[Feature] Import and export supported SQL, Power Query M, and R profiles with loss diagnostics #16

New Issue

2026-07-29T14:27:58+02:00

zemion commented

2026-07-29 14:27:58 +02:00

Parent: #1

Outcome

Power users can bring existing transformations into the canonical Dataflow graph and export supported subsets without GovOPlaN falsely claiming lossless conversion.

Scope

Define versioned compatibility profiles for supported SQL dialects, a practical Power Query M subset, and an R/dplyr-style subset or explicitly sandboxed opaque step.
Use established parsers/ASTs where available; do not build ad-hoc text rewriting for mature languages.
Translate supported expressions/operators into the typed Dataflow IR and preserve source spans, comments/labels where feasible, and parameter bindings.
Unsupported constructs produce structured blocking or lossy diagnostics with exact source locations and suggested replacements.
Exports state the target profile and every semantic loss; no silent fallback to arbitrary executable code.
A golden corpus covers joins, unions, filters, projections, type conversions, null/error behavior, grouping, windows, pivots, parameters, locale/timezone, and source-specific functions.
Round-trip tests distinguish textual equality, semantic equality, supported-loss conversion, and opaque preservation.
Imported code executes only through governed Dataflow backends and security budgets.

Parent: #1 ## Outcome Power users can bring existing transformations into the canonical Dataflow graph and export supported subsets without GovOPlaN falsely claiming lossless conversion. ## Scope - Define versioned compatibility profiles for supported SQL dialects, a practical Power Query M subset, and an R/dplyr-style subset or explicitly sandboxed opaque step. - Use established parsers/ASTs where available; do not build ad-hoc text rewriting for mature languages. - Translate supported expressions/operators into the typed Dataflow IR and preserve source spans, comments/labels where feasible, and parameter bindings. - Unsupported constructs produce structured blocking or lossy diagnostics with exact source locations and suggested replacements. - Exports state the target profile and every semantic loss; no silent fallback to arbitrary executable code. - A golden corpus covers joins, unions, filters, projections, type conversions, null/error behavior, grouping, windows, pivots, parameters, locale/timezone, and source-specific functions. - Round-trip tests distinguish textual equality, semantic equality, supported-loss conversion, and opaque preservation. - Imported code executes only through governed Dataflow backends and security budgets.