Enable table name uniquification, fix column uniquification

[AndriySvyryd]

Currently table names longer than MaxIdentifierLength are just truncated.

There's also a bug in column uniquification that prefixes duplicate columns with the type name even on the same type.

[roji]

@AndriySvyryd, after updating Npgsql to preview4 I get:

2019-04-20 14:00:03.826 CEST [6256] npgsql_tests@PartialUpdateNpgsqlTest LOG:  execute <unnamed>: CREATE UNIQUE INDEX "IX_LoginEntityTypeWithAnExtremelyLongAndOverlyConvolutedNameTh~" ON "LoginEntityTypeWithAnExtremelyLongAndOverlyConvolutedNameThat~1" ("ProfileId", "ProfileId1", "ProfileId3", "ProfileId4", "ProfileId5", "ProfileId6", "ProfileId7", "ProfileId8", "ProfileId9", "ProfileId10", "ProfileId11", "ProfileId12", "ProfileId13", "ProfileId14")                                                                                        
2019-04-20 14:00:03.840 CEST [6256] npgsql_tests@PartialUpdateNpgsqlTest LOG:  execute <unnamed>: CREATE INDEX "IX_LoginEntityTypeWithAnExtremelyLongAndOverlyConvolutedNameTh~" ON "LoginEntityTypeWithAnExtremelyLongAndOverlyConvolutedNameThatI~" ("ProfileId", "ProfileId1", "ProfileId3", "ProfileId4", "ProfileId5", "ProfileId6", "ProfileId7", "ProfileId8", "ProfileId9", "ProfileId10", "ProfileId11", "ProfileId12", "ProfileId13", "ProfileId14", "ExtraProperty")                                                                              
2019-04-20 14:00:03.841 CEST [6256] npgsql_tests@PartialUpdateNpgsqlTest ERROR:  relation "IX_LoginEntityTypeWithAnExtremelyLongAndOverlyConvolutedNameTh~" already exists                                                                                                    

Seems like there may be an issue with the uniquification for the indexes? Not sure why two are being created (one UNIQUE, one non-UNIQUE)... Anything I'm doing wrong?

[roji]

Note that the max identifier length on PostgreSQL is 63

[ErikEJ]

Should that limit not be set in the test?

[roji]

I haven't had time to dive in, but I don't think so - at least up to preview4 there was nothing specifically needed to be set up in the tests...

[AndriySvyryd]

@roji This is a bug and precisely the kind this test should catch. The indexes are being created on different tables, but the uniquification code only looks at the indexes in any single table.

[roji]

Thanks @AndriySvyryd, opened #15425 to track. Am curious why this didn't fail on SqlServer/Sqlite (identifier limitation wasn't hit for some reason?)