for server in servers:
if server.check_availability():
primary_server = server
break
else:
primary_server = backup_server
deploy_application(primary_server)
As it is shorter to do this:
primary_server = backup_server
for server in servers:
if server.check_availability():
primary_server = server
break
deploy_application(primary_server)
I would say that the main benefit of for/else is readability and ease of writing since nothing needs to be front-loaded. (The downside is an arguably redundant addition to the basic syntax, of course.)
If you replace the assignment in the else clause with something side-effectful it does make sense. But even then it harms the readability of the code to such a ridiculous extent. I've never not regretted adding an else clause to a for loop. Any cognitive benefits from Python's approach to for loops closely mirroring natural language just go out the window.
I do like the else clause with for loops. However most people are not familiar with it, and also `else:` as a keyword is confusing. I always remember it as `no break:`.
That's still confusing though. The problem here is that `else` is semantically attached to `break`, but syntactically attached to the body of the loop. The latter makes it look like it executes if the loop body didn't, if you interpret it in the most straightforward way.
IMO a better design would be to have a block that always executes at the end of the loop - there's even a reasonable keyword for it, `finally` - but gets a boolean flag indicating whether there was a break or not:
for server in servers:
if server.check_availability():
primary_server = server
break
finally did_break:
if not did_break:
primary_server = backup_server
Or better yet, make `break` take an optional argument (which defaults to `True` if unspecified), and that's what you get in `finally`. So this could be written:
for server in servers:
if server.check_availability():
break server
finally server:
primary_server = server if server is not None else backup_server
for server in servers:
if server.check_availability():
primary_server = server
break
else:
logger.warning("Cannot find a valid server") # <---
primary_server = backup_server
deploy_application(primary_server)
This kind of search can be done a variety of different ways, and is worth abstracting, e.g.:
def first(candidates, predicate, default):
try:
return next(c for c in candidates if predicate(c))
except StopIteration:
return default
deploy_application(first(servers, Server.check_availability, backup_server))
`more_itertools` is definitely on my shortlist of things I wish were in the Python standard library. (The list of things I'd like to see cut out, is probably considerably longer. But that's mainly on aesthetic principles; it's probably not creating a significant maintenance burden.)
And yet somehow I keep forgetting about it in the exact moments when it would probably make my life easier.
Indeed. I was thinking of ways to generalize for the case where a default isn't desired, then decided against introducing that complexity in the example, then forgot that I could re-simplify further.
If the team is familiar with Functional Programming concepts, I'd actually suggest
available_servers = (server for server in servers if server.check_availability())
primary_server = next(available_servers, backup_server)
deploy_application(primary_server)
which IMHO better conveys the intent to the reader (i.e. focus on "what to do" over "how to do it")
The shorter version that saves the else: line assigns primary_server twice: it is OK for an assignment but a potential problem if doing something less trivial, and very ugly style in any case.
The way the language is evolving, it seems likely to me that people in the applications camp (ML, simple web-dev, etc.) will soon need a "simple Python" fork or at least an agreed-upon subset of the language that doesn't have most of these complications (f-strings are a major success, though).
I would argue that most of these features (basically everything except metaclasses) are not advanced features. These are simple, but for some reason less well known or less used features.
Metaclasses are however quite complex (or at least lead to complex behavior) and I mostly avoid them for this reason.
And 'Proxy Properties' are not really a feature at all. Just a specific usage of dunder methods.
This is all fun and games unless you have to debug someone elses code and they use a new feature that you didnt know about.
Speaking for myself, I would be glad if there were a python_light version of the interpreter that has a simple syntax only like the early 3.x versions.
You underestimate the social complexity of a simplified language. Everyone would disagree about the design (what is worth including because it is "simple" or necessary?), everyone would avoid using an handicapped language where features they need are gratuitously missing, and nobody would care about it enough to work on implementation and maintenance.
This is a nice list of "things you might not know" that is worth skimming to add to your toolkit.
If you are really interested in "advanced Python", though, I would recommend the book Fluent Python by Ramalho. I have the first edition which is still highly relevant, including the async bits (you just have to translate the coroutines into async syntax). There is a second edition which is more up to date.
I would also recommend checking out the functools[0] and itertools[1] modules in the standard library. Just go and read the docs on them top to bottom.
It's also worth reading the first few sections of Python Data Model[2] and then bookmarking this page.
+1 to itertools especially. Absurdly powerful, and the recipes at the bottom of the doc page are terrific.
The only problem I’ve found is that for interviews, people often aren’t familiar with it, which can lead to you solving whatever puzzle they had in far less time than they intended, and without manually building whatever logic it was they assumed you would need.
> The only problem I’ve found is that for interviews, people often aren’t familiar with it, which can lead to you solving whatever puzzle they had in far less time than they intended, and without manually building whatever logic it was they assumed you would need.
My favourite example of a similar thing happening to me was when I was asked to reverse the digits in a number. I somewhat jokingly asked if I was assuming base 10, which got some awkward looks so I knew something was up. They weren't impressed at all with my answer of `"".join(reversed(str(123456789)))`. I didn't get the job.
Their loss. Expecting an arithmetic solution to be necessary is incongruous with accepting the overhead of Python. IIRC your approach will actually be faster; the string conversion implicitly number-crunches on the bare metal, whereas any arithmetic approach in Python code won't.
