From 4e764b86799d48fdb4a72fcb3b778924722d6c09 Mon Sep 17 00:00:00 2001 From: InauguralPhysicist Date: Fri, 24 Jul 2026 19:28:08 -0500 Subject: [PATCH] Bump EigenScript pin v0.30.0 -> v0.32.0 Local preflight against a v0.32.0 worktree build: make lint (clean) + make test (all headless tests passed), run with EIGS_DIR pinned to a v0.32.0 worktree Also fixes ALREADY-DRIFTED prose: CLAUDE.md claimed the pin was v0.26.0 while the Dockerfile said v0.30.0. Now both read v0.32.0. Co-Authored-By: Claude Opus 5 (1M context) --- .devcontainer/Dockerfile | 2 +- CLAUDE.md | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/.devcontainer/Dockerfile b/.devcontainer/Dockerfile index f07238b..bbd72dd 100644 --- a/.devcontainer/Dockerfile +++ b/.devcontainer/Dockerfile @@ -27,7 +27,7 @@ RUN apt-get update \ # EIGS_REF is a tag/SHA (fetched directly, so it stays reproducible). Pinned to # v0.19.0 — the release that ships the flat-buffer tensors (shaped VAL_BUFFER, # #275) the neural policy's buffer path requires. -ARG EIGS_REF=v0.30.0 +ARG EIGS_REF=v0.32.0 RUN mkdir -p /opt/eigenscript && cd /opt/eigenscript && git init -q \ && git remote add origin https://github.com/InauguralSystems/EigenScript.git \ && git fetch -q --depth 1 origin "${EIGS_REF}" && git checkout -q FETCH_HEAD \ diff --git a/CLAUDE.md b/CLAUDE.md index b7012bd..f0b9299 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -68,7 +68,7 @@ make gfx # graphical binary for the playable game (dlopens SDL2 at run weights/obs as flat shaped buffers (`buffer of [r, c]`, shaped `VAL_BUFFER` #275), which land in v0.19.0. (Background music needs `audio_music_*` from 0.18.0; the older `spawn`/`recv_timeout`/`audio_play_loop` niceties are - v0.13.0.) **CI and the devcontainer pin v0.26.0** (`.devcontainer/Dockerfile` + v0.13.0.) **CI and the devcontainer pin v0.32.0** (`.devcontainer/Dockerfile` `EIGS_REF`). The flat-buffer forward is ~7× faster than the old nested-list form and the trainer's backprop ~1.8×, both byte-identical; `models/policy.txt` stays format-compatible. Validated headless (regressions, obs-stacking,