The future of Python

Podcast transcript

Alexey Boas:

Hello and welcome to the Thoughtworks Technology podcast. My name is Alexey, I'm the head of technology for Thoughtworks Brazil, and I will be one of your hosts this time together with Mike Mason. Hello Mike.

Mike Mason:

Hello Alexey, my name is Mike and I'm a global head of technology here at Thoughtworks.

Alexey Boas:

And this time we're delighted to have Luciano Ramalho. Luciano is the author of the O'Reilly Fluent Python book. Hello Luciano, would you mind introducing yourself?

Luciano Ramalho:

Hello. Yeah, I'm proud to be a principal consultant at Thoughtworks in Brazil in the São Paulo office, and I joined Thoughtworks about four and a half years ago and I am now working on the second edition of my book as well as helping with other things at Thoughtworks such as preparing our great X-conference that's going to happen in April in São Paulo and other things.

Alexey Boas:

Oh, that's awesome. Well, looks like very good timing for us to talk about Python then since you were working on the second edition of the book. And now, Python has featured in our own Technology Radar quite a couple of times recently and many of them connected to data. So maybe why don't we start there. So Python is quite popular for data. Why is that so Luciano?

Luciano Ramalho:

Well, that's very interesting because I've given it a lot of thought. I think it's probably the main thing that has helped with Python's popularity recently together with another thing, the fact that it was even before data and Python or Python in the domain of data was so popular, many American universities, top universities like MIT and so on, adopted Python as the first language that they teach in their computer science programs and other programs. So I think those two things helped with the increase of popularity of Python recently. But the story of Python with data goes very far.

Luciano Ramalho:

In fact, just before I joined you here, I was looking at the history of the SWIG project. SWIG, S-W-I-G, is a system for generating bindings between C++ and other languages. It's mostly used by scripting languages like Python, TECO, Ruby, et cetera, to be able to connect with libraries written in C++. And SWIG was started by Dave Beasley. Dave Beasley is probably the most famous Python speaker and book author in the world. And Dave started SWIG in '95, so only four years after Python was released, and he was working at the Physics Lab. Initially, he was supporting another language, but then in '96 right in January, he released the first version that supported Python. So it's a long story.

Luciano Ramalho:

Anyway, since that happened, there was a lot of projects, scientific projects initially in the physics domain but then in other domains they started doing this integration between Python and libraries written in C++. And like I said, that happened a long time ago. And then suddenly what happened was when this new approach to artificial intelligence emerged that we call now machine learning, which is heavily based on math, when that came around, it turned out that Python already had everything that was essential for doing machine learning because of the work of the scientists with the language over the previous years.

Luciano Ramalho:

So I think that's one explanation. The fact that the libraries that people needed to do numerical computation, statistics, linear algebra, et cetera, were already in place, plus the fact that Python is an easy language to pick up, but it's not a toy language. It's a very powerful language that takes you very far. And so I think those are the two factors that made it popular in machine learning.

Mike Mason:

And machine learning and all this stuff that you can do with wiring up data pipelines and all that kind of stuff in Python is what has made it maybe more popular recently, but I mean what about beyond data? You were talking about it being ... you didn't quite say an easy language to learn, but you said it was maybe a friendly language.

Luciano Ramalho:

One of the things that I always felt attracted to Python was the fact that people use it for lots of things. There are other scripting languages that are more confined to certain domains, but Python is not like that. If you go to a large pattern conference like PyCon US, EuroPython, PythonBrasil, and so on, you're going to see talks about all kinds of different things. For instance, there's a lot of stuff happening in IoT, right? There is one alternative interpreter for Python called MicroPython that actually runs on microcontrollers. And so for instance, when a few years ago BBC launched a project called the Micro Bits, which is a tiny microcontroller with a LED display and a few sensors, sort of like an Arduino, but friendlier to get started. When that came out in Britain, they distributed that ... every student, I think of 10th grade in Britain got one of those Micro Bits.

