[patches] [schema 1/1] Allow to do not specify tail nullable index columns

[v.shpilevoy at tarantool.org]

I agree, that the feature seems to be strange, but it is requested by a customer. The only hot paths touched here is is_nullable = true comparators, which are not critical.

> 25 янв. 2018 г., в 16:16, Vladimir Davydov <vdavydov.dev at gmail.com> написал(а):
> 
> To tell the truth, I'm not quite convinced of the usefulness of this
> feature - I simply fail to see a proper use case for it - but then again
> I don't use Tarantool and so can't be impartial. I guess we should
> question our users whether they need it and what for.
> 
> Anyway, I see that this patch touches some hot paths so IMO the first
> thing we should do is run some tests (cbench?) to check if there's a
> performance degradation after it is applied.
> 
> Apart from potential performance issues and somewhat intricate
> definition of indexed field constraints, I currently don't see
> any pitfalls this feature can introduce.
> 
> On Thu, Jan 25, 2018 at 03:26:15PM +0300, Konstantin Osipov wrote:
>> What's your take on this patch?
>> 
>> * Vladislav Shpilevoy <v.shpilevoy at tarantool.org> [18/01/25 13:59]:
>> 
>> 
>>> If a column is nullable and is the last defined one (via index parts
>>> or space format), it can be omited on insertion. Such absent fields
>>> are treated as NULLs in comparators and are not stored.
>>> 
>>> Closes #2988
>>> 
>>> Signed-off-by: Vladislav Shpilevoy <v.shpilevoy at tarantool.org>
>>> ---
>>> src/box/tuple_compare.cc     |  77 +++++--
>>> src/box/tuple_extract_key.cc | 173 +++++++++++---
>>> src/box/tuple_format.c       |  11 +-
>>> src/box/tuple_format.h       |  10 +-
>>> src/box/tuple_hash.cc        |  32 ++-
>>> src/box/vy_stmt.c            |  17 +-
>>> src/box/vy_stmt.h            |   2 +-
>>> test/engine/null.result      | 526 +++++++++++++++++++++++++++++++++++++++++++
>>> test/engine/null.test.lua    | 130 +++++++++++
>>> 9 files changed, 912 insertions(+), 66 deletions(-)