I will confess that I get a sense of psychological comfort from strict typing, even though everyone agrees Python is faster for a quick hack. I usually go with Haskell for quick stuff.
I firmly believe that every language has an equal proportion of spaghetti code to clean code. The only factor that might screw with this is how much a language is used in industry, which I’d expect raises the ratio. However, there’s plenty of hobbyists writing spaghetti code too so I don’t think even that factor has much effect.
Okay, I’ll grant you brainfuck… As for assembly, I don’t think it’s inherently spaghetti. You can split it up into functions just like you can with an actual programming language. It’s not impossible to make structured code.
That said, I never coded assembly outside of a mandatory university course, so I don’t feel super confident in saying that. But I don’t think of it as a programming language anyway - it’s a 1:1 translation to/from machine code, and machine code isn’t meant to make programming easy or scalable.
And TBF neither is brainfuck. It was a bit of a cheeky example, but I wanted to really emphasise the range of differences between languages, and language-like things.
I have trouble believing that every language is exactly as easy to organise code in. I’ll give you that it’s possible in every language (and assembly) to organise code, but that’s far too low a bar for practical measurement. Technically you can dig a ditch with a rusty spoon, too…
If Roller Coaster Tycoon had well organised code, that was down to way more effort being expended to make it that way.
I find Python is quick for the first 30 minutes. If you need any kind of libraries, or assistance from an IDE, or a distribution build, or you’re more familiar with another language, then it isn’t quicker.
It does, but the lack of static typing means it is more difficult to interact with foreign code (correctly).
When I pull in a library in a JVM language or Rust etc., I quickly glance at the documentation to get a rough idea of the entrypoint for the library.
Like, let’s say I want to create a .tar file, then the short “Writing an archive” example tells me all I need to know to get started: https://crates.io/crates/tar
If I need to find out more, like how to add a directory, then having the tar Builder initialized is enough for me to just ask my code completion. It will tell me the other available functions + their documentation + what parameters they accept.
If I make a mistake, the compiler will immediately tell me.
In Python, my experience was completely different. Pulling in a library often meant genuinely reading through its documentation to figure out how to call it, because the auto-completion was always unreliable at best.
Some libraries’ functions wouldn’t even tell you what types you’re allowed to feed into them, nor what type they return, and not even even diving into their code would help, because they just never had to actually specify it.
PyCharm a super powerful IDE, VSCode has tons of Python extensions that L rival PyCharm’s functionality, lots of other IDEs have decent python support
Yes, PyCharm is a super powerful IDE when compared to Nodepad++. But it’s a trashcan fire compared to IntelliJ or even the much younger RustRover.
Half the time it can’t assist you, because no one knows what types your code even has at that point.
The other half of the time, it can’t assist you, because, for whatever reason, the Python interpreter configured in it can’t resolve the imports.
And the third half of the time, it can’t assist you, because of what I already mentioned above, that the libraries you use just don’t specify types.
These are problems I’ve encountered when working on a larger project with multiple sub-components. It cost us so much time and eventually seemed to just be impossible to fix, so I ended up coding in a plain text editor, because at least that wouldn’t constantly color everything red despite there being no errors.
These problems are lessened for smaller projects, but in that case, you also don’t need assistance from an IDE.
or a distribution build
Not sure exactly what you mean by this
What I mean by that is that Python tooling is terrible. There’s five different ways to do everything, which you have to decide between, and in the end, they all have weird limitations (which is probably why four others exist).
or you’re more familiar with another language
Yeah this can be said about any language. “You’re quickest in the language you’re most familiar with”. That’s basically a tautology.
Yes. That is all I wanted to say by that. People just often claim that Python is a great prototyping language, to the point where the guy I was responding to, felt they’re doing the wrong thing by using the familiar tool instead. I’m telling them, they’re not.
What I mean by that is that Python tooling is terrible. There’s five different ways to do everything, which you have to decide between, and in the end, they all have weird limitations (which is probably why four others exist).
