ponder-lab / Hybridize-Functions-Refactoring / #1912

80%

main: 79%

Build Type

Pull #615

travis-pro

Committed by khatchad

Commit Message

Auto-inject the union of all functions' dtype constants per file

The auto-inject path in `convertToHybrid` injected a `from tensorflow
import ...` line carrying the dtype constants of the first hybridizable
function processed in an import-less file. A later function in the same
file needing a dtype the first did not inject was gated off emission by
the reachability check (#585) and left with a bare `@function`.

Add a file-level pre-pass, `Function.planAutoInjectedImports`, that the
processor runs before transforming: it collects, per file, the union of
every to-be-hybridized function's `requiredDTypeNames` from the memoized
inferred signatures (no inference re-triggered). `convertToHybrid` unions
that plan with the function's own dtypes when fixing the injected line,
so every function in the file emits its signature unqualified. Falls back
to the function's own dtypes when no plan was computed (direct transform).

Closes #588

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Pull Request Pull Request #615: Auto-inject the union of all functions' dtype constants per file

Coverage Stats

15 of 17 new or added lines in 2 files covered. (88.24%)

1 existing line in 1 file now uncovered.

1554 of 1943 relevant lines covered (79.98%)

0.8 hits per line