[DGD] hooking jit into the dgd system and modularity

[Raymond Jennings]

As long as you decouple the generation of LLVM from the parsing of LPC
source you should be fine.

My advice, is to generate the JIT as needed on a per function+control
block basis, and to invalidate the associated JIT when a control block
is recompiled.

Some caveats to be aware of:

1.  It is possible to have DGD bytecode without corresponding LPC
source code (think warm booting)
2.  LPC-side error handling often makes use of line number information
extracted from call_trace() kfun

A possible option is to stuff the code in question in a separate .cpp file.

On Wed, Apr 12, 2017 at 11:05 AM, Carter Cheng <cartercheng at gmail.com> wrote:
> I actually looking for some advice as to how best to do this since I am a
> rather inexperienced programmer. To just get it to the point where it is
> working my first take is to include some call code to the llvm code cache
> of the jit to update in sdata.cpp and in the compiler code whenever code is
> loaded from disk or when code is freshly compiled respectively. The problem
> with this is it seems to generate an odd dependency between the
> compiler/sdata code and the jit which is not really "semantically"
> necessary I suspect.
>
> My thoughts on this is to perhaps create two new functions that each handle
> the separate cases of the compiler hook and the sdata hook and have that
> invoke the compiler code and sdata control block loading individually. But
> where do should these functions be placed to avoid creating the artificial
> dependency?
>
> Regards,
>
> Silenus
> ____________________________________________
> https://mail.dworkin.nl/mailman/listinfo/dgd