The core `duplication` check was jscpd-via-npx — it needs the target's Node toolchain, so it silently skipped on every non-Node stack (.NET, legacy ASP, Go, Python). On exactly the polyglot/legacy repos checkup targets, the duplication signal — where a bug gets fixed in one copy and not the others — was absent. Make `duplication` a two-tier engine picked by stack, mirroring the complexity ladder (#33): 1. jscpd — Node targets (package.json + npm): exact-token clone detection, the better-fit tool, unchanged. 2. lizard — every other stack: `-Eduplicate` does identifier-unified (type-2) clone detection across the languages lizard tokenises (C#, Java, Go, Python, C/C++, …). Spike (#34) weighed PMD CPD and scc against lizard. lizard wins for core: it is ALREADY baked + pinned (from the complexity work, #33), so this fills the non-Node gap at +0 MB. PMD CPD would drag a JRE into core — breaching the lean-core rule (ADR-0004) and the #4 size budget — AND still couldn't tokenise Classic ASP, so the weight buys nothing lizard doesn't already give. scc's only duplicate flag is whole-file dedup, not clone detection. - Shared LIZARD_BIN / SCAN_ROOTS / LIZARD_PROBE resolution hoisted to one early block (duplication runs before complexity, so the resolution can't live alongside the complexity engine); section 13 now consumes it. - lizard's text report is parsed (Duplicate block: locations + the overall rate) into the existing `duplication` parsed contract — renderer picks it up unchanged. Empty/non-parseable output fails honestly, never a false 0%-pass (same guard the dotnet overlay uses for CPD). - duplication is now a fifth Focus Areas axis: a file that is both hot and copy-pasted rises to the top of the "where to focus" view. Honest limitation: Classic ASP/VBScript has no tokeniser in either engine, so .asp duplication stays unmeasured — documented in the check's intent. Validated end-to-end (host + in checkup-core image, 884 MB = unchanged): a polyglot C#/Python repo routes to lizard, reports a 23-line clone, and Focus Areas fuses it with complexity on the same file; a Node target routes to jscpd; an ASP-only target skips honestly. shellcheck -S error clean, run-tool tests 29/29, banned-terms clean. Closes #34 Co-authored-by: Mark Ridley <210189+maudlin@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CI: enforce an image-size budget #4

Description

Activity

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions