From dc6463c658ba506b83d063001cd50df55e698c9f Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Ralph=20K=C3=BCpper?= Date: Thu, 12 Nov 2026 07:22:36 +0100 Subject: [PATCH] changelog: fragment for #14526 (main hygiene) --- changelog.d/14526-main-hygiene.md | 1 + 1 file changed, 1 insertion(+) create mode 100644 changelog.d/14526-main-hygiene.md diff --git a/changelog.d/14526-main-hygiene.md b/changelog.d/14526-main-hygiene.md new file mode 100644 index 0000000000..d04286e719 --- /dev/null +++ b/changelog.d/14526-main-hygiene.md @@ -0,0 +1 @@ +- **Repo checks**: main's own gates are green again. `COMPILE_ENV_VARS` (the build-cache and object-cache env inputs) is sorted and unique as documented, so its audit runs again; it found eleven unregistered compiler env reads: `PERRY_FACTS_DCE`, `PERRY_INDEX_RAW_I32`, `PERRY_SNAP_CENSUS`, `PERRY_WP_ELEMENT_READS` and `PERRY_WP_REP_P2` change emitted code and are now cache inputs (a build with one of them set could reuse a cached object made without it), six diagnostic ones are excluded with a reason. The order of the list is part of both cache keys, so existing compile caches miss once. Also: `scripts/check_file_size.sh` allowlist counts re-recorded, `scripts/check_gc_header_constants.py` registers the method and field guard masks as 39359ae502 split them (the compiler was right, the registry stale), the wasm32 runtime ABI table regenerated, the unrooted-local baseline ratcheted down, and the `held_lengths` / `named_lengths` unit tests assert the `Arr` precondition and the refusal reason with positive twins. No lowering change.