[MrBlack1995]

Summary

First slice of apache/ossie-converters#23: an opt-in, deterministic DAX to OSSIE_SQL_2026 translator for Power BI measures on import, so the most common measures leave Power BI in a portable form every Ossie consumer can read.

Today every imported measure is DAX only. With --translate-dax, a measure that is a single aggregate over one fully qualified column gets OSSIE_SQL_2026 as its primary dialect, with the original DAX kept as a fallback, so nothing is lost:

# ossie-microsoft import -i sales_model.bim --translate-dax
- name: Total Sales
  expression:
    dialects:
      - dialect: OSSIE_SQL_2026
        expression: SUM(Sales.Amount)
      - dialect: DAX
        expression: SUM ( Sales[Amount] )

Everything else is refused and kept exactly as today (DAX only), with a warning naming the reason:

[table 'Sales' measure 'Order Count'] DAX measure kept as-is; not translated to OSSIE_SQL_2026 (COUNTROWS counts rows, not a single column; it is a later tier)

Change

• New ossie_microsoft/_dax_to_sql.py, the reverse of _sql_to_dax.py. It translates exactly the functions _sql_to_dax emits, so an Ossie SQL model exported to Power BI and imported again comes back to the same expression:

  DAX	OSSIE_SQL_2026
  SUM('T'[c])	SUM(T.c)
  AVERAGE('T'[c])	AVG(T.c)
  MIN('T'[c]) / MAX('T'[c])	MIN(T.c) / MAX(T.c)
  COUNTA('T'[c])	COUNT(T.c)
  DISTINCTCOUNTNOBLANK('T'[c])	COUNT(DISTINCT T.c)

• Refused on purpose, following the reasoning already written down in _sql_to_dax: DAX COUNT (TRUE/FALSE values unsupported, so it is not SQL COUNT(x) for an arbitrary column) and DISTINCTCOUNT (counts BLANK as a distinct value). The empty-set convention (DAX BLANK vs SQL 0) is the one _sql_to_dax already accepts.

• Also refused: anything with more than one argument or a nested call (COUNTROWS, iterators, CALCULATE, DIVIDE, arithmetic, IF, time intelligence), unqualified columns, a column that does not resolve to exactly one dataset field, and names that would need SQL quoting. These are later tiers.

• translate_dax=False is threaded through convert_semantic_model_to_ossie, build_ossie_document and _convert_metrics; the import subcommand gets --translate-dax. A resolver maps 'Table'[Column] to the Ossie dataset.field built from the model.

Answers to the questions on apache/ossie-converters#23, as implemented (easy to change)

1. Lives in converters/microsoft, in Python, as a sibling of _sql_to_dax.py.
2. Opt-in, default off. Default output is unchanged, so existing Power BI round trips are unaffected. If you would rather have it on by default, that is a one-line change.
3. No catalog format yet: this slice is a six-entry mapping. The Python code vs catalog.yaml question can be settled in the PR that grows it.

Measured effect

Running the portability fixtures in #596 (M0 to M3) through Ossie to Power BI to Ossie:

	DAX only	OSSIE_SQL_2026	BLANK() placeholder
today (flag off)	16	0	2
--translate-dax	0	16	2

The 16 returned expressions are identical to the original SQL (SUM(sales.amount) comes back as SUM(sales.amount), COUNT(DISTINCT sales.customer_id) as COUNT(DISTINCT sales.customer_id), and so on). The 2 placeholders are the M3 filtered aggregates, which the SQL to DAX export already reduces to BLANK(); that is the export side and out of scope for an import translator.

Validation

• converters/microsoft: 468 passed, 2 skipped locally; ruff check clean. 28 new tests in tests/test_dax_to_sql.py (golden translations, every refusal, reasons, quoted names, and integration tests for flag off, a translated measure with DAX fallback, and a refused measure).
• On tests/fixtures/sales_model.bim: flag off output is unchanged; flag on output passes validation/validate.py; exporting the flag-on document back to Power BI gives the original DAX for every measure.
• No existing behavior changes unless the flag is passed.

Related Issues

Refs apache/ossie-converters#23. Measured against the fixtures in #596 (#576).

Checklist

Converters

• Converter logic in converters/ is updated (Microsoft import, opt-in)
• Tests added; existing suite passes

[jbonofre]

@MrBlack1995 as shared on the dev mailing list, the converters have moved to a dedicated repository: https://github.com/apache/ossie-converters

Can you please create this PR on the new repo? Thanks!

[MrBlack1995]

Thanks @jbonofre, re-opened as apache/ossie-converters#6.