scripts/repro/optional-nullish-matrix/README.md

GitHub

Optional/Nullish Reproduction Matrix

This harness checks how common tools lower optional chaining, nullish coalescing, and nullish assignment snippets, then runs wakaru on the lowered output. Babel is tested across a few meaningful lines: early proposal plugins, assumptions-era Babel 7, current transform plugins, and the Babel 8 RC line.

The matrix also includes standalone Terser rows and Babel/TypeScript/SWC/esbuild output minified through Terser, because optional/nullish lowering can become a different recoverable shape after minification. The nullish-assignment rows cover identifier, static-member, and side-effectful computed-member targets. The computed-member rows verify that compiler object/key temporaries are folded back into getTarget()[getKey()] ??= make() when those temporaries prove the receiver and key are each evaluated once. The nullish-assignment rows also opt into the execution-equivalence check (see docs/testing.md), which enforces that contract behaviorally: a recovery that calls getTarget() or getKey() twice produces a different effect log and fails the row.

The logical-AND rows include issue #166-style boolean prefixes, such as a lowered optional chain followed by ordinary suffix conditions or a second lowered optional access.

It is for investigation, not as a committed snapshot source. Use it to find reproduced tool shapes worth minimizing into focused Rust unit tests.

node scripts/repro/optional-nullish-matrix/matrix.mjs

Add --details to print full lowered and recovered code for missed cases. Add --level minimal, --level standard, or --level aggressive to run wakaru with a specific rewrite level.

Some Babel loose optional-call rows are expected to stay unrecovered at standard, including the same loose output after Terser: those lowerings read the same property twice, while optional-call syntax reads it once. Wakaru only recovers those rows at aggressive, where stable getter reads are an accepted assumption. The matrix marks those rows as gated instead of no.

Rows are grouped by distinct lowered output per snippet. The grouping key only normalizes CRLF to LF and trims leading/trailing whitespace, so exact helper shape is still preserved while duplicate tool outputs are collapsed.

By default the script uses target/debug/wakaru(.exe) when present, otherwise it falls back to cargo run -q -p wakaru-cli --. Set WAKARU to test a specific binary.

The transformer packages are installed under target/repro-tools/, so the first run may download Babel, TypeScript, SWC, esbuild, or Terser packages. The target/ directory is ignored by git.