interop: always hand a C++ bind "" for a null daslang string

[aleksisch]

A daslang string is null when empty. Whether a bind taking const char * received that null depended on whether its translation unit included ast/ast_typefactory_bind.h, which is where the cast_arg specializations substituting "" lived — fine while the call node was per bound function, broken once it became per signature: one game link has 27 objects defining the node for void (*)(char const *), 10 with the substitution and 17 without, and object order picks the winner. ImGui::PushID got a nullptr and crashed in ImHashStr. Nothing diagnoses it — it is an ODR violation, and ld.gold --detect-odr-violations and gcc -Wodr are both silent.

So substitute unconditionally, in all three lanes:

• cast_arg specializations move beside the generic in simulate/interop.h; the AOT and JIT emitters stop asking whether the callee opted in, which retires needStringCast
• the argument null-checks that can no longer fire go: module_builtin_string trades 44 stringLengthSafe for stringLength and drops 22 guards. What stays is what the substitution does not reach — a string taken by reference, non-bind helpers, and guards on lengths and FILE *
• ast_typefactory_bind.h is left empty, so typeFactory for unbound callbacks moves next to typeFactory<TT *> in ast_typedecl.h (that order matters: above it, GCC rejects the partial specialization), cast_arg<das::string> moves to interop.h, and the header, its 102 includes and the binder line emitting it all go

Cost, from the second commit's benchmark — two binds, four const char * against four void *, same body and node: 10.3 vs 10.1 ns/op interpreted, 0 allocations either way.

Two notes for review:

1. Removing the flag bit leaves a hole in MoreFunctionFlags, which rtti exposes and the AST module cache serializes — may want a cache version bump.
2. Empty results still come back as null (allocateString cannot allocate an empty string), but a null crossing into a bind and back — as through _temp_string_result, wrapped around every [temp_string_result] builtin — returns as "". So is_null_str in the byte-view pins measures length now, and one assertion asking whether a NUL-holding run was allocated at all is gone: das can no longer tell. Covering that again needs a C++-side test calling the builtin directly.

Out-of-tree generated bindings include the deleted header, so this needs borisbat/dasImguiImplot#30 and borisbat/dasImguiNodeEditor#46 first — fatman and wasmboy clone their default branches.

[borisbat]

way too slow.

[borisbat]

this is going to be slower than original implementation, which uses cast<>
cast gets inlined on C++ side, this creates interpreter node - which is a lot slower
even blanket cast_arg<char *> would be faster

[aleksisch]

Now we simply wrap it everywhere on das -> cpp path

[Copilot]

🔵 Needs a closer look

It changes core interop null-string semantics for every string bind across the engine and removes a public installed header, a high-impact cross-cutting change that warrants final human review despite being well-tested.

Pull request overview

This PR moves the "null daScript string → empty C string" conversion for interop binds from a header-wide, per-TU mechanism into the interpreter's simulate path, keyed on the bind's needStringCast flag — matching how the AOT (das_string_cast in aot_cpp.das) and JIT (llvm_jit.das) backends already behave. Previously the conversion lived in cast_arg<char*>/cast_arg<const char*> specializations inside ast_typefactory_bind.h, which applied to every bind in any TU that happened to include that header and to none elsewhere — an unstable, link-order-dependent behavior (the motivating ImGui::PushID crash). The header is deleted and its still-needed pieces relocated.

Changes:

• Add SimNode_StringArgNotNull and wrap string arguments of needStringCast binds in sv_simulateCall, on the same predicate the AOT/JIT tiers use.
• Delete ast_typefactory_bind.h, relocating typeFactory<ResT(*)(Args...)> into ast_typedecl.h and cast_arg<das::string> into interop.h; drop its include from ~90 module/tutorial/example TUs, the cbind generator, and the install list.
• Add a UnitTest bind pair (test_string_arg_length / _cast) and tests/handle_types/string_arg_never_null.das covering both halves.

File summaries

File	Description
include/daScript/simulate/simulate_nodes.h	New SimNode_StringArgNotNull node (null → "").
src/simulate/simulate_visit.cpp	Visitor plumbing for the new node (matches KeepAlive).
src/ast/ast_simulate.cpp	Wraps needStringCast string args in sv_simulateCall.
include/daScript/simulate/interop.h	Adds relocated cast_arg<das::string>.
include/daScript/ast/ast_typedecl.h	Adds relocated typeFactory<ResT(*)(Args...)>.
include/daScript/ast/ast_typefactory_bind.h	Deleted (specializations relocated / dropped).
modules/dasUnitTest/test_handles.cpp	Adds testStringArgLength bound with and without needStringCast.
tests/handle_types/string_arg_never_null.das	New test covering cast, non-cast, empty, null, non-empty.
CMakeLists.txt	Removes deleted header from release AST include list.
modules/dasClangBind/cbind/cbind_boost.das	Stops emitting the deleted header in generated binds.
~90 module/tutorial/example .cpp files	Drop the now-unused header include.

One minor, non-blocking note (not filable — the file isn't in the diff): tests/README.md asks that every .das under tests/ be listed, but its handle_types/ section is already stale (missing several files); consider adding string_arg_never_null.das there.

Review details

• Files reviewed: 112/112 changed files
• Comments generated: 0
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

changed been made