miniml/TASKS.md

A minimal, embeddable semantic data modeling language for generating SQL queries from YAML model definitions. Inspired by LookML.

- [ ] date-granularity not applied as exoected if date-field is specified as "date" when date_field is something else (see tmp/date-granularity-problem.md)

- [ ] Category level for models for 3-stage breakdowns

- [ ] Implement tags scheme that limits metadata rendering and model access

- [ ] Specifying order_by with a date field that has an underlying expression and that is not in the select list will cause a SQL error, fix is to expand the underlying expression in the order_by clause in this case

- [x] Specifying a date_field that is not of type TIMESTAMP will cause errors when SQL is generated from the following call tree:
    - appendDefaultDateRange [query.ts]
    - constructDateRangeExpression [dialect.ts]
    - constructCurrentTimeOffsetExpression [dialect.ts] 
    - generates an expression like `date >= CURRENT_TIMESTAMP - INTERVAL 1 DAY` which fails when `date` is not of type TIMESTAMP
    - Consider adding an additional miniml field named `date_type` so type can be specified as either TIMESTAMP or DATE and generates CURRENT_TIMESTAMP or CURRENT_DATE appropriately
    - FIXED: added `date_type` (DATE|TIMESTAMP, default TIMESTAMP). Default-range path now emits CURRENT_DATE vs CURRENT_TIMESTAMP and a DATE-appropriate upper bound accordingly. Separately, the explicit date_from/date_to upper bound was changed from `DATE_ADD('<lit>', INTERVAL 1 DAY)` (which forces a DATE result and mismatched TIMESTAMP columns) to a bare next-day string literal that coerces to either type; the lower bound now also strips any time component so a DATE column never sees a datetime literal.

- [ ] Consider refactoring dialect.ts to break each dialect into a seperate module file (one file for bigquery, one file for snowflake, etc.)

- [ ] Split validation.test into the following
    - generic.test ARRAY_TO_STRING(colors, ', ')
    - bigquery.test 'red' IN UNNEST(colors)
    - snowflake-test ARRAY_CONTAINS('red'::VARIANT, colors)

- [ ] Consider a new feature to join between models


- [x] If at least one dimension or measure is not specified, then render as `SELECT *`
- [x] When `date_from` and `date_to` are the same no rows are returned which is a common scenario
- [x] Add MAX_EXPRESSION_LENGTH 10X
- [x] Support function calls in where clause like ARRAY_CONTAINS('At Risk'::VARIANT, trackers) = TRUE
- [x] refactor expandWhereReferences to use AST instead of regexp's
- [x] validate default_date_range parameter
- [x] validate date_field parameter
- [x] validate that all sql inputs end with "as <name>", a lot of assumptions this is the case
- [x] validate dialect is only bigquery or snowflake
- [x] check that alias matches the key name (is this a problem?)
- [x] add params to inputs
- [x] allow formulas in dimensions like `CONCAT(first_name, ' ', last_name) AS person_name`
- [x] also allow jinja expansion like `CONCAT(retailer_url, '{{product_path}}') AS page_url`
- [x] tests to verify important notes are appended to info field

- [x] HAVING on a ratio-of-aggregates measure (e.g. `SAFE_DIVIDE(SUM(a), SUM(b))`) failed with "Aggregations of aggregations are not allowed" because HAVING re-expanded the measure SQL. Fixed by referencing the measure alias in HAVING instead (matching ORDER BY), and auto-projecting any HAVING-referenced measure that was not explicitly selected so its alias resolves.