Saturday, February 9, 2019

Düsseldorf Sprint Report 2019

Hello everyone!

We are happy to report a successful and well attended sprint that is wrapping up in Düsseldorf, Germany. In the last week we had eighteen people sprinting at the Heinrich-Heine-Universität Düsseldorf on various topics.

Totally serious work going on here constantly.

A big chunk of the sprint was dedicated to various discussions, since we did not manage to gather the core developers in one room in quite a while. Discussion topics included:

  • Funding and general sustainability of open source.
  • Catching up with CPython 3.7/3.8 – we are planning to release 3.6 some time in the next few months and we will continue working on 3.7/3.8.
  • What to do with VMprof
  • How can we support Cython inside PyPy in a way that will be understood by the JIT, hence fast.
  • The future of supporting the numeric stack on pypy – we have made significant progress in the past few years and most of the numeric stack works out of the box, but deployment and performance remain problems. Improving on those problems remains a very important focus for PyPy as a project.
  • Using the presence of a CPython developer (Łukasz Langa) and a Graal Python developer (Tim Felgentreff) we discussed ways to collaborate in order to improve Python ecosystem across implementations.
  • Pierre-Yves David and Georges Racinet from octobus gave us an exciting demo on Heptapod, which adds mercurial support to gitlab.
  • Maciej and Armin gave demos of their current (non-PyPy-related) project VRSketch.

Visiting the Landschaftspark Duisburg Nord on the break day

Some highlights of the coding tasks worked on:

  • Aarch64 (ARM64) JIT backend work has been started, we are able to run the first test! Tobias Oberstein from Crossbar GmbH and Rodolph Perfetta from ARM joined the sprint to help kickstart the project.
  • The long running math-improvements branch that was started by Stian Andreassen got merged after bugfixes done by Alexander Schremmer. It should improve operations on large integers.
  • The arcane art of necromancy was used to revive long dormant regalloc branch started and nearly finished by Carl Friedrich Bolz-Tereick. The branch got merged and gives some modest speedups across the board.
  • Andrew Lawrence worked on MSI installer for PyPy on windows.
  • Łukasz worked on improving failing tests on the PyPy 3.6 branch. He knows very obscure details of CPython (e.g. how pickling works), hence we managed to progress very quickly.
  • Matti Picus set up a new benchmarking server for PyPy 3 branches.
  • The Utf8 branch, which changes the internal representation of unicode might be finally merged at some point very soon. We discussed and improved upon the last few blockers. It gives significant speedups in a lot of cases handling strings.
  • Zlib was missing couple methods, which were added by Ronan Lamy and Julian Berman.
  • Manuel Jacob fixed RevDB failures.
  • Antonio Cuni and Matti Picus worked on 7.0 release which should happen in a few days.

Now we are all quite exhausted, and are looking forward to catching up on sleep.

Best regards, Maciej Fijałkowski, Carl Friedrich Bolz-Tereick and the whole PyPy team.

4 comments:

  1. Congratulations for the sprint, folks! Any plans to leverage the manylinux2010 infrastructure and about producing PyPy compatible wheels soon?

    ReplyDelete
  2. Nice work, looking forward to Python 3.6 and beyond! Is there anywhere to view the Python 3 benchmarks like there is for PyPy2?

    ReplyDelete
  3. Hi Juan! Yes, we are going to work on manylinux2010 support to have PyPy wheels soon.

    ReplyDelete
  4. @Anonymous yes, being able to view PyPy3 benchmarking results is the goal of the new benchmarking server, will still take a bit of work to hook everything up.

    ReplyDelete

See also PyPy's IRC channel: #pypy at freenode.net, or the pypy-dev mailing list.
If the blog post is old, it is pointless to ask questions here about it---you're unlikely to get an answer.