I went to PyCon 2019 last week, from Wednesday, May 1st through Thursday, May 9th. It was very fun, and I got to meet a lot of interesting people and learn a lot of new things. People were super friendly and open and willing to share ideas. IMHO, and from talking to other people at the conference, I think Python as a community does a great job at welcoming people and being observable about things like financially supporting women in tech and underrepresented minorities. For example, Guido van Rossum, the creator of Python, personally mentored Mariatta Wijaya to become a CPython core developer to increase female representation in the core developers community, and the best Django tutorial besides the official Django documentation comes from Django Girls. PyLadies is also an extremely vibrant group promoting women using Python; it raised ~\$45,000 during the PyLadies charity auction this year. Promoting community health, probably more than anything else the Python community does, is why Python is the dominant scripting language today.
Here’s what I learned about how to best make use of conference time:
You absolutely don’t have to go to everything listed on the schedule: I came on the first day believing that I could just go to a tutorial. Turns out they’re \$150 a pop, and require pre-registration, and they were all sold out. I was terrified that I would have wasted an entire day just milling about on the convention floor and doing nothing.
I needn’t have worried. I randomly walked up to Curtis Maloney, and we just started talking about random things; he was demonstrating a toy compression algorithm he was building, and the principles behind arithmetic coding. I also got to meet Amber Brown, as well as Russell Keith-Magee. The Python language summit was also this first day, and many of the core CPython developers were there to discuss the future of the language. Guido van Rossum himself stopped by and we got a picture:
I also got to meet Toshio Kuratomi, as well as Barry Warsaw; together, we got dinner with the Instagram/Facebook team at PyCon, including Carl Meyer and Jennifer Taylor, where I got to be a fly on the wall for a very interesting discussion about Python at scale. This was all completely random! And everybody’s super nice!
I should also say that the conference talks are published on YouTube (materials and documentation are released on a speaker-by-speaker basis); for example, this year’s talks are published here. This definitely impacted what I chose to do during the conference by making it less painful to not attend a particular talk and look for other things at the conference.
Definitely check out the various Open Spaces hosted during the body of the conference: This ties into my previous point of not planning out everything prior to the conference. Open Spaces are decided by people who just show up at the conference, and schedules are filled out with pen on index card. Many times, these are run by speakers who want to do discussions on their talks in small groups, or talks by speakers who didn’t get a slot this time around for one reason or another, or just FOSS project leaders giving status reports. I went to three and a half open spaces; one on CPython profiling and performance by Itamar Turner-Trauring, one on the Python
cryptographyby Alex Gaynor and Paul Kehrer, one on the Recurse Center by Allison Kaptur and Elizabeth Sander et al., and the tail end of one on Python at Netflix by Matthew Seal.
Volunteer! I was session chair for three sessions; my job was to hold up signs indicating time for the speaker and to moderate Q&A sessions if necessary. It was nice to give back to the conference. It was also a good way to meet the speakers and have something to talk about in a follow-up email. PyCon offers many different ways in order to volunteer, so there should be something for everybody.
Go to talks by speaker, not topic: Good speakers can make a meh topic extremely interesting. Mediocre speakers can make a good topic very meh. I missed a talk on API evolution by A. Jesse Jiryu Davis to listen to another talk at the same time. It did not go the way I thought it would; it was the only time during the conference I saw people streaming out the door in the middle of the talk.
On the other hand, some talks are filled, with people banging on the doors to get in. Raymond Hettinger gave a talk on solvers in Python. I was walking with a guy to the room. He went to the bathroom. I walked into the room. It was packed, people were beginning to line up against the walls. The session chairs and runners were telling people to leave; due to liability reasons PyCon couldn’t have more people in the room than the fire code specified. Some people waved to indicate an open seat. I went there and sat down. Damn, I thought. That dude’s still in the bathroom.
While PyCon doesn’t have hard and fast registration for conference talks, there are a few rules of thumb you can follow to determine which speakers are the most popular. At a convention center, there are a number of differently-sized conference rooms, at differing distances from the main entrance. The talks in the beginning of the conference (e.g. morning on opening day), in the biggest rooms, with recognizable names, are generally going to be good talks. Talks scheduled during those times in small rooms far away from everything else are less likely to be as good.