There’s actually at least 15 different ways (the fifteenth one is called rye and it’s where I got that article from). And yes your entire post is super accurate. The pycharm thing is ridiculous too because RubyMine is excellent in comparison. You just pull in a library with Ruby’s excellent (singular) package manager, and then RubyMine is able to autocomplete it pretty much perfectly. PyCharm can’t even manage to figure out that you added a new dependency to whatever flavor of the week package manager you’re using this time.
Look, it’s fine if you prefer other languages to python, I won’t besmirch anyone’s preferences. But literally everything in your post exists in nearly every programming language (minus some of the typing stuff, I’ll give you that, but it’s getting a lot better). Like, every language has some learning curve to setting up tooling, or configuring your IDE the way you like it, or learning how to navigate documentation so that it’s useful, or trying to decide on one of the multiple ways of doing things. I guarantee, as someone with limited experience with Java, I’d have a difficult time setting up and using IntelliJ, and figuring out which build/packaging system I need to use, and figuring out how to use whatever libraries I need, simply because I’m unfamiliar with the ecosystem. That’s all you’re describing - the initial learning curve in getting familiar with a new language. Which is why I pointed out all the things I pointed out. It’s where I start when I’m introducing developers to python.
No, it really is unique to python. Most other languages have one or two package managers, not 15 (15 is not an exaggeration). Ruby has one. Rust has one. Java has two (maven and gradle). Elixir has one. Swift has one.
Python programmers think it’s normal when it most definitely is not. Even your IntelliJ example isn’t correct because IntelliJ will literally install and set up the jdk for you, but pycharm is completely unable to do that and it’s not because JetBrains hasn’t tried. Python tooling is just really really really bad.
A lot of people feel that way. If I need to generate a set of numbers or a certain string, though, it’s pretty easy to punch out a one-liner in GHCi, and that’s usually my use case.
I don’t know if you’re trying to be funny or your autocorrect is, but in Germany, when they switched from diploma to the bachelor/master system, both bachelor and master come with a theses. Many people leave with a bachelor, especially in computer science, so that might be why. Don’t know about other countries
Neither of those provide type inference? Type inference is when you give the compiler only occasional type hints and it can still figure out what the types are behind the scenes.
For example:
name = "World"greeting = "Hello " + name
compile_error = greeting / 42
In a type-inferred language, the compiler would automatically know that:
the first line is of type String, because it’s trivially initiated as one.
the second line is of type String, because String + String results in a String.
the third line is non-sense, because greeting is a String and cannot be divided by a number. It could tell this before you run the program.
Mypy on the other hand can only tell these things, if you give the first two lines an explicit type hint:
name:String = "World"greeting:String = "Hello " + name
Having to do this on every line of code is extremely noisy and makes refactoring annoying. I can absolutely understand that Python folks think you get productivity gains from duck typing, if this is the version of static typing they’re presented.
And we did excessively use mypy + type hints + pydantic on my most recent Python project. These are not the silver bullet you think they are…
If you’re going to post a code example, at least check that it works. Here’s your example, with no type hints, giving me errors both from the LSP, and when trying to run via mypy: https://imgur.com/a/Hq5Y5Gt.
I find it’s possible to operate Python as a statically typed language if you wanted, though it takes some setup with external tooling. It wasn’t hard, but had to set up pyright, editor integration, configuration to type check strictly and along with tests, and CI.
I even find the type system to be far more powerful than how I remembered Java’s to be (though I’m not familiar with the newest Java versions).
Did everyone become stupid in the last 10 or so years? We used to write huge apps in Python without any type checkers or static analysis tools and never had any problems we wouldn’t have had in statically typed languages.
But one compiling error is Java is 7 run time errors in python.
There is a type error and you couldn’t have known it beforehand? Thanks for nothing
I will confess that I get a sense of psychological comfort from strict typing, even though everyone agrees Python is faster for a quick hack. I usually go with Haskell for quick stuff.
And then the quick hack gets a permanent solution and the next employee has to fight trough the spagetti.
sounds like not my problem
This may be true, but it’s equally true in any programming language, so not really relevant.
I’d guess it’s less true for something statically typed, just because that reduces the ways it can be unintuitive.
