fix(cache): key the object cache on all codegen-affecting env vars (#6394)

[proggeramlug]

Problem

The object cache (node_modules/.cache/perry) keys on source + compiler identity + a curated set of codegen-affecting environment variables — but that set was incomplete. Nine vars were keyed (PERRY_WRITE_BARRIERS, PERRY_SHADOW_STACK, PERRY_TARGET_CPU, …), while ~16 other PERRY_* vars that perry-codegen reads at compile time — and that change the emitted .o — were not. Setting any of those against a warm cache silently reused an object built without it: the switch appears to do nothing, and any A/B or bisection built on it is vacuous.

(Contrary to the issue title, PERRY_WRITE_BARRIERS itself is already keyed — the env block exists. This PR closes the remaining gap so that no codegen env var can silently no-op with a warm cache.)

Fix

Add the missing compile-time codegen env vars to compute_object_cache_key_with_env
(crates/perry/src/commands/compile/object_cache.rs):

PERRY_TYPED_FEEDBACK, PERRY_TYPED_FEEDBACK_TRACE, PERRY_FULL_OUTLINE_IC,
PERRY_FULL_OUTLINE_IC_MIN_FUNCS, PERRY_OUTLINE_METHOD_DISPATCH, PERRY_INLINE_NEW,
PERRY_INLINE_CTOR, PERRY_STRING_INIT_CHUNK_SIZE, PERRY_LL_O0_THRESHOLD_BYTES,
PERRY_LL_SIZE_OPT, PERRY_LL_SIZE_OPT_MAX_FN_BYTES, PERRY_ENTRY_SYMBOL,
PERRY_CODEGEN_UNITS, PERRY_CODEGEN_UNIT_SIZE, PERRY_SETJMP_VOLATILE,
PERRY_GC_MOVING_LOOP_POLLS.

Each is a std::env::var read in perry-codegen that changes emitted IR or the
.ll→.o compile decision (typed-feedback site emission, IC inline/outline, method-dispatch
outlining, inline new/ctor, string-init chunking, per-unit -O0/-Os selection, entry-symbol
rename, codegen-unit partitioning, volatile-setjmp promotion, and loop GC-poll emission).

Deliberately excluded

• Runtime globals, not compile-time env: PERRY_CLASS_FIELD_INLINE_GUARD_DISABLED,
  PERRY_TA_VIEW_GUARD, PERRY_TA_KIND_CACHE — codegen emits loads of @PERRY_* globals
  resolved at run time, so the emitted bytes are identical regardless of the compile-time
  environment.
• Diagnostics / side files only: PERRY_LLVM_KEEP_IR, PERRY_SAVE_LL, PERRY_NATIVE_REPS*,
  PERRY_NO_CLANG_PROBE.
• Final-link step, not per-module .o: PERRY_LD.
• Already captured indirectly: PERRY_WATCHOS_ARM64_32 (folds into the resolved target
  triple, already hashed as tgt); PERRY_GLOBAL_SCRIPT_THIS / PERRY_EVAL_CSP /
  PERRY_ALLOW_UNIMPLEMENTED (affect AST→HIR lowering, captured by the hir fingerprint).

Over-inclusion only costs an unnecessary cache miss when a rarely-set var is present;
under-inclusion is the silent-wrong-result bug — so the classification errs toward inclusion
for anything that changes emitted bytes.

The auto-optimize archive cache (target/perry-auto-*, keyed by
auto_optimized_cache_key) is a separate mechanism that keys the Rust build of the
runtime/stdlib static archives, not the per-module TS .o; it does not share this blind spot
and is out of scope.

Test

Extends key_changes_with_codegen_env_vars to assert every newly-keyed var (unset vs set vs a
different value) produces a distinct key, via the existing compute_object_cache_key_with_env
seam (no process-environment mutation).

Closes #6394

Summary by CodeRabbit

• Bug Fixes
  • Improved build cache accuracy by including additional code-generation and tracing-related environment settings in the object-cache key.
  • Prevented reuse of previously cached object files when compile-time tuning options differ between builds.
• Tests
  • Expanded cache invalidation tests to cover more code-generation configuration environment variables, verifying distinct cache keys for different unset/set combinations.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 88c8dbf6-9137-482e-a30f-95f875829146

📥 Commits

Reviewing files that changed from the base of the PR and between de8a087 and 20b143e.

📒 Files selected for processing (2)

• crates/perry/src/commands/compile/object_cache.rs
• crates/perry/src/commands/compile/object_cache/object_cache_tests.rs

🚧 Files skipped from review as they are similar to previous changes (1)

• crates/perry/src/commands/compile/object_cache/object_cache_tests.rs

---

📝 Walkthrough

Walkthrough

The object-cache key now incorporates additional compile-time perry-codegen environment variables, and its invalidation test covers the expanded set of tuning and emission controls.

Changes

Codegen cache key invalidation

Layer / File(s)	Summary
Hash codegen settings and test invalidation crates/perry/src/commands/compile/object_cache.rs, crates/perry/src/commands/compile/object_cache/object_cache_tests.rs	compute_object_cache_key_with_env hashes additional codegen environment variables, while key_changes_with_codegen_env_vars verifies that setting or changing each variable produces distinct cache keys.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: andrewtdiz

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is concise and accurately summarizes the main cache-key change.
Description check	✅ Passed	The description covers the problem, fix, exclusions, and test approach, though it doesn't use the exact template headings.
Linked Issues check	✅ Passed	The change keys the cache on the added compile-time codegen env vars and excludes runtime-only ones, matching #6394's objective.
Out of Scope Changes check	✅ Passed	The patch stays focused on cache-keying and the corresponding test updates, with no clear unrelated code changes.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/object-cache-key-codegen-env-vars

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}