Use Twitter: I always thought Twitter was a complete garbage fire; when I purged my social media accounts, Twitter was the first to go. Parts of it probably are garbage fires, but a lot of tech people still use it to keep in touch with their network. By contrast, LinkedIn didn’t have a whole lot of usage, probably because many people there were advanced enough in their careers to just reach out to their network for a job, and because LinkedIn has so many third-party recruiters sending unsolicited requests. On Twitter, it’s also much easier to put your personal brand first and foremost, and much easier to just “get connected” there and then since you can follow anybody without their consent.
My modus operandi was to collect business cards and then follow up with email and LinkedIn requests afterwards, but it took a lot of time and energy to do so, and I’m not sure how valuable it is in comparison to a Twitter feed. On the other hand, it’s much less likely that LinkedIn will fill your inbox with too many ads, and if you have somebody’s email, they will generally respond to a request as opposed to something thrown on Twitter, you have initial context as to how you got in touch with somebody, and people may appreciate you’re deliberate about constructing a relationship. Also, if you don’t use social media regularly, other people may find it less valuable to follow you since there’s no interesting content posted on a regular basis.
I opened a Twitter account. I locked myself out after the first five minutes accidentally. I have gotten quite old.
Honestly the most valuable social media network for developers these days may be GitHub.
Try out a bunch of different open source projects: I knew nothing about open-source contributions when I first came to PyCon, and the experience I had during the development sprints there was fairly encouraging. My goal was to ship something during the four days I had.
(As a strong disclaimer, this is my experience after working professionally with Python for 2.5 years, and should not be interpreted as a benchmark to measure people’s contributions by. If you do get this much done or more, props to you, and if you don’t, it’s ok, nobody bites! I’m also kind of just writing this as a log of what I physically did).
The first project I went to was the BeeWare project, which is led by a number of former Django core developers like Russell. I went to BeeWare instead of another project explicitly because the BeeWare community is extremely friendly to beginners (everybody who merges a pull request gets a challenge coin, and everybody in the room claps for them), and because the BeeWare project is still experimental meaning newcomers have more opportunity to make meaningful feature merges. I merged in a documentation fix the first day, and then I merged in the ability to select and open multiple folders on
toga-cocoa. This led to an interesting interface design; Nehar Arora mentioned how file and folder selection used the same endpoints in macOS, while in Gtk different endpoints were used, necessitating a change in in top-level interface for opening multiple folders after discussion with the BeeWare core developers.
The next project I took a look at was
securedrop, an electronic dead drop for whistleblowers. I merged in a new ability to leverage property-based testing with
hypothesis, which I thought could help the team find any bugs more cheaply. After talking to Zac Hatfield-Dodds, I added in a test for ensuring encrypting and decrypting strings was working properly. Since this leveraged primarily well-tested cryptography code such as
python-gnupg, I didn’t find any bugs. The tests also took a lot longer to run on the CI pipeline than on my local machine (1 minute vs. 400 milliseconds), which Zac said could be due to the lack of support for cryptographic hardware on the CI virtual machines. Since hypothesis runs about a hundred tests for every stub created, this incurred a non-trivial penalty. Finally, since securedrop (wisely, IMHO) mostly glues together well-tested third-party libraries, there wasn’t a whole lot to test with regards to unit tests. The vast majority of tests were in the integration test suite. Since
hypothesisis best at running unit tests, I had to lift the deadline for execution in order to avoid the test stub from timing out. Otherwise, it was a pleasant experience. The securedrop team is very nice, and since everybody was there in the room, changes were merged in quickly.