Unless they wanted you to convert back to int afterwards. Or wanted you to reverse the string with the slicing trick (which is, indeed, quite a bit faster yet).
Good list. Some of these i knew already, but the typing overloading and keyword/positional-only arguments were new to me.
One personal favorite of mine is __all__ for use in __init__.py files. It specifies which items are imported whenever uses from x import *. Especially useful when you have other people working on your codebase with the tendency to always import everything, which is rarely a good idea.
It’s never a good idea. I use `__all__` to explicitely list my exports in libraries, so that when one write `from mylib import `, the IDE auto-completes only the public classes and functions.
This sounds fun if you have 10x programmers or at least IQ > 140 programmers in charge. Last place I worked, I was told never use "smart" tricks if you can do the same thing in a simpler way. For-else and f-strings and := sound like harmless enough (though the last one is controversial); "with" is useful for resources that need to be deallocated but "yield"? Really? "more readable" when you slap a decorator on the thing? Why are we packing the enter and exit operations into a single function?
Playing with the typesystem and generics like this makes me worry I'm about to have a panic attack.
Give me code that I can understand and debug easily, even when I didn't write it; don't do implicit magical control flow changes unless you have a very good excuse, and then document both the how and the why - and you'll get a product that launches earlier and has fewer bugs.
Sometimes, a few more if statements here and there make code that is easier to understand, even if there's a clever hack that could cut a line or two here and there.
I don’t really see "smart tricks" here, just features of the language that you may not be aware of. I use most of them in my daily work without thinking about it: `@overload` and other typing features are useful to document the code, prevent mistakes, and help IDEs generate more helpful suggestions; keyword-only arguments force users to pass options as keywords, so you avoid calls like `foo(x, z, z, true, true, false, true)` and ensure the code is more explicit; for-else can sometimes clarify some complicated code; the walrus operator, short-circuiting with `or`, and chaining operators are just normal Python.
overload is very useful, but I find some people mistakenly think it actually modifies the runtime function signature to do multiple dispatch, so I do try to avoid it in favour of simpler functions whenever possible
too many overloads can be a code smell IMHO, it's ok when you are implementing a very common pattern and want proper types (decorators come to mind, so that both @decorator and @decorator() work, where the decorator might also have optional args) but I think in cases like the example in the article it should almost always be two separate functions
When I code, I try to make everything I write not clever. I am not saving a byte here and there because I am not typing in a program from a magazine in the 1980s (I was there, I am not going back). No code golf. I did my time on the Timex-Sinclair with its miserable two kilobytes of memory and reserved keywords as special-function strokes at a character each.
Each line should do one thing, in general, and it ought to be obvious. Cleverness is held in reserve for when it is truly needed, namely data structures and algorithms.
One of my accomplishments which seems to have mattered only to me is when my apartment-finding/renting system, written entirely in Perl, was transferred into the hands of students, new programming students. Perl is famously "write-once, read-never" and seems to have a culture favoring code golf and executable line noise. Still, the students got back to me and told me how easily-ported everything was, because I had done just one thing per line, avoided $_ and other such shortcuts, and other practices. They were very happy to take it over because I had avoided being cryptic and terse.
Test fixtures as opposed to production code forgive a multitude of sins. Sometimes you have to mock up an object that could never exist in the real world, to assert that the main code doesn't just accept it and move on. Sometimes you poke at individual variables in an object that in the real world should only be accessed via the methods in its API. All in a day's work if you know what you're doing and never try that in the real code.
The simple contextlib.contextmanager example doesn't really sell the benefits.
The main one is that it makes error handling and clean-up simpler, because you can just wrap the yield with a normal try/catch/finally, whereas to do this with __enter__ and __exit__ you have to work out what to do with the exception information in __exit__, which is easy to get wrong:
Suppressing or passing the exception is also mysterious, whereas with contextlib you just raise it as normal.
Another is that it makes managing state more obvious. If data is passed into the context manager, and needs to be saved between __enter__ and __exit__, that ends up in instance variables, whereas with contextlib you just use function parameters and local variables.
Finally, it makes it much easier to use other context managers, which also makes it look more like normal code.
Here's a more real-world-like example in both styles:
You can describe it in english as "open a file, then try to write a header, run the code inside the with statement, write a footer, and if anything fails truncate the file and pass the exception to the caller". This maps exactly to the lines in the contextlib version, whereas this logic is all spread out in the other one.
It's also more correct, as the file will be closed if any operation on it fails -- you'd need to add two more try/catch/finally blocks to the second example to make it as good.
Working in the ML field, I can't hate Python. But the type system (pre-3.12, of course) cost me a lot of nerves. Hoping for a better post-3.12 experience once all libraries are usable in 3.12+. After that experience, I’ve come to truly appreciate TypeScript’s type system. Never thought I’d say that.
Same experience here, Python’s typing experience is awful compared to TypeScript, even post-3.12. Mypy’s type inference is so dumb you have to write arguments like `i: int = 0`; `TypedDict`s seems promisable at first and then end up as a nightmare where you have to `cast` everything. I miss TypeScript’s `unknown` as well.
