This does seem like a sane approach – and a much more direct way to get high performance Python than what PyPy does (i.e. the second Futamura projection<http://en.wikipedia.org/wiki/Partial_evaluation#Futamura_projections>). This kind of thing is really hard to pull off for an existing dynamic language, however – especially one is complex as Python (c.f. Lua). V8 certainly proves that it can be done, but for that project Google hired Lars Bak and a team of people who already had a lot of experience with implementing successful high-performance VMs like Smalltalk, Self and Java. In general, the extremely fractured nature of the Python ecosystem – especially when it comes to performance – seems a bit toxic and this just adds to that. Arguably, this should have been the first high-performance Python project, not the Nth, where N has gotten rather large. You might be able to get your Python to run fast, but your first problem is figuring out which non-standard Python implementation and/or package to try.
On Fri, Apr 4, 2014 at 3:59 PM, Jake Bolewski <[email protected]>wrote: > I think that this is great news for Julia, especially if Dropbox puts > serious engineering effort into this project. Julia is dynamically typed > just like Python, so all the high level optimizations PySton needs to make > a Python LLVM JIT fast will be aplicable to Julia. Both will sit ontop of > LLVM's MCJIT and they will likely have to contribute patches to make this > project sucessful, which will again benefit Julia. It looks like they are > going to attempt interesting things like tiered compilation, escape > analysis for GC'd memory, and backpatching of JIT'ed code, all things Julia > does not do at the moment. > > -Jake > > > > > On Friday, April 4, 2014 2:46:08 PM UTC-4, Cristóvão Duarte Sousa wrote: >> >> Once again pythonistas feel the need for a single high-level-high-performance >> language: >> https://tech.dropbox.com/2014/04/introducing-pyston-an- >> upcoming-jit-based-python-implementation/ >> >
