A long overdue change is now committed dicussed around more then a year ago.
The short story:
* interpreter->code is now a PackFile_ByteCode pointer
* the "shortcut" pointers to prederef or jit_info are gone
* switching a bytecode segment for a function call or return is now basically just one C instruction:
interpreter->code = prev_segment; // return only
The longer story:
interpreter->code used to be a PackFile pointer. The PackFile had shortcuts to byte_code or const_table. But code segments are created or loaded in terms of PackFile_ByteCode, which has it's own pointers to code start, constant table and what not.
We had a mixture of both bytecode addressing schemes and additional shortcuts to get rid of some indirections. This caused a rather complicated sequence to switch to another code segment. See Parrot_switch_to_cs() (or the diff thereof).
Unfortunately but not surprisingly interpreter->code is used all over the interpreter and in platform code like JIT or EXEC core. This makes the patch rather big and worse: I couldn't test all the changes.
I did replace all the obvious pieces in JIT and EXEC code, but I might have missed some so that it won't even compile. Needed changes are basically:
interpreter->code->byte_code => interpreter->code->base.data interpreter->jit_info => interpreter->code->jit_info
But some of the former might be hidden in assembly generating code or even macros.
I've tested it on i386/linux and ppc/OS X. If some other platforms stop compiling or JIT doesn't work anymore please have a look at changes for i386 or ppc - or just mail a bug report.
Sorry for the inconvenience, leo