I believe mypy infers i as an integer in i = 0. I remember I had to do i = 0.0 to make it accept i += someFloat later on. Or of course i:float = 0 but I preferred the former.
Because it shouldn’t in function arguments. The one defining the function should be responsible enough to know what input they want and actually properly type it. Assuming an int or number type here is wrong (it could be optional int for example).
In TypeScript arguments with a default value "inherit" the type of that value, unless you explicitely mark it otherwise. I believe this is how Pyright works as well.
Because you provided a default value so clearly it’s not required to provide an input parameter. It’s also wrong to assume `0` is an int. There’s other valid types it could be. If the default was say `42`, I’d be pushing back a little less (outside of the Optional part), but this contrived example from GP had 0, which is ambiguous on what the inferred typing must be.
I was merely giving an example that strong typing has nothing to do with having to write the types. (and, obviously, the inferred type (int -> int) is correct. )
It's guaranteed to be correct if you use different operators for ints and floats, which is what at least some ML dialects (notably, OCaml) do precisely so that types can be inferred from usage.
That's the downside of operator overloading - since it relies on types to resolve, they need to be known and can't be inferred.
I want a typing system with a good inference that doesn’t require me to type each and every variable, just like in any good statically-typed language like OCaml or Typescript. Strong typing and explicit typing are two very different things.
How would a typing system know if the right type is `int` or `Optional[int]` or `Union[int, str]` or something else? The only right thing is to type the argument as `Any` in the absence of a type declaration.
The typing system should use the most specific valid type, and the code author can broaden it with explicit typing if needed. No good typing system should ever infer a variable as `Any`: "I don’t know the type of this" (`unknown` in TypeScript) is not the same thing as "This function accepts anything". Conflating these things is one of the main reasons why Mypy is so annoying.
A typing system should only infer things that it knows are true, it should never invent restrictions. In a language like Python that is duck-typed, `Any` is the only reasonable choice in the absence of other real constraints like a type-annotation.
def f(i=0) -> None:
if i is None:
do_something()
else:
do_something_else()
Yeah, I know it's retarded. I don't expect high quality code in a code base missing type annotation like that. Assuming `i` is `int` or `float` just makes incrementally adoption of a type checker harder.
No it’s not. The typing system should use the most specific type available, and it’s your responsability to broaden it if needed. That’s how it works in all statically-typed languages.
It still has a special case for dataclass-like things. I don't see how Python type checking (I haven't tried Red Knot) could let you do semi-magical things like Zod schema validation from TypeScript.
Mypy was designed to enable gradual adoption. There is definitely Python code out there with `def f(i=0)` where `i` could be any numeric type including floats, complex, numpy etc.. This is called duck typing. It's wrong for a type checker to assume `i: int` in such a case.
Pyright probably works if you use it for a new project from the start or invest a lot of time "fixing" an existing project. But it's a totally different tool and it's silly to criticise mypy without understanding its use case.
One of the most frustrating bugs I had encountered was when I was using memory mapped CSC format sparse arrays.
I needed the array indices to be int64 and specified them as such during initialization.
Downstreams, however, it would look at the actual index values and dynamically cast them to int32 if it judged there would be no loss in precision. This would completely screw up the roundtrip through a module implemented in C.
For me, the biggest benefit to python is that it feels like executable pseudocode. The language gets out of the way of your domain level instructions. This is probably why most non-programmers have the easiest time with python.
The more fancy stuff you add to it, the less attractive it becomes. Sure, most of these things have some sort of use, but I reckon most people do not get deep enough into python to understand all these little things.
I disagree. Adding "fancy stuff" by definition does not remove what made it attractive to you in the first place. You even acknowledge this in your second sentence. Something you don't know of is not capable to influence your opinion in any way. And when you know of it -- and dislike it -- just keep doing it the way you did before knowing it.
Lua is like this only moreso, and there's little fear of "progress" as large parts of the community have stuck with Lua5.1.
Of course, Lua is batteries-not-included so there may be the problem of "progress" in external libraries; in practice things like Penlight barely change though.
Yes, please. Given I'm forced to use Python, I'd welcome any "compile time" tools I can get. These days, it's using pyright with strict mode, which is pretty good, but there's still a long way to go.
Nitpick about 9.3 Short Circuit Evaluation: both things evaluate differently if you have empty strings. The if-else clause treats empty strings as valid while the or operator will treat them equivalent with None.
It has an effect, and is usually worth including anyway. I used to omit it by default; now I include it by default. Also, you say "nowadays" but it's been almost 13 years now (https://peps.python.org/pep-0420/).
> since 3.13 there is a @deprecated decorator that does what you think it does
Nice find. Probably worth mentioning it comes from the `warnings` standard library.
> the time package has functions for monotonic clocks and others not just time()
There's quite a bit in there, but I question how many people need it.
Anyway, it's always surprising to me how when other people make these lists, such a large fraction is taken up by tricks with type annotations. I was skeptical of the functionality when the `typing` standard library was introduced; I've only grown more and more wary of it, even as people continue to insist to me that it's somehow necessary.
Python has evolved into quite a complex language considering the amount of features that come built-in. The documentation, while complete, does not facilitate discovery of many of those features.
