pragma index_info

[pgulutzan]

I have Ubuntu 18.04. I have Tarantool 2.1, pulled from source today.

I execute these requests:
CREATE TABLE f (s1 CHAR(1) PRIMARY KEY);
CREATE INDEX i ON f (s1);
PRAGMA index_info(f.i);
CREATE TABLE g (s1 CHAR(1) COLLATE "BINARY");

The PRAGMA statement says that the collation is BINARY (upper case).
The CREATE TABLE g statement says that there is no such collation as BINARY (upper case).

The correct name is binary (lower case), as seen in PRAGMA collation_list.

Also, the PRAGMA statement says the data type is "string" -- why not CHAR?

Also, the PRAGMA statement would say that the collation of an integer column
is BINARY -- why not NULL or 'none'? I don't know what 'none' means, so I'm
not sure whether it is correct, but I know what 'binary' means, and it won't
work on integers (assuming MessagePack high-endian storage and assuming
unsigned but not assuming everything has the same length).

[Korablev77]

Well, 'none' collation is @kostja 's idea. It was needed for this issue: #3185 to tell binary collation from total absence of collation and meanwhile operate on collation objects in the same way.

Other concerns are related mostly to output format, so I guess it is easy to fix.

[kostja]

This is still not documented, so can hardly be a bug. I think we should support an information schema view for indexes, not a pragma.

[pgulutzan]

Well, kostja is right that we should have a view similar to what information_schema has, but now PRAGMA is documented, so I think this could be regarded as a bug.