[apotema]

Closes #540.

Pre-existing latent robustness issues in the flow/script scan (surfaced during the #539 split, present verbatim on main before it): three paths swallowed error.OutOfMemory, violating the module's documented "OOM is a hard failure" goal. All three are behavior changes on the OOM path only — normal codegen output is byte-identical (golden suite unchanged).

The three sites

1. src/codegen/scan/flow_decls.zig — the plugin + game-script passes did scanFlowDeclsInSource(...) catch continue, masking an allocation failure as a tolerated per-file parse failure. A swallowed OOM silently leaves the generated flow registry incomplete → wrong codegen with no error. Fix: catch |err| switch (err) { error.OutOfMemory => return err } at both call sites.

   • Note discovered while implementing: the scan's inferred error set is exactly error{OutOfMemory} — std.zig.Ast.parse records parse errors in the AST rather than raising them, so there was never a genuine per-file parse error to tolerate; the old continue only ever swallowed OOM. The switch is therefore exhaustive (no else); if the scan ever grows a tolerable error it will stop compiling and force a deliberate else => continue.

2. src/codegen/scan/promote.zig — a freshly-duped rel_path leaked if out.append(...) OOMed before ownership transferred to the list (the function-scope errdefer walks out.items, which doesn't yet include the pending entry). Fix: errdefer allocator.free(rel_path); right after the dupe.

3. src/root/game_shim.zig (from the refactor: split root.zig into focused sub-modules (behavior-preserving) #542 root split) — arr_list.toOwnedSlice(allocator) leaked the buffer on OOM: toArrayList() reset the writer so the function-scope errdefer aw.deinit() covers nothing. Fix: errdefer arr_list.deinit(allocator); before the toOwnedSlice (safe — the reset writer's deinit is a no-op).

Tests

FailingAllocator regression tests (the toolkit's failing-allocator idiom) for each site:

• discoverPluginFlowDecls: OOM propagates, never a silent partial scan — drives a game script with a real FlowNodes block through a FailingAllocator at every alloc index; asserts each forced failure surfaces as error.OutOfMemory and never returns a value with a truncated node set (exactly what the old catch continue produced). Balances freed==allocated at every point.
• collectPromotedScripts: no allocator leak at any OOM point
• generateGameShim: no allocator leak at any OOM point

Bar

• zig build + zig build test green.
• Golden suite unchanged (zero diff — only OOM paths changed; no output-producing code touched).
• zig fmt --check clean.

https://claude.ai/code/session_017pW3ifKf9wgxNg4viy6okw

Summary by CodeRabbit

• Bug Fixes

  • Improved handling of out-of-memory failures so discovery and generation steps now fail cleanly instead of producing partial results.
  • Fixed a memory leak in script promotion when allocation fails during list updates.
  • Fixed a memory leak during game shim creation when final buffer conversion fails.

• Tests

  • Added regression coverage for out-of-memory scenarios, including checks that cleanup fully releases allocated memory.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR fixes out-of-memory handling in three scan/shim-generation modules: flow_decls.zig now propagates error.OutOfMemory instead of swallowing it in catch continue, promote.zig adds an errdefer to prevent a rel_path leak on append failure, and game_shim.zig adds an errdefer to prevent an arr_list leak on toOwnedSlice failure. Each fix is accompanied by a FailingAllocator-based regression test.

Changes

OOM propagation and leak fixes

Layer / File(s)	Summary
Flow decl scan: propagate OOM instead of catch continue src/codegen/scan/flow_decls.zig	Plugin and game-script scan loops rethrow error.OutOfMemory via exhaustive switch instead of catch continue; a new test forces allocation failures at every index and asserts either complete results or error.OutOfMemory, with balanced allocated/freed bytes.
Promote scan: fix rel_path leak on append OOM src/codegen/scan/promote.zig	Adds errdefer allocator.free(rel_path) after allocation, before out.append; a new test forces allocation failures at every index and asserts no leaks.
Game shim generation: fix arr_list leak on toOwnedSlice OOM src/root/game_shim.zig	Adds errdefer arr_list.deinit(allocator) before toOwnedSlice; a new test forces allocation failures at every index and asserts balanced allocated/freed bytes, accepting error.OutOfMemory or error.WriteFailed.

