[tarantool-patches] Re: [PATCH 0/2] sql: store statistics in statN as an array of integers

[n.pettik]

Travis status is entirely negative - it can’t be compiled.
Please, make sure that at least build is successfully finished
before sending patch.

> On 27 Jan 2019, at 03:28, Roman Khabibov <roman.habibov at tarantool.org> wrote:
> 
> It was necessary to add function for msg pack result in sqlite3 VDBE API.

Actually, it wasn’t. You still can use result_blob and
set subtype after its invocation. Now it doesn’t matter much.

> The patchset is very raw.
> There are problems with the old tests that need
> to be solved.

Ok, could you at least explain or enumerate existing problems
(in our char, in ticket comments, wherever you want) and
possible solutions.

> I do not know if a piece of code with keyword checking works.
> Could you send in the reply letter test samples for "unordered" and "noskipscan"
> keywords?

What kind of samples do you need? To check that it works
you need get plan of query (explain query plan …), manually
update statisctics in _sql_stat space, probably restart instance
to load to in-memory structs new statistics and make sure
that plan of query has changed. Ofc not all plans will change,
but only those which are related to skip-scan optimisation
(don’t remember what the second token is responsible for).