Luciano Ramalho:

And the Micro Bits is programmable in Python, natively. And there's other things like Fruits, this company in the US that makes very interesting gear for playing with electronics and programming. They built on top of MicroPython and did a whole environment called a secret Python and now they're releasing one board after another that you can program in Python. So that's one example of a domain. But also in biotech, there's a lot of work being done in biotech for several years and ... of course web programming, right? Web development. It's important to remember that YouTube was built on Python and more recently, Instagram. It started basically as a Python. And those are two projects that we're able to achieve in a web scale, really global scale with essentially a Python code base.

Alexey Boas:

Cool. Well, that speaks for widespread use and several different applications. What about the language itself, Luciano? Of course, you're very passionate about the language. What is it in the language that you love? What are the main features that you feel are unique to Python and that make a difference and make it stand out?

Luciano Ramalho:

Yeah, well, there is several things. And in fact, recently I presented a keynote in China and another one in Brazil that was just about that. They were called the Beauty of Python, and in it I tried to pick up what I like most. And so one of the things is the fact that Python has the concept of an iterator built into the language at a very deep level. So the for loop for instance... remember when in Java, I think it was Java 1.5 there was the enhanced for loop, also known as the for-each loop. So Python was always like that. The for looping pattern was always that way.

Luciano Ramalho:

And it worked with a lot of native data structures that come implemented in the language. But it's also super easy for you to create your own data structure that will play along with the for-each kind of for loop that Python has.

Luciano Ramalho:

And by the way, when I was studying this subject for a talk a few years ago, I discovered that this idea of having the iterator Python built into the language is usually with a key word called yield that you put in a function that makes the function suspend and produce a result and then resume when needed. So that's how the functioning interacts with the for loop.

Luciano Ramalho:

Anyway, this was another invention by the great Barbara Liskov, did you know that? So the famous, Barbara Liskov of the Liskov substitution principle, she created the language along with her PhD students called CLU or CLU, and CLU was the first real language that had this idea of built-in iterator and the yield keyword. So the way Python does that is very similar to the way it was done in CLU.

Luciano Ramalho:

So another thing that I like is, I think Python is easy to learn because the syntax is very simple. There's a very good document in the Python documentation called, I think it's called PEP 3000 and it's about things that we are not changing Python 3000. That's like forever, right?

Mike Mason:

Like when we get to version 3000 this will be the same.

Luciano Ramalho:

Yes, exactly. Exactly. And one of the things there is that the parser is an LL(1) parser. So that's a very simple parser, And it means that the code developers, they want to be bound by a simple parser. Because if the parser is simple, then usually that means that the language is also easy to read. Of course, there's a balance there, right? Because the simplest syntax of a famous language is the Lisp syntax, right? Of the s-expressions. But that's actually a problem that's so simple that makes it hard to read because everything this looks the same and that's becomes a problem.

Luciano Ramalho:

But anyway, Python finds ... I think Guido van Rossum the creator of Python, now retired. He has a very good sense of finding the middle way between alternatives that may be too extreme. The syntax is simple so that makes it easy to learn. Another thing is the fact that the language has this idea of the special methods also known as the dunder methods because they are declared with two double scores in front and back, so they're called dunder methods.

Luciano Ramalho:

Anyway those dunder methods, when you reach a certain level of usage in the language, and that's basically the focus of my book is to bring people to that level. When you understand how the dunder methods work, then you realize why the language is so consistent. Because for instance, to implement something that is iterable, you implement a method called dunder iter. If you want to implement addition, the plus operator, then you implement a method called dunder add. And because those methods are pretty fines, they're pretty established in the language, everybody knows how to grow new things in a way that's consistent with the traditional ways and one-

Alexey Boas:

So like that you can use common language constructs and they are going to behave appropriately for different types of things. But you can read that quite easily because the syntax is familiar but it's behaving as it should for that specific case, right?

