Summary

PR #659's template-warning suppression pass in vera/codegen/core.py::compile_program keys compiled_mono_bases on bare base function names. If two modules in the same compile unit declare a forall<T> function with the same name, the suppression set conflates them — if either compiles, BOTH templates' warnings get suppressed.

Reproducer (hypothetical)

-- prelude:
private forall<T> fn map(@Option<T>, ...) -> ...

-- user module:
public forall<T> fn map(@Array<T>, ...) -> ...

If only one of these has working mono clones in the user's program, the other's template warning gets suppressed too (because compiled_mono_bases contains "map" without module disambiguation).

Current mitigations

• Today's user-vs-prelude namespace separation: users can't redeclare prelude names like option_map, etc. But user-defined module imports CAN share names.
• The failure mode is "warning suppressed when it shouldn't be" — not a runtime bug. The actual functions still compile correctly; users just lose the per-function diagnostic.

Source location

vera/codegen/core.py around lines 500-580. The relevant code:

compiled_mono_bases: set[str] = set()
for mdecl in mono_decls:
    orig_name = mdecl.name.split("$")[0]
    # ... compile ...
    if fn_wat is not None:
        compiled_mono_bases.add(orig_name)

orig_name is the bare base name ("map" from "map$Int"). No module qualification.

Suggested fix

Key on (module, name) rather than just name. Requires tracking module attribution through Pass 1.5's mono-decl synthesis. Alternatively, scope the suppression to the current program's forall_decl_names set only (which excludes imported module declarations).

Investigation context

Surfaced as F4 in the multi-agent review of PR #659. Not blocking for #659 — the failure mode is benign (silenced warning) and only reachable when users intentionally shadow names across modules.

Acceptance criteria

• Reproducer (two modules each declaring forall<T> fn shared_name) compiles correctly, and BOTH templates emit warnings if their mono clones don't compile
• Existing 4 tests in tests/test_codegen.py::TestGenericMonoSuffixFromSlotRef604 still pass

References

• PR Fix mono-suffix bug for FnType-alias SlotRef args + release v0.0.145 #659 — introduced compiled_mono_bases
• audit recommendation 2 from #604 investigation comment — the original "suppress when at least one mono compiles" idea