Victor Stinner blog 3

I am currently at a CPython sprint 2017 at Facebook.

This is the third part of my contributions to CPython during 2017 Q2 (april, may, june): Security Trick bug: Clang 4.0, dtoa and strict aliasing sigwaitinfo() race condition in test_eintr FreeBSD test_subprocess core dump Previous reports: My contributions

This is the second part of my contributions to CPython during 2017 Q2 (april, may, june): Mentoring Reference and memory leaks Contributions Enhancements Bugfixes Stars of the CPython GitHub project Previous report: My contributions to CPython during 2017…

This is the first part of my contributions to CPython during 2017 Q2 (april, may, june): Statistics Buidbots and test.bisect Python 3.6.0 regression struct.Struct.format type Optimization: one less syscall per open() call make regen-all Previous report: My

I spent the last 6 months on working on buildbots: reduce the failure rate, send email notitication on failure, fix random bugs, detect more bugs using warnings, backport fixes to older branches, etc.

This article tells the story of the new CPython test.bisect tool to identify failing tests in the CPython test suite.

My contributions to CPython during 2017 Q1 (january, februrary, march): Statistics Optimization Tricky bug FASTCALL optimizations Stack consumption Contributions os.urandom() and getrandom() Migration to GitHub Enhancements Security regrtest Bugfixes Prev…

In feburary 2017, CPython from Bitbucket with Mercurial moved to GitHub with Git: read [Python-Dev] CPython is now on GitHub by Brett Cannon.

Here is the raw list of the 46 CPython issues I opended between 2016-04-21 and 2017-02-10 to implement my FASTCALL optimization.

For my FASTCALL project (CPython optimization avoiding temporary tuples and dictionaries to pass arguments), I wrote many short microbenchmarks.

False start In April 2016, I experimented a Python change to avoid temporary tuple to call functions.

My contributions to CPython during 2016 Q4 (october, november, december): hg log -r 'date("2016-10-01"):date("2016-12-31")' --no-merges -u Stinner Statistics: 105 non-merge commits + 31 merge commits (total: 136 commits).

My contributions to CPython during 2016 Q3 (july, august, september): hg log -r 'date("2016-07-01"):date("2016-09-30")' --no-merges -u Stinner Statistics: 161 non-merge commits + 29 merge commits (total: 190 commits).

I was invited at my first CPython sprint in September!

getrandom() avoids file descriptors Last years, I'm making sometimes enhancements in the Python code used to generate random numbers, the C implementation of os.urandom().

My contributions to CPython during 2016 Q2 (april, may, june): hg log -r 'date("2016-04-01"):date("2016-06-30")' --no-merges -u Stinner Statistics: 52 non-merge commits + 22 merge commits (total: 74 commits).

My contributions to CPython during 2016 Q1 (january, februrary, march): hg log -r 'date("2016-01-01"):date("2016-03-31")' --no-merges -u Stinner Statistics: 196 non-merge commits + 33 merge commits (total: 229 commits).

I am working on the CPython benchmark suite (performance) and I run the benchmark suite to upload results to speed.python.org.

My first article Intel CPUs is a general introduction on modern CPU technologies having an impact on benchmarks.

Ten years ago, most computers were desktop computers designed for best performances and their CPU frequency was fixed.

I developed a new perf module designed to run stable benchmarks, give fine control on benchmark parameters and compute statistics on results.

Stable benchmarks are so close, but ...

With the system tuning (part 1), I expected to get very stable benchmarks and so I started to benchmark seriously my FASTCALL branch of CPython (a new calling convention avoiding temporary tuples).

Background In the CPython development, it became common to require the result of the CPython benchmark suite ("The Grand Unified Python Benchmark Suite") to evaluate the effect of an optimization patch.

Now that most OpenStack services have reached feature freeze for the Mitaka cycle (November 2015-April 2016), it's time to look back on the progress made for Python 3 support.

This article described the _PyBytesWriter and _PyUnicodeWriter private APIs of CPython.

My contributions to CPython during 2015 Q4 (october, november, december): hg log -r 'date("2015-10-01"):date("2015-12-31")' --no-merges -u Stinner Statistics: 100 non-merge commits + 25 merge commits (total: 125 commits).

A few years ago, someone asked me: "Why do you contribute to CPython?

I added functions to the private "pytime" library to convert timestamps from/to various formats.

Previous status: Status of the FAT Python project, November 26, 2015.

Previous status: [python-dev] Second milestone of FAT Python (Nov 4, 2015).

From 2to3 to 2to6 When Python 3.0 was released, the official statement was to port your application using 2to3 and drop Python 2 support.