R).    {sqlite3AddNotNull(pParse, R);}
>   ccons ::= PRIMARY KEY sortorder(Z) onconf(R) autoinc(I).
>                                    {sqlite3AddPrimaryKey(pParse,0,R,I,Z);}