No way to support databases with tables that have exotic names

[ricomariani]

e.g. a database with a table named [x y]and a column named [a b] is fundamentally not usable by CQL today.

The idea here to add basic support for this:

• Introduce `x y` as a valid quoted name format
• support this in sql statements where a sql name can be used
  • DONE insert, update, select, delete forms
  • DONE create table (table and column names and constraints)
  • DONE create index
  • DONE create view
  • DONE create trigger
  • DONE constraint names including references
  • DONE drop [all the above]
  • DONE alter table add column

These names leak into the algebra... these areas have to be enlightened.

• DONE

  • allow `x y` in dot expressions e.g. `x y`.a, `x y`.`a b`
  • auto-rename the exotic names into something that C can support where they appear in structs
  • when producing output for sqlite use [x y].[a b] etc.
  • support this in * and T.* expansion
  • support declare C cursor LIKE `abc def` and other shape references
  • support general names in expansions such as FROM C
  • support general names in shape narrowing e.g.LIKE C(-`x y`)
  • support field names like `x y` in cursors, so C.`x y` has to work
  • enlighted gen_sql to the presence of this new kind of identifier and format it appropriately
  • because of proc foo(LIKE table_name) locals can have exotic names
  • since locals have to be supported then extend LET and SET so that they work on exotic names too
  • support decoding the names in JSON output
  • support query plan generating queries that use these in the table output
  • support adding and removing exotic tables and columns in schema upgrader
  • support these names in backed columns and backing columns
  • support quoted names in arg bundles X like `foo bar` where `foo bar` might have `a column`
  • support autotest creating such tables as needed
  • support typed columns with exotic names (a column integer, another column real)
  • support quoted id in misc attributes
  • support quoted names in generated exports output
  • support cursor diff rewrite (turns out it's best to use the C field name)
  • support cursor format rewrite (turns out it's best to use the C field name)
  • support dynamic cursors (give the cursor call the nice name in the name argument) (turns out it's best to use the C field name)
  • support these names in blob storage (this might be a no-op)
  • support fetch cursor USING form (a no-op, this falls out once INSERT works)
  • support these names in the incremental @recreate update case, column pattern matching might need to be generalized (this might be a no-op) (OSS version of CQL doesn't support rebuild so this is a no-op, this was a Meta internal only feature)

• PENDING nothing

[ricomariani]

`abc def` is escaped to X_abcX20def for use in C/Lua. This transformation is reversable. In principle it can collide with user names but this is not a new thing for CQL. Note that `abc` will map to just abc. This means if you are doing this just to avoid reserved word collisions then output quality does not suffer. In short anything that isn't an alphanumeric is hex encoded in the C/Lua. This really only happens for fields in cursors. The C code doesn't include table names that are not in strings.

[ricomariani]

To see progress on this item check out https://github.com/ricomariani/CG-SQL-author/pull/53

This is not going to be a five-minute fix by any means...

[ricomariani]

Further progress: https://github.com/ricomariani/CG-SQL-author/commit/dba95f4717320faf31d19b004812f1f05b16d809

What's left is a lot more fringe at this point. With query plan, upgrade, and JSON plus the main codegen supporting these it's quite usable.

Test helpers should be very straightforward, that will be next.

[ricomariani]

Test helpers are done and it was less straightforward than I expected because test helpers force the issue of:

• quoted names in typed names which was already a thing really
• quoted names in attributes so you can name your dummy test columns

There are not that many things left that can't have an exotic name.

[ricomariani]

I'm not aware of any missing cases now. I may do another exploration pass just to make sure but I think everything that needs to be done is done at this point. I will open separate issues for anything non-trivial that I discover while I play with it.