Alex Wells

I think we left the batched version alone (not using hashes where possible) mostly to just limit the scope of the code changes with the original hashcode changes which were...

> There's now more code to support ustringhash_pod for batched. Some string variables from .osl files get passed as ustrings into liboslexec/wide functions while ustringhash is expected, I didn't find...

Just one comment on converting to hashcode vs. ustrings, for "real" string operations that are not constant folded (CPU only), there will be more overhead when we do the lookup...

I am able to reproduce the problem, think the caching of get_userdata might be broken vs. obtaining the value from the renderer.

Adding `--options dump_forced_llvm_bool_symbols=1 ` we can see the results of the Batched Analysis: `face_idx is forced llvm bool.` Which means face_idx can be only be 0 or 1. When an...

Its a bug in Batched Analysis, it is not properly considering get_userdata could write any integer value to face_idx. It is only looking at the lack of writers to face_idx...

I have a fix now, just need to clean it up, should be able to post PR tomorrow.

This issue should be fixed by https://github.com/AcademySoftwareFoundation/OpenShadingLanguage/pull/1949

About the union approach. I don't think the solution to C++ undefined behavior is to rely on another C++ undefined behavior even if it "appears" to work for most compilers....

I've done some prototyping to explore implementing dynamic index array subscript support and created a godbolt project ontop of Imath that has a wrapper class Vec3Subscript where enum class Subscript...