normalperson / yhbt.net wrote:
>  But I am thinking of trying an entirely different approach to this
>  optimization (also able to target non-cfunc)

I had an idea to do lazy/late rb_str_resurrect before ci->call by
defining + setting a STR_LITERAL flag (FL_USER19), but that is also
complicated and I haven't been able to implement it, yet.

The current patches can only handle cfunc on core classes, and even
handling of Time (T_DATA) is ugly.  I'm afraid this type of transparent
optimization is not effective for real-world performance unless it can
be applied more broadly to non-core classes.