The quality of documentation is a known, highlighted concern in the community. There is particular interest in adopting Diataxis principles to improve documentation, both for Python itself and for packaging (see e.g. the user guide at https://packaging.python.org/en/latest/ , or the PyPA main site https://www.pypa.io/en/latest/ ).
It's more complex that decade ago, but still a relatively simple language. I can understand the article without much effort, while I scratch my head really hard when read about advance feature of Typescript or Scala.
If only. I suspect very few Python programmers can even fully explain what `a + b` does.
If `a` and `b` are instances of classes, many would say it's equivalent to `a.__add__(b)` or `type(a).__add__(a, b)`, but in fact it's much more complex.
Yes, the next level of complexity is that `a + b` will sometimes fall back to `b.__radd__(a)` if `a.__add__(b)` returns `NotImplemented`. But also:
- There are situations where `__radd__` takes priority over `__add__`. The rules for determining that priority are complex (and IIRC subtly different from the rules that determine whether `a < b` prioritises `a.__lt__(b)` or `b.__gt__(a)`).
- The lookup of `__add__` etc uses a special form of attribute lookup that's neither equivalent to `a.__add__` nor `type(a).__add__`. This special lookup only searches `type(a)` whereas the first would find an `__add__` function on `a`, and the second on `type(type(a))`.
I once broke some Python by changing a = a + b to a += b.
If a and b are lists, the latter modifies the existing list (which may be referenced elsewhere) instead of creating a new one.
I think Python is the only language I've encountered that uses the + operator with mutable reference semantics like this. It seems like a poor design choice.
This is one of the absolute worst design mistakes in Python. The example you give (that `a = a + b` and `a += b` aren't equivalent) is bad enough:
>>> a = b = [1, 2, 3] >>> a = b = [1, 2, 3]
>>> a = a + [4] >>> a += [4]
>>> a, b >>> a, b
([1, 2, 3, 4], [1, 2, 3]) ([1, 2, 3, 4], [1, 2, 3, 4])
What's worse is that sometimes, they are equivalent:
>>> a = b = (1, 2, 3) >>> a = b = (1, 2, 3)
>>> a = a + (4,) >>> a += (4,)
>>> a, b >>> a, b
((1, 2, 3, 4), (1, 2, 3)) ((1, 2, 3, 4), (1, 2, 3))
And even worse, in order to support a version of `a += b` that sometimes modifies `a` (e.g. with lists), and sometimes doesn't (with tuples), the implementation of the `+=` operator is convoluted, which can lead to:
>>> t = ([1, 2, 3], ['a'])
>>> t[0] += [4]
TypeError: 'tuple' object does not support item assignment
>>> t
([1, 2, 3, 4], ['a'])
The operation raises a TypeError, despite having succeeded!
I doubt many 4gl language developers could explain what a + b does in their respective language. That rabbit hole goes all the way down to the physical instruction execution on silicon.
Just like Microsoft maintains the Typescript handbook, I think the Python Foundation needs a "Idiomatic Modern Python Handbook" which always shows the most modern and idiomatic way to use Python features.
Every time I try to use Python I get this mixed feeling of liking how few guard rails there are between me and the logic I am making and this lingering worry that my code looks like fools attempt to use Python. So much exists as convention or loosely followed rules. Whenever I read articles like this I am wowed by the depth of things I didn't know about Python or how much has changed. It makes something like Go feel like a comfort blanket because I have reasonable certainty code I write won't feel deprecated or left behind a year or two later. Excellent article showing off even more I didn't know.
There are plenty of language features which let you re-use or build your own guardrails in python.
I do with the concept of truthiness and falsiness would be taken out and shot though. It's been responsible for far too many nasty bugs IME and it only cuts out a few extra characters. Not a great trade off.
Yeah, everybody uses it. That's partly how it ends up causing so many bugs - e.g. some user puts 0 in a text box and instead of the program reacting as if you put 0 in it reacts the same way as if no value was supplied.
Then everybody loses their minds because the system suddenly started doing something it was never supposed to.
Keep programming in python long enough and you'll see it eventually. It might be hard to recognize at first.
On the python bug tracker somewhere a while back there was a guy who wrote "if time_variable:" which resolved to false but only when it was called at midnight. It was initally resolved as wontfix.
(midnight being the datetime's conceptual equivalent to empty string, 0, empty list, etc.)
The most common case is when people use `if not s` to test for None in an otherwise string value (e.g. from JSON) without realizing that empty strings are also false.
... thank you for informing me that the code I wrote the other day is buggy. This sort of BS is exactly why I avoid JS. Truly one of the worst convenience "features" ever.
We need a unary "truthy" operator so that code remains compact and readable without the surprising behavior.
The "truthy" operator is bool(); the problem here is that empty strings are falsy.
For myself, I settled on these patterns for various kinds of tests:
# check for None
if x is not None: ...
# check for empty string
if x != "": ...
# check for empty collection other than string
if not len(x): ...
In this last case I rely on 0 being falsy to make this idiom distinct from checking length for a specific value via equality. So that when I'm scanning the code later and I see "not len", I know right away it's an empty check specifically.
But, of course, this only works if you're doing it consistently. And since the language doesn't enforce it, and there's no idiomatic Python standard for it, the footgun remains...
Those seem like opinionated rules that are easily added to ruff or other checkers to beat you over the head about. I know there are a bunch of truthiness opinions in ruff's catalog already that help me catch goofs.
bool( thing ) is overly verbose. It would be nice to have a unary operator.