Luciano Ramalho:

Exactly, yeah. You know that wonderful feeling that you have when you're learning in some kind of language or framework that is very well designed and then you start making educated guesses and then you're right most of the time, oh, if that works with strings, maybe it works with this as well. And it does. And it works with result sets from a jungle query on a database as well. There's a lot of stuff like that in the language. So that's another thing that I like very much. And just to contrast, I don't want to bash any language but I like Go very much, but one thing that I feel that I miss when I'm programming in Go, is that Go has, for instance, it has a kind of a for-each syntax is called for range in goal, but that only works with five built-in types in the language. You cast yourself, make it work with your own data structures.

Luciano Ramalho:

So that kind of frustrates me. Python you don't feel like your constraints because pretty much everything that's built-in that's written in seeing the language you can emulate right in Python as well.

Alexey Boas:

Okay. So you're saying by is easy to learn. And you mentioned three things. So simple syntax was the first one. Examples like the dunder operator and bringing that consistent feel to the language. Is there a third one?

Luciano Ramalho:

Oh, here's the thing about the dunder methods is important to understand that somebody learning Python doesn't need to know that they exist, but they are already benefiting from them. Because of the fact that the way the language is structured for people who develop libraries and frameworks, it makes it easy for them to provide APIs that are consistent to the novice users. So the novice user doesn't need to know about dunder methods at all or operator overloading, we don't expect people ... we don't want to be doing that on a daily basis, but yeah.

Luciano Ramalho:

And so that's how I think the language is very well designed because it's simple to get started, but then you start uncovering other things about it and then you get other insights and then suddenly you become somebody that can do anything that the pros can do in the language because it doesn't prevent you for doing that.

Mike Mason:

One piece of it that you, you mentioned when we were figuring out what to talk about on the podcast, you said that fails fast.

Luciano Ramalho:

Oh, yeah.

Mike Mason:

Can you talk a bit more about that because I think that's something that might be useful to people?

Luciano Ramalho:

Oh, totally. Yeah. So I've always been a big fan of the so-called scripting languages. I don't like that word too much because it's usually used in a derogatory manner. But anyway, I've always been a fan of very high level dynamic languages. And one problem that many of them have is the fact that sometimes bugs are hidden, for instance, because the typing is weak. So JavaScript has that problem of doing crazy conversion. So you have to learn that you can never trust the equals operator. You have to use the triple equals because otherwise you're going to be beaten, right?

Luciano Ramalho:

Now, in contrast with pretty much every other scripting language that I've learned, Python has a more strict philosophy that we call fail-fast. And so for instance, one example is when you retrieve a value from a dictionary, right? You provide a key and then you want to get a value from the dictionary back or from the hash or mapping or whatever you call it. Anyway, every language, pretty much this days has this constrict, but in Python, if the key is not present, it raises an exception. Other languages give you something that's undefined or some default value like none, that sometimes will bite you later because you were expecting something, the thing wasn't there and you got this value so you didn't get an error and then you have to figure it out later.

Luciano Ramalho:

So another example is called the splat operator or unpacking. When you have several values that you want to assign in parallel or when you went to pass a list of arguments to a function, other languages, sometimes if you pass more arguments they will be silently discarded, if you pass not enough arguments, then if you assign some default value like undefined or none to the remaining arguments, Python doesn't do any of those things. It is strict in many ways where other languages let bugs pass. So I think that's very good. It's really a pleasure because you're working with this highly productive dynamic language, but a lot of bugs are easy to catch because of this characteristic.

Alexey Boas:

Cool. So you talked a little bit about the history of Python with the scientific community and some of the fundamentals that you love and that are cool and make it a joy to use Python. But what about the new and shiny Luciano? What's the cool stuff in Python? What's, what's the new things?

Luciano Ramalho:

Well, for us who work in big systems for our clients, for Thoughtworks clients, one thing that is probably a lot of people are going to welcome is the introduction of a static typing in the language. It's an optional thing, it's actually very similar to the way that TypeScript works and also Dart the new language that Google is using in their flutter framework. So the idea is that you're not forced to declare types and also that type errors are not caught at runtime and the type annotations have no effect at runtime. But the idea is you put those annotations and then you can use in your ... well, they help your ID to detect errors just by looking at the source code. And also you can put ... there are command line type checkers like my PI, there's another one from Microsoft and other one from Google.

