From: Clif F. <cl...@cf...> - 2010-04-05 18:32:24
|
On Mon, Apr 05, 2010 at 07:24:26PM +0100, Donal K. Fellows wrote: > On 05/04/2010 18:02, Guilherme Polo wrote: >> To keep it short, my idea for a gsoc project would involve creating a >> machine code generator based on Tcl bytecodes so in future projects it >> would be possible to do the other tasks related to a good JIT compiler >> like: select what to compile (a tracing JIT could be a good idea), >> when to compile, recompile, deoptimize, and etc.. >> >> What do you think ? Stop it and go use LLVM or something else ? > > I think "go for it!" I think there's a substantial fraction (i.e. well > over 50%) of Tcl programs that can be usefully compiled this way, but > even experimental work here is really valuable. > My first thought was that I'm aware of at least 2 compile-to-machine code Tcl projects that floundered because of the dynamic nature of Tcl. But I agree with Donal - at least 80% of the code I've written doesn't create new procs on the fly or otherwise need self-modifying code type solutions. If you try to solve 80% of the applications, instead of supporting the entire Tcl capability set, it's a solveable problem. Being able to compile a set of Tcl support libraries to C/Object code could be a big gain for Tcl speed. I say go for it. Clif -- ... Clif Flynt ... http://www.cwflynt.com ... cl...@cf... ... .. Tcl/Tk: A Developer's Guide (2nd edition) - Morgan Kauffman .. .. 17'th Annual Tcl/Tk Conference: 2010, Oak Brook, IL USA .. ............. http://www.tcl.tk/community/tcl2010/ ............ |