Re: [PATCH 5/5] Allow to execute non-yielding DDL statements in transactions

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

On Mon, Jul 08, 2019 at 11:21:16AM +0300, Konstantin Osipov wrote:
> * Vladimir Davydov <vdavydov.dev@gmail.com> [19/07/08 11:14]:
> > On Sat, Jul 06, 2019 at 01:56:35AM +0300, Konstantin Osipov wrote:
> > > * Vladimir Davydov <vdavydov.dev@gmail.com> [19/07/05 23:27]:
> > > 
> > > Please work on a test case which not only ensures the statements
> > > are allowed, but also produce desired results - as well as leave
> > > no traces/partial effects on rollback.
> > 
> > As a matter of fact, this simple test does check that rollback doesn't
> > leave any traces - if it did, an attempt to commit the same statements
> > would fail:
> > 
> >   box.begin() create() box.rollback() -- suppose it leaved some traces
> >   box.begin() create() box.commit() -- then this would fail
> > 
> > It also checks that the DDL operations produce the desired results,
> > otherwise an attempt to drop created objects would fail:
> > 
> >   box.begin() create() box.commit() -- suppose it didn't create some objects
> >   box.begin() drop() box.commit() -- then this would fail
> 
> In any case you test a fixed number of combinations, and should
> come up with a randomized test. This feature encompasses all DDL
> operators, including:
> 
> - roles
> - users
> - functions
> - SQL triggers
> - SQL foreign key constraints
> - SQL views
> - SQL check constraints
> 
> Every object should be involved in the test, and the generated
> transactions should be random.

Sounds tricky :-/ May be, hand this over to Q&A?