Luciano Ramalho:

Some big companies have their own type checkers now, but they are now all consistent because the language adopted a standard syntax and semantics for doing that. And so you can put in your pipeline to have my Pycheck, do a static type check on the source code before I commit even. So that I think is something that's going to be big in the context where we at Thoughtworks work.

Luciano Ramalho:

Another thing is the ecosystem around data and scientific computing in Python and analytics and so on never ceases to amaze me. It's incredible. One thing is that's really super impressive is the Dask project, D-A-S-K. So Dask is a way for you to program in an API that is very similar to the NumPy API that everybody that uses Python for those things know. So they emulate part of the NumPy API, they are raised and so on. But they do it in a way that you can distribute the computation on clusters of machines.

Luciano Ramalho:

And I saw a demo of that in at a talk here in São Paulo and it was super impressive because the person who opened a Jupyter notebook and then was doing Dask computations and then he typed a special commands in Jupyter notebook and then suddenly a big chunk of the screen was taken over by a dashboard showing in real time the computation happening in the nodes. What nodes were idle, what nodes were busy so that you could reconfigure, maybe you had a pipeline with a several steps of map reduce for instance, and then you could graphically see that some steps needed more machines and other steps you could reduce the number of machines.

Luciano Ramalho:

So that's Dask and it's super impressive and it's all open source. One of the things that really I love about Python that I need to mention is that it's really open source language to the core. There was never a huge corporate backer in the history of Python that controlled it. And so it's a community project and the governance is very transparent and of course people in this domain of analytics and scientific computation have always had their voice. And that's why the language also adapted to their needs over the years. But I think it's a great example of a language that shows real open source governance that works.

Mike Mason:

That's pretty cool.

Luciano Ramalho:

Yep.

Mike Mason:

Those are some of the exciting things now. What about the future? You mentioned Python version 3000 earlier. Is there a new version of Python coming? Do we need to look out for that?

Luciano Ramalho:

Well, the switch from Python 2 to Python 3 was very traumatic. I started using Python in '98 when it was Python 1.5. So I went through the change from 1.5 to 2.0 and that was very smooth. Pretty much most programs just continued working with no problems. But in Python 3 they decided to fix some things that were very fundamental. The most fundamental things that they wanted to fix that was impossible to do without breaking backward compatibility was reorganizing the support of Unicode in the language because the strings in Python, the default string literals were not Unicodes and there was an alternative Unicode type in Python 2.7 that was okay, but it was not the default so people missed.

Luciano Ramalho:

Anyway, with that change, with what required breaking a lot of stuff, they decided to do other breaking changes and it took about 10 years for ... yeah, Python 2.7 lasted 10 years. It was frozen last January. Since January the community doesn't support Python 2.7 anymore. I think lots of vendors are still going to support it like Red Hat and other vendors because there isn't a huge install base. But anyway-

Mike Mason:

And what was-

Luciano Ramalho:

... because of ... go ahead.

Mike Mason:

Why were people stuck so long? Well, we've got all these libraries and frameworks and code bases and we-

Luciano Ramalho:

That's interesting. The first thing that took a long time was that people needed ... if I am an app developer, I need all of my dependencies to be compatible with Python 3, right? So I had to wait for that. So that took about five years, I would say for the main projects, even the whole universe of the NumPy libraries, the SciPy libraries and Django and so on, took them a few years to be able to release versions that supported both Python 2.7 and Python 3. But now, it's just the problem of priorities when you have a system in production and you have a ... yeah, maybe-

Mike Mason:

Do you think like ... sorry, do you think commercial language vendors have done any better on that? It would seem ... I actually don't know the answer to this, I'm just curious. Whether, for example, Microsoft has done a bad job on getting people to stay up to date on versions of C# or whether Java has done any better to encourage people to move to new versions. I guess it depends on whether there was a major breaking change like the one that you talked about.

