00:41 <sb0> the _ prefixes are one thing, the main issues is the inline transform does too many things at the same time (inlining functions, encoding RPC/exceptions, resolving references, initializing variables, creating a single flat namespace) and things get complicated

00:43 <sb0> perhaps a good option is have transform #1 do the function inlining only and use some placeholders for all the references, that contain enough information for transform #2 to replace them properly

02:19 <rjo> sb0: hmm. one would think that e.g. llvm has a similar problem. they must be doing quite a bit of metadata annotation to transfer information between the passes.

02:34 <sb0> yeah, that should be doable... and it's possible to attach arbitrary objects to python ast nodes, even use arbitrary objects as the nodes themselves