I don't think the problem is that empty strings are falsy per say (although that might not be to your personal preference). Rather I think the problem is the implicit coercion to bool, hence my desire for a unary operator.
I think this is yet another entry in the (extremely) long list of examples of why implicit type conversions are a bad thing.
> I don't think the problem is that empty strings are falsy per say (although that might not be to your personal preference).
Let me correct that. The problem isn't so much that they are falsy per se, but rather that they share this property with so many unrelated things - and this is then combined with dynamic typing, so any given `x` can be any of those things. None/'' in particular is exceedingly common because None is the standard way to report absence of a value in Python.
As far as having a symbolic unary operator for such checks, I think that would go stylistically contrary to the overall feel of Python syntax - I mean, this is a language in which you have to spell out things like "not", and len() is also a function unlike say Lua. It feels like the most Pythonic interface for this would be as an instance property with a descriptive name.
Instead of comfort I feel constantly annoyed that the language gives me so little that I have to produce more lines and very little in the way of convenience. It's a language that benefits a lot from AI completion, at least. Yeah, old code still looks the same as new code but that is due to a lack of progress on the language though.
I write an unfortunate amount of Go code.. because Go supports 2(?) of the features in this post (structural typing and generics, sort of, if you're feeling generous).
For example, circular imports aren't supported in Go at all. I run into this from time to time even if using an interface sometimes consts or other types are defined in the package as well and the whole thing has to be refactored. No way around it, has to be done.
Circular imports aren't encouraged in Python but Python never leaves you without options. Code can import types only while type checking or move imports out of the module level are quick workarounds. (I hope for typescripts `import type` someday.) That's what gives me a "comfort" feeling in Python, knowing that the language isn't likely to force me to work around the language design.
pick up a book on how to program pythonically and it will feel better, I promise. A lot of people program in python like they do in c++ or rust or javascript and it doesn't feel "right" to them. If you follow the pythonistas take I think you'll appreciate it better.
Turns out I managed to use almost all of these during a refactor of a project at work, even metaclasses... (Metaclass usage is justified in my case: we have a sort of language evaluator and using a metaclass lets us define function arguments with their types and validators in a very coincise and obvious way similar to Pydantic.)
I think this list should also include descriptors[0]: it's another metaprogramming feature that allows running code when accessing or setting class attributes similar to @property but more powerful. (edit: nvm, I saw that they are covered in the proxy properties section!)
I think the type system is quite good actually, even if you end up having to sidestep it when doing this kind of meta-programming. The errors I do get are generally the library's fault (old versions of SQLAlchemy make it impossible to assign types anywhere...) and there's a few gotchas (like mutable collections being invariant, so if you take a list as an argument you may have to type it as `Sequence[]` or you'll get type errors) but it's functional and makes the language usable for me.
I stopped using Ruby because upstream would not commit on type checking (yes I know you have a few choices if you want typing, but they're a bit too much overhead for what I usually use Ruby for, which is writing scripts), and I'm glad Python is committing here.
I enjoyed reading the article. I'm far from a Python expert, but as an observation, most of these features are actually just typing module features. In particular, I wasn't sold on Generics or Protocols as I would have just used duck typing in both cases... Does modern, production-level python code use types everywhere? Is duck typing frowned upon?
> Does modern, production-level python code use types everywhere?
I don't know about code inside companies, but most new open source projects I encounter use typing. Many old ones have been converted too.
> Is duck typing frowned upon?
No. You can use Protocols to define what shape you expect your duck to be. There's some discussion about whether you should use abstract classes or protocols, though.
Structural typing is a static equivalent of dynamic duck typing. It is compatible with type checkers. Protocols mentioned in the article work without inheritance. A duck doesn't need to know that there is SupportsQuack protocol. If it quacks, it passes the type checks.
That feature I'm using regularly and can recommend doing so, e.g. when passing boolean flags to functions, to clarify their meaning at the call-site by ensuring they have to be passed by name.
I did not expect to wake up at 4am seeing my post on front page HN, but here we are nevertheless :D
As the intro mentioned, these started off as 14 small tweets I wrote a month prior to starting my blog. When I finally got that set up, I just thought, "hey, I just spent the better part of two weeks writing these nifty Python tricks, might as well reuse them as a fun first post!"
That's why the flow might seem a little weird (as some pointed out, proxy properties are not really a Python "feature" in of itself). They were just whatever I found cool that day. I tried to find something more esoteric if it was a Friday, and something useful if it was a Monday. I was also kinda improving the entire series as it was going on, so that was also a factor.
Same goes with the title. These were just 14 feature I found interesting while writing Python both professionally and as a hobby. Some people mentioned these are not very "advanced" per se, and fair enough. I think I spent a total of 5 second thinking of a title. Oh well!
Hey, I really enjoyed reading your post. It's straight to the point, the list is very rich and well illustrated with examples. I have been coding in python for the last 20 years and I nowadays rarely like reading these posts as much as I did for yours. Kudos.
The tricks are more advanced than what I've seen in most people's code at some jobs before! The post is refreshing because it sits in the neglected middle ground between the tons of Python-for-beginners content out there and the arcane "I'm a core developer writing about very specific things" content.