I firmly believe that every language has an equal proportion of spaghetti code to clean code. The only factor that might screw with this is how much a language is used in industry, which I’d expect raises the ratio. However, there’s plenty of hobbyists writing spaghetti code too so I don’t think even that factor has much effect.
Really? Doesn’t that imply non-spaghetti brainfuck or assembly?
Okay, I’ll grant you brainfuck… As for assembly, I don’t think it’s inherently spaghetti. You can split it up into functions just like you can with an actual programming language. It’s not impossible to make structured code.
That said, I never coded assembly outside of a mandatory university course, so I don’t feel super confident in saying that. But I don’t think of it as a programming language anyway - it’s a 1:1 translation to/from machine code, and machine code isn’t meant to make programming easy or scalable.
And TBF neither is brainfuck. It was a bit of a cheeky example, but I wanted to really emphasise the range of differences between languages, and language-like things.
I have trouble believing that every language is exactly as easy to organise code in. I’ll give you that it’s possible in every language (and assembly) to organise code, but that’s far too low a bar for practical measurement. Technically you can dig a ditch with a rusty spoon, too…
If Roller Coaster Tycoon had well organised code, that was down to way more effort being expended to make it that way.
Such is life.
I find Python is quick for the first 30 minutes. If you need any kind of libraries, or assistance from an IDE, or a distribution build, or you’re more familiar with another language, then it isn’t quicker.
PyPI has a huge selection of libraries
PyCharm a super powerful IDE, VSCode has tons of Python extensions that L rival PyCharm’s functionality, lots of other IDEs have decent python support
Not sure exactly what you mean by this
Yeah this can be said about any language. “You’re quickest in the language you’re most familiar with”. That’s basically a tautology.
Oh boy, you really wanna talk about it?
It does, but the lack of static typing means it is more difficult to interact with foreign code (correctly).
When I pull in a library in a JVM language or Rust etc., I quickly glance at the documentation to get a rough idea of the entrypoint for the library.
Like, let’s say I want to create a .tar file, then the short “Writing an archive” example tells me all I need to know to get started: https://crates.io/crates/tar
If I need to find out more, like how to add a directory, then having the tar
Builderinitialized is enough for me to just ask my code completion. It will tell me the other available functions + their documentation + what parameters they accept.If I make a mistake, the compiler will immediately tell me.
In Python, my experience was completely different. Pulling in a library often meant genuinely reading through its documentation to figure out how to call it, because the auto-completion was always unreliable at best.
Some libraries’ functions wouldn’t even tell you what types you’re allowed to feed into them, nor what type they return, and not even even diving into their code would help, because they just never had to actually specify it.
Yes, PyCharm is a super powerful IDE when compared to Nodepad++. But it’s a trashcan fire compared to IntelliJ or even the much younger RustRover.
Half the time it can’t assist you, because no one knows what types your code even has at that point.
The other half of the time, it can’t assist you, because, for whatever reason, the Python interpreter configured in it can’t resolve the imports.
And the third half of the time, it can’t assist you, because of what I already mentioned above, that the libraries you use just don’t specify types.
These are problems I’ve encountered when working on a larger project with multiple sub-components. It cost us so much time and eventually seemed to just be impossible to fix, so I ended up coding in a plain text editor, because at least that wouldn’t constantly color everything red despite there being no errors.
These problems are lessened for smaller projects, but in that case, you also don’t need assistance from an IDE.
What I mean by that is that Python tooling is terrible. There’s five different ways to do everything, which you have to decide between, and in the end, they all have weird limitations (which is probably why four others exist).
Yes. That is all I wanted to say by that. People just often claim that Python is a great prototyping language, to the point where the guy I was responding to, felt they’re doing the wrong thing by using the familiar tool instead. I’m telling them, they’re not.
There’s actually at least 15 different ways (the fifteenth one is called rye and it’s where I got that article from). And yes your entire post is super accurate. The pycharm thing is ridiculous too because RubyMine is excellent in comparison. You just pull in a library with Ruby’s excellent (singular) package manager, and then RubyMine is able to autocomplete it pretty much perfectly. PyCharm can’t even manage to figure out that you added a new dependency to whatever flavor of the week package manager you’re using this time.
