Pyston Now 95% Faster Than CPython, But Dropbox Just Stopped Supporting It

[phoronix]

Phoronix: Pyston Now 95% Faster Than CPython, But Dropbox Just Stopped Supporting It

Back in 2014 Dropbox announced the Pyston project as an open-source JIT compiler to Python focusing upon maximum performance. With this newest Pyston release (v0.6.1) they are now 95% faster than CPython, but Dropbox is ending their involvement in the project...

http://www.phoronix.com/scan.php?page=news_item&px=Pyston-Dropbox-Stops

[Azrael5]

excellent news.

[chuckula]

So you're saying DropBox Dropped it like it was hot?

[timofonic]

So you're saying DropBox Dropped it like it was hot?

Maybe it was too good for them. Now they need to reinvent the wheel with Go.

These are the wonders of proprietary software

[defaultUser]

Maybe it was too good for them. Now they need to reinvent the wheel with Go.

These are the wonders of proprietary software

It appears more to be letting go a hopeless cause...

Not only go but also rust (https://news.ycombinator.com/item?id=11283688, https://blogs.dropbox.com/tech/2016/...-magic-pocket/)

[JPFSanders]

In my opinion the Python++ people should be backing is Nuitka

[stephen82]

I have mixed feelings right now for this announcement.

First of all, Guido van Rossum works for DropBox who happens to be a person that surely knows more than the rest of us around Python. If I were him and heard such an announcement, not about the abandoning of Pyston, but choosing Go over other tools like Cython, Nuitka, and PyPy I would have got offended for sure, seriously now!

Another excellent scientist named Jessical McKellar works for DropBox as well who happens to know far too much about how to push the limits of Python AND the Linux kernel.

My mind tells me that this has something to do with the Google announcement about grumpy, thus their choice of using Go for their most critical parts. I understand that as a company they are trying to keep their costs at the lowest level possible, which means with the use of grumpy they can convert their most critical parts from Python to Go and they could gain incredible performance improvements by orders of magnitude, while using the least hardware possible to produce the same results as before.

Of course, no one would have used such a powerful tool such as Go for the sake of maintaining the same old mediocre performance Python code could produced WITHOUT important customizations to their algorithms and / or to its inner guts.

I have a feeling that in the past some of them inside DropBox have read the amazing article "How we went from 30 servers to 2" and has played a role on their decision to switch from Python to Go eventually.

For some reason though, my instinct tells me that it won't take a long time we will hear or read about Guido's departure from DropBox. This announcement shows me or at least that's the impression I'm getting that they are pushing their developers towards quantity over quality.

Maybe I'm imagining things...well, I hope I'm wrong ¯\_(ツ)_/¯

[defaultUser]

I have mixed feelings right now for this announcement.

First of all, Guido van Rossum works for DropBox who happens to be a person that surely knows more than the rest of us around Python. If I were him and heard such an announcement, not about the abandoning of Pyston, but choosing Go over other tools like Cython, Nuitka, and PyPy I would have got offended for sure, seriously now!

Another excellent scientist named Jessical McKellar works for DropBox as well who happens to know far too much about how to push the limits of Python AND the Linux kernel.

My mind tells me that this has something to do with the Google announcement about grumpy, thus their choice of using Go for their most critical parts. I understand that as a company they are trying to keep their costs at the lowest level possible, which means with the use of grumpy they can convert their most critical parts from Python to Go and they could gain incredible performance improvements by orders of magnitude, while using the least hardware possible to produce the same results as before.

Of course, no one would have used such a powerful tool such as Go for the sake of maintaining the same old mediocre performance Python code could produced WITHOUT important customizations to their algorithms and / or to its inner guts.

I have a feeling that in the past some of them inside DropBox have read the amazing article "How we went from 30 servers to 2" and has played a role on their decision to switch from Python to Go eventually.

For some reason though, my instinct tells me that it won't take a long time we will hear or read about Guido's departure from DropBox. This announcement shows me or at least that's the impression I'm getting that they are pushing their developers towards quantity over quality.

Maybe I'm imagining things...well, I hope I'm wrong ¯\_(ツ)_/¯

I am under the impression that they are changing their architecture, since they got out of the cloud to o premisse servers and in this new phase the are changing things and using new tools, when this make sense. When dropbox started python was pretty much the only game in town.

[ldo17]

Seems from the FAQ that Pyston was only for Python 2 anyway.

Anything that still tries to build on Python 2 is a waste of time.

[nevion]

And nothing of value was lost. CPython and something better than Cython to directly integrate C (that avoids the whole thou shall not use C directly limit Cython has in place) is all we really need to address performance issues of compute and avoid all this magic crap that isn't 3.0 compatible and repeatedly failing overly complex lost in limbo. Pretending the massive amount of software (including all the numerical processing stuff) can work on non CPython is an unproductive exorcise in masochism and at best second class citizenship. Jython is cool when you need it but really most stuff should be CPython and a Cython-like replacement. A polyglot language would be pretty cool and maybe you could allow Go for those googlers too. But C/C++-first as the core integration backbone of everything. Boo to all this magic solution holy grail risk taking - just give me something that works, integrates, and has predictable execution wrt source.