I think the background of the post is very well explained on the blog.
But I’m surprised to read you went from idea to posting each day. I had expected maybe a week of preparation before starting execution.
Regardless it’s a great post, regardless of wether one think there is too much typing or too many new features in modern Python or one is a lover of niche solutions to shorten code.
180 comments
[ 139 ms ] story [ 994 ms ] threadIMO a better design would be to have a block that always executes at the end of the loop - there's even a reasonable keyword for it, `finally` - but gets a boolean flag indicating whether there was a break or not:
Or better yet, make `break` take an optional argument (which defaults to `True` if unspecified), and that's what you get in `finally`. So this could be written:And yet somehow I keep forgetting about it in the exact moments when it would probably make my life easier.
The itertools package.
never used it but it seems worth a try
Metaclasses are however quite complex (or at least lead to complex behavior) and I mostly avoid them for this reason.
And 'Proxy Properties' are not really a feature at all. Just a specific usage of dunder methods.
just my 2 ct
What’s wrong about learning things by looking at code from more experienced people?
I would add assert_never to the pattern matching section for exhaustiveness checks: https://typing.python.org/en/latest/guides/unreachable.html#...
If you are really interested in "advanced Python", though, I would recommend the book Fluent Python by Ramalho. I have the first edition which is still highly relevant, including the async bits (you just have to translate the coroutines into async syntax). There is a second edition which is more up to date.
I would also recommend checking out the functools[0] and itertools[1] modules in the standard library. Just go and read the docs on them top to bottom.
It's also worth reading the first few sections of Python Data Model[2] and then bookmarking this page.
[0] https://docs.python.org/3/library/functools.html
[1] https://docs.python.org/3/library/itertools.html
[2] https://docs.python.org/3/reference/datamodel.html
The only problem I’ve found is that for interviews, people often aren’t familiar with it, which can lead to you solving whatever puzzle they had in far less time than they intended, and without manually building whatever logic it was they assumed you would need.
My favourite example of a similar thing happening to me was when I was asked to reverse the digits in a number. I somewhat jokingly asked if I was assuming base 10, which got some awkward looks so I knew something was up. They weren't impressed at all with my answer of `"".join(reversed(str(123456789)))`. I didn't get the job.
Their loss. Expecting an arithmetic solution to be necessary is incongruous with accepting the overhead of Python. IIRC your approach will actually be faster; the string conversion implicitly number-crunches on the bare metal, whereas any arithmetic approach in Python code won't.
Unless they wanted you to convert back to int afterwards. Or wanted you to reverse the string with the slicing trick (which is, indeed, quite a bit faster yet).
One personal favorite of mine is __all__ for use in __init__.py files. It specifies which items are imported whenever uses from x import *. Especially useful when you have other people working on your codebase with the tendency to always import everything, which is rarely a good idea.
Playing with the typesystem and generics like this makes me worry I'm about to have a panic attack.
Give me code that I can understand and debug easily, even when I didn't write it; don't do implicit magical control flow changes unless you have a very good excuse, and then document both the how and the why - and you'll get a product that launches earlier and has fewer bugs.
Sometimes, a few more if statements here and there make code that is easier to understand, even if there's a clever hack that could cut a line or two here and there.
too many overloads can be a code smell IMHO, it's ok when you are implementing a very common pattern and want proper types (decorators come to mind, so that both @decorator and @decorator() work, where the decorator might also have optional args) but I think in cases like the example in the article it should almost always be two separate functions
When I code, I try to make everything I write not clever. I am not saving a byte here and there because I am not typing in a program from a magazine in the 1980s (I was there, I am not going back). No code golf. I did my time on the Timex-Sinclair with its miserable two kilobytes of memory and reserved keywords as special-function strokes at a character each.
Each line should do one thing, in general, and it ought to be obvious. Cleverness is held in reserve for when it is truly needed, namely data structures and algorithms.
One of my accomplishments which seems to have mattered only to me is when my apartment-finding/renting system, written entirely in Perl, was transferred into the hands of students, new programming students. Perl is famously "write-once, read-never" and seems to have a culture favoring code golf and executable line noise. Still, the students got back to me and told me how easily-ported everything was, because I had done just one thing per line, avoided $_ and other such shortcuts, and other practices. They were very happy to take it over because I had avoided being cryptic and terse.
The main one is that it makes error handling and clean-up simpler, because you can just wrap the yield with a normal try/catch/finally, whereas to do this with __enter__ and __exit__ you have to work out what to do with the exception information in __exit__, which is easy to get wrong:
https://docs.python.org/3/reference/datamodel.html#object.__...
Suppressing or passing the exception is also mysterious, whereas with contextlib you just raise it as normal.
Another is that it makes managing state more obvious. If data is passed into the context manager, and needs to be saved between __enter__ and __exit__, that ends up in instance variables, whereas with contextlib you just use function parameters and local variables.
Finally, it makes it much easier to use other context managers, which also makes it look more like normal code.
Here's a more real-world-like example in both styles:
https://gist.github.com/tomjnixon/e84c9254ab6d00542a22b7d799...
I think the first is much more obvious.