Luciano Ramalho:

Well, I don't know. I don't follow those languages so closely anymore so, but let me just say that this whole thing, I think it's passed us, it's amazing that the language survived and even grew a lot during those years of troubles and recently I was talking about that on Twitter with some other people and actually quoted, Windu in one of my tweets and Windu came and said probably there won't be a Python four at all because there's no need for that. We are going to do incremental releases and then we can use just ... now 3.9 is in alpha and the next is going to be 3.10 and 3.11 and so on. So I don't think we're going to have ... the community sufficiently traumatized to not make that mistake again. Yeah.

Alexey Boas:

Okay, great. And how about the book Luciana? So you're writing a second edition of your quite popular Fluent Python book. We can mention that, right? That's not a secret is it?

Luciano Ramalho:

No, no, it's not a secret.

Alexey Boas:

Okay, fine. So why a second edition then? What's in this new edition? What are you working on?

Luciano Ramalho:

Well, the first edition came out in 2015 so although I focused on Python 3 at the time and it was Python 3.4 at the time, there were some things that most of ... almost every line of code in the book, in the first edition still works except for one part that was there asyncio library because it was provisional at the time. So I ran the risk of covering it because it was a hot new things, but because it was provisional and everybody knew that the API could change and it did change. And also the syntax evolved to become nicer. They introduced two keywords, async and await that are very similar to the similar keywords in C#. So they introduced those impact and then made all of that kind of asynchronous programming more readable. And so that was a major reason to do the updates. And of course, the whole new topic of type hints that did not exist at the time. So I think those are the most important changes in the second edition, covering type hints and talking about the new way of doing asyncio.

Mike Mason:

And just for people who may know have read the first edition, it's called Fluent Python, tell me more about your approach in the book in general.

Luciano Ramalho:

Yeah, so the book is aimed at people who already know Python. All right? Yeah, it's actually evolved from a series of courses that I used to present that were called Python for people who know Python. And so, the idea is, you know Python, you use it every day, but do you really take the most advantage of it? Because being an easy language to learn means that sometimes you don't study it so deeply because you can get away doing things in a simple way, but then often you're using it with a very strong accent from another language and missing things from the other language. But you don't know that there are other things that do the same things, but in different ways in Python. So that's the focus of the work and that's why it's called a Fluent Python. But it's-

Mike Mason:

I-

Luciano Ramalho:

Yeah, go ahead.

Mike Mason:

I can relate. I'm sure all my Ruby code looked like Java. [crosstalk 00:31:23] anyone who actually knew Ruby,

Luciano Ramalho:

Exactly, it a book is about idiomatic Python. And I think it was the first book in the market. I don't know if ... it was if the first book in the market that covered special methods in the first chapter because I assumed that you know Python, so here's something that if you don't know, you need to know so that you get a deeper understanding of all the language and why it's so consistent. So it starts with special methods, but then it goes all the way to meta programming meta classes. It gets really weird by the end. I really enjoy writing. It's one of my favorite things professionally.

Mike Mason:

Yeah, we are glad that you do because then we get those wonderful books, so.

Luciano Ramalho:

And I want to thank Thoughtwork sincerely from the bottom of my heart for supporting this because I'm working part time at Thoughtworks because they let me work on the book as well, so that's really great.

Alexey Boas:

Okay, Luciano, thank you very much. I guess we're coming to the end of the episode. It was a great conversation and great to have you with us. Thank you so much.

Luciano Ramalho:

Well thank you so much Mike and Alexey Boas. It was a pleasure talking to you.

Mike Mason:

Thank you Luciana.

Luciano Ramalho:

Bye-bye.

Alexey Boas:

And thank you all for joining and if you have any feedback for us, don't hesitate to reach out or leave a rating or comments on your preferred platform. Thank you so much for listening. Bye.

Neal Ford:

On the next episode of the Thoughtworks technology podcast. We're joined by two of our colleagues Prassana and Bharani to talk about the distinction between observability and monitoring and some good ideas of what to do and what not to do. So please join us on our next episode.