Oh man, this isn’t meant to shit talk Rye (from what I’ve heard of it, it actually is a decent effort), but it’s literally this comic then:
Yup
Look, it’s fine if you prefer other languages to python, I won’t besmirch anyone’s preferences. But literally everything in your post exists in nearly every programming language (minus some of the typing stuff, I’ll give you that, but it’s getting a lot better). Like, every language has some learning curve to setting up tooling, or configuring your IDE the way you like it, or learning how to navigate documentation so that it’s useful, or trying to decide on one of the multiple ways of doing things. I guarantee, as someone with limited experience with Java, I’d have a difficult time setting up and using IntelliJ, and figuring out which build/packaging system I need to use, and figuring out how to use whatever libraries I need, simply because I’m unfamiliar with the ecosystem. That’s all you’re describing - the initial learning curve in getting familiar with a new language. Which is why I pointed out all the things I pointed out. It’s where I start when I’m introducing developers to python.
No, it really is unique to python. Most other languages have one or two package managers, not 15 (15 is not an exaggeration). Ruby has one. Rust has one. Java has two (maven and gradle). Elixir has one. Swift has one.
Python programmers think it’s normal when it most definitely is not. Even your IntelliJ example isn’t correct because IntelliJ will literally install and set up the jdk for you, but pycharm is completely unable to do that and it’s not because JetBrains hasn’t tried. Python tooling is just really really really bad.
I won’t feel bad about it then, lol. At least not until I’m collabing on something and they want to use Python.
I wrote my bachelor’s thesis in Haskell and have never touched it again.
A lot of people feel that way. If I need to generate a set of numbers or a certain string, though, it’s pretty easy to punch out a one-liner in GHCi, and that’s usually my use case.
When did baccalaureate theses become a thing?
I don’t know if you’re trying to be funny or your autocorrect is, but in Germany, when they switched from diploma to the bachelor/master system, both bachelor and master come with a theses. Many people leave with a bachelor, especially in computer science, so that might be why. Don’t know about other countries
Ah, I see. They aren’t standard at US Universities, though I did have to complete capstone projects for each of my BS degrees.
It’s half a semester worth of credit points while the master theses is a whole semester here so it might be similar to your system
With type annotations, this problem is mostly alleviated in practice, while still keeping the productivity gains of duck typing.
In my experience, Python’s type annotations are annoying as hell, because there’s no inference.
You can use mypy and/or Pydantic.
Neither of those provide type inference? Type inference is when you give the compiler only occasional type hints and it can still figure out what the types are behind the scenes.
For example:
name = "World" greeting = "Hello " + name compile_error = greeting / 42In a type-inferred language, the compiler would automatically know that:
Mypy on the other hand can only tell these things, if you give the first two lines an explicit type hint:
name: String = "World" greeting: String = "Hello " + nameHaving to do this on every line of code is extremely noisy and makes refactoring annoying. I can absolutely understand that Python folks think you get productivity gains from duck typing, if this is the version of static typing they’re presented.
And we did excessively use mypy + type hints + pydantic on my most recent Python project. These are not the silver bullet you think they are…
If you’re going to post a code example, at least check that it works. Here’s your example, with no type hints, giving me errors both from the LSP, and when trying to run via mypy: https://imgur.com/a/Hq5Y5Gt.
I find it’s possible to operate Python as a statically typed language if you wanted, though it takes some setup with external tooling. It wasn’t hard, but had to set up pyright, editor integration, configuration to type check strictly and along with tests, and CI.
I even find the type system to be far more powerful than how I remembered Java’s to be (though I’m not familiar with the newest Java versions).
Ime anyone who is experienced enough knows how to avoid these Types of issues anyway (through tooling and good convention etc)
In Java you get a bunch of unexpected NullPointerExceptions instead…
usually this means you made a booboo, try catching these exceptions if it’s a possibility.
The IDE can assist you before it even becomes a problem, so even more time saved.
Did everyone become stupid in the last 10 or so years? We used to write huge apps in Python without any type checkers or static analysis tools and never had any problems we wouldn’t have had in statically typed languages.
no, we didn’t, we made a bunch of small programs