You can describe it in english as "open a file, then try to write a header, run the code inside the with statement, write a footer, and if anything fails truncate the file and pass the exception to the caller". This maps exactly to the lines in the contextlib version, whereas this logic is all spread out in the other one.
It's also more correct, as the file will be closed if any operation on it fails -- you'd need to add two more try/catch/finally blocks to the second example to make it as good.
int -> int
Is wrong. At minimum it’s:
Optional[int] -> int
Because you provided a default value so clearly it’s not required to provide an input parameter. It’s also wrong to assume `0` is an int. There’s other valid types it could be. If the default was say `42`, I’d be pushing back a little less (outside of the Optional part), but this contrived example from GP had 0, which is ambiguous on what the inferred typing must be.
Pyright correctly deduces the type as int.
In any case it's a bad example as function signatures should always be typed.
2) inferring the type is int isn’t guaranteed to be correct in this case
No it’s not. It’s Optional[int] -> int at minimum. There are other completely valid signatures beyond that too.
That's the downside of operator overloading - since it relies on types to resolve, they need to be known and can't be inferred.
It could be:
Yeah, I know it's retarded. I don't expect high quality code in a code base missing type annotation like that. Assuming `i` is `int` or `float` just makes incrementally adoption of a type checker harder.I mentioned pyright because (some of) the specific concerns by OP are addressed by it.
Pyright probably works if you use it for a new project from the start or invest a lot of time "fixing" an existing project. But it's a totally different tool and it's silly to criticise mypy without understanding its use case.
I tried Pyright but as you say on an existing project you need a looot of time to "fix" it.
You don’t know but you are addicted to types
Come to the light - Haskell!
Or embrace logic + functional programming: Curry. https://curry-language.org/
I needed the array indices to be int64 and specified them as such during initialization.
Downstreams, however, it would look at the actual index values and dynamically cast them to int32 if it judged there would be no loss in precision. This would completely screw up the roundtrip through a module implemented in C.
Being an intermittent bug it was quite a hell.
The more fancy stuff you add to it, the less attractive it becomes. Sure, most of these things have some sort of use, but I reckon most people do not get deep enough into python to understand all these little things.
That breaks down as soon as you need to work with anyone else's code that uses the "fancy stuff".
Of course, Lua is batteries-not-included so there may be the problem of "progress" in external libraries; in practice things like Penlight barely change though.
* did you know __init__.py is optional nowadays?
* you can do relative imports with things like "from ..other import foo"
* since 3.13 there is a @deprecated decorator that does what you think it does
* the new generics syntax also works on methods/functions: "def method[T](...)" very cool
* you can type kwargs with typeddicts and unpack: "def fn(*kwargs: Unpack[MyKwargs])"
* dataclasses (and pydantic) support immutable objects with: "class MyModel(BaseModel, frozen=True)" or "@dataclass(frozen=True)"
* class attributes on dataclasses, etc. can be defined with "MY_STATIC: ClassVar[int] = 42" this also supports abstract base classes (ABC)
* TypeVar supports binding to enforce subtypes: "TypeVar['T', bound=X]", and also a default since 3.13: "TypeVar['T', bound=X, default=int]"
* @overload is especially useful for get() methods to express that the return can't be none if the default isn't None
* instead of Union[a, b] or Optional[a] you can write "a | b" or "a | None" nowadays
* with match you can use assert_never() to ensure exhaustive matching in a "case _:" block
* typing has reveal_type() which lets mypy print the type it thinks something is
* typing's "Self" allows you to more properly annotate class method return types
* the time package has functions for monotonic clocks and others not just time()
anyone know more things?
It has an effect, and is usually worth including anyway. I used to omit it by default; now I include it by default. Also, you say "nowadays" but it's been almost 13 years now (https://peps.python.org/pep-0420/).
> since 3.13 there is a @deprecated decorator that does what you think it does
Nice find. Probably worth mentioning it comes from the `warnings` standard library.
> the time package has functions for monotonic clocks and others not just time()
There's quite a bit in there, but I question how many people need it.
Anyway, it's always surprising to me how when other people make these lists, such a large fraction is taken up by tricks with type annotations. I was skeptical of the functionality when the `typing` standard library was introduced; I've only grown more and more wary of it, even as people continue to insist to me that it's somehow necessary.
It's not optional. Omitting it gets you a namespace package, which is probably not what you want.
> TypeVar supports binding to enforce subtypes: "TypeVar['T', bound=X]",
Using the new generics syntax you mentioned above you can now do:
If you want to help, there's a section on the Python forum (https://discuss.python.org/c/documentation/26) and a Discord server, and issues with documentation can also be reported on the main Python GitHub issue tracker (https://github.com/python/cpython/labels/docs).
If only. I suspect very few Python programmers can even fully explain what `a + b` does.
If `a` and `b` are instances of classes, many would say it's equivalent to `a.__add__(b)` or `type(a).__add__(a, b)`, but in fact it's much more complex.
- There are situations where `__radd__` takes priority over `__add__`. The rules for determining that priority are complex (and IIRC subtly different from the rules that determine whether `a < b` prioritises `a.__lt__(b)` or `b.__gt__(a)`).
- The lookup of `__add__` etc uses a special form of attribute lookup that's neither equivalent to `a.__add__` nor `type(a).__add__`. This special lookup only searches `type(a)` whereas the first would find an `__add__` function on `a`, and the second on `type(type(a))`.
