docs/superpowers/notes/ws6b-complete.md
Status: DONE on branch 6.0-rc (pushed to fork). All four layered workflows green. Spec: docs/superpowers/specs/2026-05-27-ws6-platform-cicd-design.md §8–§9, decisions D4/D8. Plan: docs/superpowers/plans/2026-05-27-ws6b-web-and-sanitizers.md. Completes WS6 (with ws6-prep-complete.md + ws6a-complete.md).
e7727ad, 1809707, cb43410)Green on ubuntu. emsdk 3.1.64 (via mymindstorm/setup-emsdk@v14, no bump needed). Builds raylib-sys and raylib for wasm32-unknown-emscripten. Build-verify only — no tests (wasm can't open a window in CI), no Pages deploy (WS9).
bits/libc-header-start.h not found); added BINDGEN_EXTRA_CLANG_ARGS=--sysroot=$EMSDK/upstream/emscripten/cache/sysroot (via $GITHUB_ENV since $EMSDK is dynamic).raylib-sys/build.rs needed no changes; the libraylib.bc→.a copy is a guarded no-op (6.0 emits .a directly).software_renderer-wasm build step (spec §8 step 6 deviation): software_renderer forces Platform::Memory in build.rs before the wasm-target check, so it generates Memory bindings + compiles the C shims as native x86 linked into a wasm binary — an architecture mismatch. rlsw (Memory platform) compiled-to-wasm is a genuinely different, unsupported build config; headless rlsw testing already runs on desktop CI. → deferred follow-up (§ below). WS6's actual done-criterion ("web build verified continuously") is met by the sys+safe wasm builds.32c3b18, 539b901)Workflow concludes success (whole job continue-on-error, D8: informational, never a gate). Retargeted from model_animation_raii (in the stale raylib-test — see spike) to the FFI-heavy software_renderer render tests, which compile and exercise raylib's C rasterizer.
render_shapes + render_text pass under -Zsanitizer=address + ENABLE_ASAN (C sanitized via cmake, Rust via -Z build-std). No ASAN violations at the FFI boundary. Genuine, useful coverage.-Zsanitizer=undefined is not a valid rustc value (UBSan isn't in -Zsanitizer), so it was dropped; C-side UBSAN via ENABLE_UBSAN then fails to link (the cmake-built C's __ubsan_handle_* calls aren't resolved by rust-lld under -Z build-std). Absorbed by continue-on-error. A real UBSAN run would need -C linker=gcc or libubsan on the runner — out of scope for an informational job.check/test/web/sanitizers) replace baseline.yml.unit ×6, no-default ×3, software-render ×3); web (wasm32-unknown-emscripten) build-verified.ws6a-complete.md §Gate-reality): fmt, clippy -Dwarnings, deny(missing_docs), doctests/doc-link, cargo-deny, MSRV-1.85.Sound impls from #277).release.yml deferred to WS8; Pages deploy deferred to WS9 (per the confirmed decisions).6.0-rc)| Workflow | Jobs | Status |
|---|---|---|
check | fmt, clippy, docs, cargo-deny, msrv | ✅ gates enforce + fail-on-violation |
test | unit ×6, no-default ×3, software-render ×3, integration-xvfb (non-required) | ✅ |
web | wasm-build (sys + safe) | ✅ build-verify |
sanitizers | asan-ubsan (informational) | ✅ (ASAN clean; UBSAN best-effort) |
get_gamepad_button_pressed transmute (input.rs) — latent UB on out-of-range; small fix.notes/spike-raylib-test-delete-or-fix.md (decide at WS9); integration-xvfb non-required until then.structopt→clap, consider paste alternative.software_renderer for the wasm target (fix build.rs platform_from_target ordering + wasm C-shim compilation), or add a compile_error! guard for the software_renderer + emscripten combo (mirrors the existing software_renderer+opengl_* guard).-C linker=gcc/libubsan) if a real UBSAN gate is ever wanted (informational only per D8).WS6 done. Next: WS7 (docs & book authoring) — the doc stubs from WS6a are the seed; WS7 enriches into the mdBook. WS5/WS6/WS7 parallelize; the crate release is WS8; the showcase + Pages site is the WS9 finale.