[Detail Bug] Parser v2: Empty module-def IDs leak into allocator/argument indexes

Detail Bug Report

https://app.detail.dev/org_62aa40f5-2c23-4914-a665-3bb2068af20e/bugs/bug_214c29be-8766-4126-b6be-31f37050e347

Introduced in 12d59c28aed1f86872a3abd2ffc9285eac62c8e5 by @kewde on Jun 28, 2026

Summary

• Context: ApplicationIndexer builds lookup tables (module_defs, allocators, arg_alloc, parameter_refs, …) from an ApplicationProgram so downstream code in dynamic.py, encode.py, and EvalContext (parser_v2) can do id-keyed lookups by module-def id.
• Bug: In _index_module_def, the if md.id: guard that skips empty/None-id module defs is applied to module_defs only. The two sibling registrations — allocators and arg_alloc — use md.id as the dict key with no such guard, so an empty-id ModuleDef that carries an allocators or arguments block is registered under the "" key in those dicts even though it was deliberately dropped from module_defs.
• Actual vs. expected: Expected: empty-id module defs are not registered anywhere (as documented by test_module_def_with_empty_id_is_not_registered). Actual: module_defs omits them, but allocators[""] / arg_alloc[""] are created when the module def has allocators/arguments.
• Impact: Inconsistent index state and a test gap. The leaked "" entries are currently not reachable by production read paths (allocation is gated by module_defs presence), but this will surface as soon as any code iterates or consumes allocators/arg_alloc by module-def id without re-checking module_defs.

Code with Bug

def _index_module_def(self, md: ModuleDef) -> None:
    if md.id:                                  # guard applied here (line 124)
        self.module_defs[md.id] = md
    ...
    if md.static.allocators is not None:
        self.allocators[md.id] = {             # <-- BUG 🔴 no `md.id` guard; writes `""` key
            a.id: Allocator(id=a.id, start=a.start, max_inclusive=a.max_inclusive)
            for a in md.static.allocators.allocator
        }
    if md.arguments is not None:
        self.arg_alloc[md.id] = {              # <-- BUG 🔴 no `md.id` guard; writes `""` key
            a.id: (a.allocates if a.allocates is not None else 1, a.alignment.value)
            for a in md.arguments.argument
        }
    if md.sub_module_defs is not None:
        for sub in md.sub_module_defs.module_def:
            self._index_module_def(sub)        # same gap reached recursively for empty-id sub-module defs

Explanation

• Empty module-def IDs are schema-legal, and ApplicationIndexer intentionally skips them via if md.id: when populating module_defs.
• The same contract is not applied to allocators and arg_alloc, so an empty-id ModuleDef with allocators/arguments populates those dicts under the "" key.
• This is demonstrable by extending the existing empty-id test to include allocators/arguments: idx.module_defs == {} passes, but idx.allocators and idx.arg_alloc contain "".

Codebase Inconsistency

• dynamic.py only creates ModuleNode when ref_id is present in idx.module_defs, and encode.py errors on unknown module-def IDs. This gating makes the "" entries dead state today, but it also highlights the intended invariant: allocation data should be keyed only by registered module-def IDs.

Recommended Fix

• Apply the same if md.id: guard to allocator/argument indexing (or fold all per-md registrations into a single if md.id: block), and extend test_module_def_with_empty_id_is_not_registered to also assert idx.allocators == {} and idx.arg_alloc == {}.

History

This bug was introduced in commit 12d59c28. The commit "feat: add allocator support" extended _index_module_def with two new md.id-keyed registrations (allocators and arg_alloc) but added them outside the pre-existing if md.id: guard. A later commit (1b84e21) reformatted via ruff format without logic changes.

Start at ApplicationIndexer._index_module_def and the existing test_module_def_with_empty_id_is_not_registered test. Cover empty-id module definitions with allocators and arguments, then run the relevant parser_v2 tests; done means module_defs, allocators, and arg_alloc all omit the empty ID, including recursively indexed sub-module definitions.