The last project I took a look at was CPython itself. That was a far more bureaucratic process than the prior two projects. I went ahead and addressed a documentation issue opened in
bugs.python.org, but a core developer came in and said the underlying issue was unnecessary, and closed my pull request. That was mildly upsetting, since I would have liked to merge in a change to CPython and become a CPython contributor, but his reasoning was sound. Then I took a look at BPO-36850, which fundamentally was an issue with copying over extended attributes and permissions from files on network filesystems, and which errors could be ignored without data loss. There is a pull request open addressing the underlying issue here, but it hasn’t been merged in yet for over six months. No response from PR owner, who also happened to be the module owner. Then the clock ran out. I asked other developers how CPython would test something like connecting and copying files from a network filesystem, and there weren’t many great answers. CPython uses a patchwork of free CI pipelines from multiple providers, and I don’t think there would be any money in setting up an NFS or CIFS client just for testing. We could have added unit tests mocking the new ignored error, but it would give low visibility into how the connection would actually play out, and so we didn’t add the test.
It’s pretty insane just how few CPython developers are working on CPython full-time: 1 guy (who works at Red Hat). Others are part-time contributors whose companies are generous enough to donate 1 day per week for an employee to add code, and others just contribute in their free time. BPO, the issue tracker, is also pretty archaic. I think if they switched from Mercurial to Git, and used a self-hosted version of GitLab, they may get a lot of good features to reduce issue management overhead.
The good things about open-source is that it’s very easy to meet extremely smart people, and easy to earn their respect, which is huge for building up social capital. The bad thing about open-source is that it’s hard to maintain the discipline for establishing and executing best practices, since the projects are far more resource-constrained than I could have imagined.
This was like a tasting menu for what different projects are like, and what different contribution policies and pipelines are like. You can always do serious contributions after introductions and orientations are done together!
Things are expensive: I stayed with two friends in the area, and saved quite a bit on hotel costs, but since I didn’t know how safe Cleveland was at night, and since I was time-constrained in the morning to get to the convention center on time, I took a lot of Lyft. And there aren’t enough Lyfts and traffic in the area in order to use Lyft Line, the carpooling service. Yea, killed my budget planning for the month. I talked to a Lyft driver and he said he only got half of my fee; Lyft took the other half.
I think next time, I will try and take Lyft for the first few days, in order to find a driver I like. If I like a driver, I might just offer them my fee directly, in exchange for driving to/from the same locations. That way, the driver can get more money, and I may be able to negotiate to pay less total. The problem is of course coordination and availability, but it shouldn’t be a problem since Lyft is always there.
Food and beverage is apparently insanely expensive. I was talking to Kushal Das and he said every day of the development sprints, PyCon spends \$2000 on coffee in the first hour, and \$6000 for the day. PyCon spends \$35,000 on just coffee during the duration of the conference. This is because the coffee has to come from the convention center, which has one vendor (StarBucks), meaning no competition; if outside coffee was allowed this cost could be cut by half. This is after the Python Software Foundation negotiates with cities and convention venues to lower the cost of hosting the event, resulting in the convention center having to nickel-and-dime the Python Software Foundation (PSF) to try and recoup costs. Without money, it’s a lot harder to make everybody happy. PSF has no money because Python is not a pay-to-play project. Apparently, other open source foundations, like the Linux Foundation, get corporate money to include their pet features into the project, affecting the technical direction of the project, and for Python, that could kill the contributor base.
I signed up to be a PSF financial contributor at PyCon for \$20/mo., and I think it’d be great if returning PyCon members had the option to pay the difference for their PyCon tickets between early-bird and regular pricing in order to supply grants to people of lesser means and underrepresented minorities to attend the conference.
The corporate tutorials are pretty good: Tying into my previous point, because of a strong separation between corporations and PSF, there’s a stronger focus during the corporate tutorials (which are free to attend) on useful things to learn about Python in production, rather than pushing a marketing product such as a corporate API. Capital One hosted an absolutely smashing tutorial on Python performance, represented by Steven Lott and Gil Forsyth, which I think had a lot of value and didn’t have really anything in terms of Capital One branding/marketing.
(Capital One also paid for coffee during the development sprints after PyCon ran out in the first hour, so 👏 👏 👏 to them there)
Don’t fly United Basic Economy: Don’t. Just don’t.