Обсуждение: Incorrect START TRANSACTION implementation
SQL99 part 2 clause 16.1 contains this note: NOTE 327 - The characteristics of a transaction begun by a <start transaction statement> are as specified in these General Rules regardless of the characteristics specified by any preceding <set transaction statement>. That is, evenif one or more characteristics are omitted by the <start transaction statement>, the defaults specified in the SyntaxRules of this Subclause are effective and are not affected by any (preceding) <set transaction statement>. In other words, this should work: peter=# set session characteristics as transaction read only; SET peter=# start transaction; START TRANSACTION peter=# create table test (a int); ERROR: transaction is read-only because the default of a bare "start transaction" is read write. I propose that we change this to follow the standard and keep the traditional behavior for BEGIN only.
Peter Eisentraut <peter_e@gmx.net> writes:
> SQL99 part 2 clause 16.1 contains this note:
> NOTE 327 - The characteristics of a transaction begun by a
> <start transaction statement> are as specified in these General
> Rules regardless of the characteristics specified by any
> preceding <set transaction statement>. That is, even if one
> or more characteristics are omitted by the <start transaction
> statement>, the defaults specified in the Syntax Rules of this
> Subclause are effective and are not affected by any (preceding)
> <set transaction statement>.
> In other words, this should work:
> peter=# set session characteristics as transaction read only;
> SET
> peter=# start transaction;
> START TRANSACTION
> peter=# create table test (a int);
> ERROR: transaction is read-only
Are you sure you're reading that correctly? It says <set transaction
statement> not <set session characteristics statement>. What use would
SET SESSION CHARACTERISTICS have at all, if it's ignored by START
TRANSACTION?
It appears to me that the spec is saying that *this* should work:
regression=# set transaction read only;
SET
regression=# start transaction;
START TRANSACTION
regression=# create table test (a int);
CREATE TABLE
... which it does.
regards, tom lane
Tom Lane wrote: > Are you sure you're reading that correctly? Not anymore... :-/ Sorry for the noise. I had remembered that in some context <set transaction> and <set session characteristics> were interchangeable, but apparently I got it all mixed up.