I've also heard of further complications caused by implementation details leaking into the language semantics - for example, see Armin Ronacher's blog post: https://lucumr.pocoo.org/2014/8/16/the-python-i-would-like-t...
If a and b are lists, the latter modifies the existing list (which may be referenced elsewhere) instead of creating a new one.
I think Python is the only language I've encountered that uses the + operator with mutable reference semantics like this. It seems like a poor design choice.
Using it since version 1.6.
I do with the concept of truthiness and falsiness would be taken out and shot though. It's been responsible for far too many nasty bugs IME and it only cuts out a few extra characters. Not a great trade off.
Then everybody loses their minds because the system suddenly started doing something it was never supposed to.
On the python bug tracker somewhere a while back there was a guy who wrote "if time_variable:" which resolved to false but only when it was called at midnight. It was initally resolved as wontfix.
(midnight being the datetime's conceptual equivalent to empty string, 0, empty list, etc.)
We need a unary "truthy" operator so that code remains compact and readable without the surprising behavior.
For myself, I settled on these patterns for various kinds of tests:
In this last case I rely on 0 being falsy to make this idiom distinct from checking length for a specific value via equality. So that when I'm scanning the code later and I see "not len", I know right away it's an empty check specifically.But, of course, this only works if you're doing it consistently. And since the language doesn't enforce it, and there's no idiomatic Python standard for it, the footgun remains...
I don't think the problem is that empty strings are falsy per say (although that might not be to your personal preference). Rather I think the problem is the implicit coercion to bool, hence my desire for a unary operator.
I think this is yet another entry in the (extremely) long list of examples of why implicit type conversions are a bad thing.
Let me correct that. The problem isn't so much that they are falsy per se, but rather that they share this property with so many unrelated things - and this is then combined with dynamic typing, so any given `x` can be any of those things. None/'' in particular is exceedingly common because None is the standard way to report absence of a value in Python.
As far as having a symbolic unary operator for such checks, I think that would go stylistically contrary to the overall feel of Python syntax - I mean, this is a language in which you have to spell out things like "not", and len() is also a function unlike say Lua. It feels like the most Pythonic interface for this would be as an instance property with a descriptive name.
I write an unfortunate amount of Go code.. because Go supports 2(?) of the features in this post (structural typing and generics, sort of, if you're feeling generous).
For example, circular imports aren't supported in Go at all. I run into this from time to time even if using an interface sometimes consts or other types are defined in the package as well and the whole thing has to be refactored. No way around it, has to be done.
Circular imports aren't encouraged in Python but Python never leaves you without options. Code can import types only while type checking or move imports out of the module level are quick workarounds. (I hope for typescripts `import type` someday.) That's what gives me a "comfort" feeling in Python, knowing that the language isn't likely to force me to work around the language design.
I think this list should also include descriptors[0]: it's another metaprogramming feature that allows running code when accessing or setting class attributes similar to @property but more powerful. (edit: nvm, I saw that they are covered in the proxy properties section!)
I think the type system is quite good actually, even if you end up having to sidestep it when doing this kind of meta-programming. The errors I do get are generally the library's fault (old versions of SQLAlchemy make it impossible to assign types anywhere...) and there's a few gotchas (like mutable collections being invariant, so if you take a list as an argument you may have to type it as `Sequence[]` or you'll get type errors) but it's functional and makes the language usable for me.
I stopped using Ruby because upstream would not commit on type checking (yes I know you have a few choices if you want typing, but they're a bit too much overhead for what I usually use Ruby for, which is writing scripts), and I'm glad Python is committing here.
[0]: https://docs.python.org/3/howto/descriptor.html
I don't know about code inside companies, but most new open source projects I encounter use typing. Many old ones have been converted too.
> Is duck typing frowned upon?
No. You can use Protocols to define what shape you expect your duck to be. There's some discussion about whether you should use abstract classes or protocols, though.
https://blog.edward-li.com/tech/advanced-python-features/#2-...
https://docs.python.org/3.12/reference/compound_stmts.html#f...
I did not expect to wake up at 4am seeing my post on front page HN, but here we are nevertheless :D
As the intro mentioned, these started off as 14 small tweets I wrote a month prior to starting my blog. When I finally got that set up, I just thought, "hey, I just spent the better part of two weeks writing these nifty Python tricks, might as well reuse them as a fun first post!"
That's why the flow might seem a little weird (as some pointed out, proxy properties are not really a Python "feature" in of itself). They were just whatever I found cool that day. I tried to find something more esoteric if it was a Friday, and something useful if it was a Monday. I was also kinda improving the entire series as it was going on, so that was also a factor.
Same goes with the title. These were just 14 feature I found interesting while writing Python both professionally and as a hobby. Some people mentioned these are not very "advanced" per se, and fair enough. I think I spent a total of 5 second thinking of a title. Oh well!
One I think you missed is getters and setters on attributes!
Congratulations for ending up on the front page! (I hope the server hosting your blog is okay!)
But I’m surprised to read you went from idea to posting each day. I had expected maybe a week of preparation before starting execution.
Regardless it’s a great post, regardless of wether one think there is too much typing or too many new features in modern Python or one is a lover of niche solutions to shorten code.