Estimated code review effort: 2 (Simple) | ~12 minutes

Poem

Three small burrows, each with a leak,
now sealed tight so OOM can't sneak.
catch continue? Not on my watch,
errdefer stands as the guarding notch.
Hop, hop, hooray — memory's safe today! 🐇

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Out of Scope Changes check	⚠️ Warning	src/root/game_shim.zig adds an extra OOM leak fix and test that are not part of linked issue #540.	Either move the game_shim OOM fix into a separate PR or add it to the linked issue scope with explicit justification.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: hard-failing OOM handling and a leak fix in scan paths.
Linked Issues check	✅ Passed	The PR addresses #540 by propagating OOM in flow-decls, freeing rel_path on append OOM, and adding failing-allocator regression tests.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/scan-oom

---

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

[gemini-code-assist]

Code Review

This pull request fixes several out-of-memory (OOM) handling issues across the codebase. In flow_decls.zig, OOM errors are now correctly propagated as hard failures instead of being silently swallowed. In promote.zig and game_shim.zig, local errdefer blocks have been added to prevent memory leaks when OOM occurs during list appending or slice conversion. Regression tests using FailingAllocator have been added for each of these cases to ensure proper OOM propagation and leak prevention. There are no review comments, and I have no feedback to provide.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (2)

src/codegen/scan/flow_decls.zig (2)

255-262: 🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

Propagate extractConstructsString allocation failures. allocator.dupe(...) catch null still turns error.OutOfMemory into “no .constructs field,” so the scanner can silently drop a real value on memory pressure. Make this helper return !?[]u8 and try the duplicate so OOM bubbles up like the rest of the scan.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/codegen/scan/flow_decls.zig` around lines 255 - 262,
`extractConstructsString` is swallowing allocation failures by converting
`allocator.dupe(...)` errors into `null`, which makes a real `.constructs` value
look absent under memory pressure. Update this helper to return `!?[]u8`, use
`try` for the duplicate, and propagate `error.OutOfMemory` through the caller
chain in `scan/flow_decls.zig` so the scan fails consistently instead of
silently dropping the value.

---

510-512: 🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

cache.resolvePlugin(...) catch continue should not swallow OOM src/cache/resolve.zig allocates and joins paths here, so error.OutOfMemory is reachable; let this fail hard instead of silently skipping the plugin from the flow registry.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/codegen/scan/flow_decls.zig` around lines 510 - 512, The plugin
resolution loop is swallowing allocator failures by using catch continue on
cache.resolvePlugin, which can hide error.OutOfMemory. Update the flow registry
logic in the for (cfg.plugins) loop around plugin_dir so only expected non-fatal
resolution errors are skipped, and let OutOfMemory propagate instead of
continuing silently; use the cache.resolvePlugin and plugin_dir handling in this
block to locate the fix.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Outside diff comments:
In `@src/codegen/scan/flow_decls.zig`:
- Around line 255-262: `extractConstructsString` is swallowing allocation
failures by converting `allocator.dupe(...)` errors into `null`, which makes a
real `.constructs` value look absent under memory pressure. Update this helper
to return `!?[]u8`, use `try` for the duplicate, and propagate
`error.OutOfMemory` through the caller chain in `scan/flow_decls.zig` so the
scan fails consistently instead of silently dropping the value.
- Around line 510-512: The plugin resolution loop is swallowing allocator
failures by using catch continue on cache.resolvePlugin, which can hide
error.OutOfMemory. Update the flow registry logic in the for (cfg.plugins) loop
around plugin_dir so only expected non-fatal resolution errors are skipped, and
let OutOfMemory propagate instead of continuing silently; use the
cache.resolvePlugin and plugin_dir handling in this block to locate the fix.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3f1d3bb0-02d4-4231-a557-441c261c13e3

📥 Commits

Reviewing files that changed from the base of the PR and between 1142b69 and 6f6f9dd.

📒 Files selected for processing (3)

• src/codegen/scan/flow_decls.zig
• src/codegen/scan/promote.zig
• src/root/game_shim.zig