net.box schema contains collation_id (number) instead of collation (string)

[Totktonada]

How to reproduce (on clean instance, because no box.once is used):

box.cfg{listen = 3301}
box.schema.user.grant('guest', 'read,write,execute,create', 'universe')
box.schema.space.create('s')
box.space.s:create_index('pk')
box.space.s:create_index('sk', {type = 'tree', unique = false, parts = {{field = 2, type = 'string', is_nullable = false, collation = 'unicode_ci'}}})
net_box = require('net.box')
c = net_box.connect('localhost:3301')
box.space.s.index.sk.parts[1]
---
- type: string
  is_nullable: false
  collation: unicode_ci
  fieldno: 2
...
c.space.s.index.sk.parts[1]
---
- type: string
  is_nullable: false
  collation_id: 2
  fieldno: 2
...

As we see local box provides collation field, but net.box provides collation_id. There are two reasons why this looks as a problem for me:

• It complicates box/lua/key_def.lua code (merger patchset) a bit (need to handle both formats).
• Lack of some kind of a primary identifier for a collation. This should be its string name I think.

How can be implemented:

net.box can acquire and use info from _collation space.

[Totktonada]

It also worth to handle the case when we connect to a quite old tarantool w/o _collation system space at all. It should not lead to an error, but parts should be shown w/o 'collation' field instead.

[Totktonada]

If the key_def module created in the scope of #3276 will land before this patch it is necessary to remove collation_id parameter support in the module.

[Totktonada]

Merger (#3276) will be pushed to 1.10, it depends on key_def module (it is part of merger's patchset). The reason of this issue is to simplify lua/key_def.c code and remove collation_id from it at all. So this issue should be fixed in 1.10 too.

@kyukhin Can you please review the milestone?

[locker]

Merger (#3276) will be pushed to 1.10, it depends on key_def module (it is part of merger's patchset). The reason of this issue is to simplify lua/key_def.c code and remove collation_id from it at all. So this issue should be fixed in 1.10 too.

@kyukhin Can you please review the milestone?

Pushing this one to 1.10 would be tricky as it upgrades the schema to create _vcollation space. If we had to update the upgrade script both in 1.10 and 2.1 and master, then in master and 2.1 we'd have to handle the case when the space already exists (i.e. it was created by upgrade to a previous version), which is kinda ugly.

Do we really need _vcollation to backport the merger to 1.10? As a matter of fact, do we really need to backport the merger to 1.10?

[Totktonada]

Now it seems that we should not backport merger to 1.10, the demand is gone. So this patch can be implemented for master only.

[locker]

Now it seems that we should not backport merger to 1.10, the demand is gone. So this patch can be implemented for master only.

Okay then. Moving back to 2.2.